MASTER’S DEGREE THESIS

Master’s degree in Management Engineering

DEVELOPMENT OF A
CRYPTOCURRENCY BOT

REPORT
January 12th , 2023

Author’s full name:

José Manuel Rodrı́guez
Lomeña

Mentor’s full name:

Rafel Amer Ramon

Call for delivery:

2022/23 - 1
Acknowledgments

The development of this project would not have been possible with the support of
some people. These lines are to express them my gratitude.
Firstly, I want to thank my project director, Rafel Amer, whose knowledge and
passion for cryptocurrency has helped me in the crossroads that have been presenting
during all these months. You are the definition of what a great tutor is.
I also gratefully acknowledge the support of my friends Sergio Sánchez and Guillem
Soriano. They helped me whenever I have needed it with code programming. I am
sure that your desire to learn and innovate will take you very far.
Finally, I want to thank all my loved ones for the constant support and patience. All
my work is also yours.
Declaration of Honour

I declare that,
• the work in this Master’s Degree Thesis is completely my own work,
• no part of this Master’s Degree Thesis is taken from other people’s work without
giving them credit,
• all references have been clearly cited,
I understand that an infringement of this declaration leaves me subject to the foreseen
disciplinary actions by The Universitat Politècnica de Catalunya - BarcelonaTECH.

Title of the Thesis

Development of a Cryptocurrency Bot
Contents

Nomenclature xiii

Abstract xv

1 Aim 1

2 Scope 3
2.1 In Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Out of Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 Justification 5

4 Introduction 9

5 Budget 11
5.1 Breakdown of Unit Costs . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.2 Total Budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

6 Environmental Impact 15

7 State of the Art 17

7.1 Capfolio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
7.2 3Commas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.3 CCXT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.4 Blackbird . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.5 StockSharp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.6 Freqtrade . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.7 CryptoSignal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
7.8 Ctubio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
7.9 Catalyst . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
7.10 Using Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . 20

8 Technical Analysis of the Financial Markets 21

8.1 Philosophy of Technical Analysis . . . . . . . . . . . . . . . . . . . . 21
8.2 Dow Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
8.3 Basic Concepts of Trends . . . . . . . . . . . . . . . . . . . . . . . . . 25
8.4 Volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
8.5 Chart Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

v
CONTENTS

8.5.1 Reversal Patterns . . . . . . . . . . . . . . . . . . . . . . . . . 32

8.5.1.1 The Head and Shoulders . . . . . . . . . . . . . . . . 32
8.5.1.2 Triple Tops and Bottoms . . . . . . . . . . . . . . . . 34
8.5.1.3 Double Tops and Bottoms . . . . . . . . . . . . . . . 35
8.5.2 Continuation Patterns . . . . . . . . . . . . . . . . . . . . . . 36
8.5.2.1 Triangles . . . . . . . . . . . . . . . . . . . . . . . . 36
8.5.2.2 Broadening Formation . . . . . . . . . . . . . . . . . 38
8.5.2.3 Flags and Pennants . . . . . . . . . . . . . . . . . . 38
8.5.2.4 Wedge Formation . . . . . . . . . . . . . . . . . . . . 40
8.5.2.5 Rectangle Formation . . . . . . . . . . . . . . . . . . 41
8.5.3 Confirmation and Divergence . . . . . . . . . . . . . . . . . . 42
8.6 Chart Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
8.6.1 Candlestick Charting . . . . . . . . . . . . . . . . . . . . . . . 44
8.6.2 Arithmetic versus Logarithmic Scale . . . . . . . . . . . . . . 45
8.6.3 Long Term Charts . . . . . . . . . . . . . . . . . . . . . . . . 46
8.7 Moving Averages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
8.7.1 The Simple Moving Average . . . . . . . . . . . . . . . . . . . 47
8.7.2 The Linearly Weighted Moving Average . . . . . . . . . . . . . 47
8.7.3 The Exponentially Smoothed Moving Average . . . . . . . . . 48
8.7.4 Using Two Moving Averages . . . . . . . . . . . . . . . . . . . 48
8.8 Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
8.8.1 Measuring Momentum . . . . . . . . . . . . . . . . . . . . . . 51
8.8.2 Oscillator Using Two Moving Averages . . . . . . . . . . . . . 52
8.8.3 The Relative Strength Index (RSI) . . . . . . . . . . . . . . . 53
8.8.4 The Importance of Trend . . . . . . . . . . . . . . . . . . . . . 54
8.8.5 When Oscillators Are Most Useful . . . . . . . . . . . . . . . . 54

9 Introduction to Cryptocurrency 55
9.1 Cryptocurrency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
9.1.1 Meaning and Definition . . . . . . . . . . . . . . . . . . . . . 55
9.1.2 Blockchain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
9.1.2.1 Transaction Process . . . . . . . . . . . . . . . . . . 56
9.1.3 Cryptography . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
9.1.3.1 Cryptographic Algorithms . . . . . . . . . . . . . . . 58
9.1.4 Cryptocurrency Wallet . . . . . . . . . . . . . . . . . . . . . . 62
9.1.4.1 Types of Cryptocurrency Wallets . . . . . . . . . . . 63
9.1.5 Proof-of-Work against Proof-of-Stake . . . . . . . . . . . . . . 64
9.1.5.1 Proof-of-Work . . . . . . . . . . . . . . . . . . . . . . 64
9.1.5.2 Proof-of-Stake . . . . . . . . . . . . . . . . . . . . . 65
9.1.5.3 PoW & PoS Comparison . . . . . . . . . . . . . . . . 65
9.1.6 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
9.1.6.1 The Idea for Cryptocurrency . . . . . . . . . . . . . 67
9.1.6.2 The Beginning (2008-2010) . . . . . . . . . . . . . . 67
9.1.6.3 The Market Begins to Form (2010-2014) . . . . . . . 67
9.1.6.4 Scams Dominate Headlines (2014-2016) . . . . . . . 68
9.1.6.5 Bitcoin Ascends to Worldwide Phenomenon (2016-2018) 68

José Manuel Rodrı́guez Lomeña vi

 CONTENTS

9.1.6.6 Bust and Recovery, and Bust and Recovery (2018-

Present) . . . . . . . . . . . . . . . . . . . . . . . . . 69
9.1.7 Types of cryptocurrency . . . . . . . . . . . . . . . . . . . . . 70
9.1.7.1 Coins . . . . . . . . . . . . . . . . . . . . . . . . . . 70
9.1.7.2 Tokens . . . . . . . . . . . . . . . . . . . . . . . . . . 71
9.1.7.3 Largest Cryptocurrencies by Market Cap . . . . . . . 72
9.1.8 Cryptocurrency Exchange . . . . . . . . . . . . . . . . . . . . 74
9.1.8.1 Centralized Cryptocurrency Exchange . . . . . . . . 74
9.1.8.2 Decentralized Cryptocurrency Exchange . . . . . . . 75
9.1.8.3 CEX against DEX . . . . . . . . . . . . . . . . . . . 75
9.2 Cryptocurrency Bot . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
9.2.1 Backtesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
9.2.2 Yahoo! Finance . . . . . . . . . . . . . . . . . . . . . . . . . . 78

10 Machine Learning and Statistics 79

10.1 Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
10.1.1 Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . 80
10.1.2 Recurrent Neural Network . . . . . . . . . . . . . . . . . . . . 81
10.1.3 Long Short-Term Memory Network . . . . . . . . . . . . . . . 81
10.1.4 Modular Neural Network . . . . . . . . . . . . . . . . . . . . . 82
10.2 Statistical Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
10.2.1 ARIMA model . . . . . . . . . . . . . . . . . . . . . . . . . . 83
10.2.2 Combination Without Repetition . . . . . . . . . . . . . . . . 84

11 Python Libraries 85
11.1 General Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
11.2 Data Analysis Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . 86
11.3 Visualization Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . 87
11.4 Trading Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
11.5 Machine Learning Libraries . . . . . . . . . . . . . . . . . . . . . . . 88

12 Studies 89
12.1 Simulation on Random Trade Closures . . . . . . . . . . . . . . . . . 89
12.2 Analysis of Strategy Variables . . . . . . . . . . . . . . . . . . . . . . 92

13 Software Description 97
13.1 Atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
13.2 GitHub Desktop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
13.3 Command Prompt . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

14 Software Implementation 99
14.1 Requirements of the Algorithm . . . . . . . . . . . . . . . . . . . . . 99
14.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
14.2.1 Bot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
14.2.1.1 botData . . . . . . . . . . . . . . . . . . . . . . . . . 101
14.2.1.2 botCore . . . . . . . . . . . . . . . . . . . . . . . . . 102
14.2.1.3 botMain . . . . . . . . . . . . . . . . . . . . . . . . . 107

vii José Manuel Rodrı́guez Lomeña

CONTENTS

14.2.2 Backtesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

14.2.2.1 settings . . . . . . . . . . . . . . . . . . . . . . . . . 110
14.2.2.2 sizers . . . . . . . . . . . . . . . . . . . . . . . . . . 111
14.2.2.3 downloadData . . . . . . . . . . . . . . . . . . . . . 111
14.2.2.4 strategy . . . . . . . . . . . . . . . . . . . . . . . . . 113
14.2.2.5 backtestingCore . . . . . . . . . . . . . . . . . . . . . 114
14.2.2.6 backtestingMain . . . . . . . . . . . . . . . . . . . . 119

15 Results 121
15.1 Backtesting for a Frequency of 15 Minutes . . . . . . . . . . . . . . . 122
15.2 Backtesting for a Frequency of 5 Minutes . . . . . . . . . . . . . . . . 122
15.3 Backtesting for a Frequency of 30 Minutes . . . . . . . . . . . . . . . 123
15.4 Strategy Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

16 Conclusions 125

17 Future Work 127

Appendices 128

A Bot Code 129

A.1 botData . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
A.2 botCore . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
A.3 botMain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
A.4 settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
A.5 sizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
A.6 downloadData . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
A.7 strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
A.8 backtestingCore . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
A.9 backtestingMain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

B Studies Codes 149

B.1 Simulation on Random Trade Closures . . . . . . . . . . . . . . . . . 149
B.2 Combination Without Repetition . . . . . . . . . . . . . . . . . . . . 152
B.3 ROI vs Number of Trades for ωp Range . . . . . . . . . . . . . . . . . 153
B.4 ROI vs Number of Trades for p Range . . . . . . . . . . . . . . . . . 154
B.5 ROI vs Number of Trades for A Range . . . . . . . . . . . . . . . . . 155

José Manuel Rodrı́guez Lomeña viii

List of Figures

Figures from articles, books, these or other sources are cited in their respective
caption. In any other case, they correspond to figures of own elaboration.

8.1 Failure Swing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

8.2 Nonfailure Swing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
8.3 Failure Swing Bottom . . . . . . . . . . . . . . . . . . . . . . . . . . 24
8.4 Nonfailure Swing Bottom . . . . . . . . . . . . . . . . . . . . . . . . . 24
8.5 Example of an uptrend . . . . . . . . . . . . . . . . . . . . . . . . . . 25
8.6 Example of a downtrend . . . . . . . . . . . . . . . . . . . . . . . . . 25
8.7 Example of a sideways trend . . . . . . . . . . . . . . . . . . . . . . . 26
8.8 Example the three degrees of trend . . . . . . . . . . . . . . . . . . . 26
8.9 Support and resistance levels in uptrend . . . . . . . . . . . . . . . . 27
8.10 Example of an up trendline . . . . . . . . . . . . . . . . . . . . . . . 28
8.11 Example of an down trendline . . . . . . . . . . . . . . . . . . . . . . 28
8.12 Breaking of an up trendline . . . . . . . . . . . . . . . . . . . . . . . 29
8.13 Breaking of a down trendline . . . . . . . . . . . . . . . . . . . . . . . 29
8.14 Example of an uptrend channel . . . . . . . . . . . . . . . . . . . . . 30
8.15 Example of a downtrend channel . . . . . . . . . . . . . . . . . . . . 30
8.16 Example of volume bars . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.17 Example of a head and shoulders top . . . . . . . . . . . . . . . . . . 33
8.18 Example of an inverse head and shoulders . . . . . . . . . . . . . . . 34
8.19 Example of a triple top . . . . . . . . . . . . . . . . . . . . . . . . . . 34
8.20 Example of a triple bottom . . . . . . . . . . . . . . . . . . . . . . . 35
8.21 Example of a double top . . . . . . . . . . . . . . . . . . . . . . . . . 35
8.22 Example of a double bottom . . . . . . . . . . . . . . . . . . . . . . . 35
8.23 Example of a bullish symmetrical triangle . . . . . . . . . . . . . . . 36
8.24 Example of an ascending triangle . . . . . . . . . . . . . . . . . . . . 37
8.25 Example of a descending triangle . . . . . . . . . . . . . . . . . . . . 37
8.26 Example of a broadening top . . . . . . . . . . . . . . . . . . . . . . . 38
8.27 Example of a bullish flag . . . . . . . . . . . . . . . . . . . . . . . . . 39
8.28 Example of a bullish pennant . . . . . . . . . . . . . . . . . . . . . . 40
8.29 Example of a bullish falling wedge . . . . . . . . . . . . . . . . . . . . 41
8.30 Example of a bearish wedge . . . . . . . . . . . . . . . . . . . . . . . 41
8.31 Example of a bullish falling rectangle . . . . . . . . . . . . . . . . . . 42
8.32 Example of a bearish rectangle . . . . . . . . . . . . . . . . . . . . . 42
8.33 Example of a bar chart . . . . . . . . . . . . . . . . . . . . . . . . . . 43
8.34 Example of line chart . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
8.35 Example of candlestick . . . . . . . . . . . . . . . . . . . . . . . . . . 44

ix
LIST OF FIGURES

8.36 Bar chart versus candlestick chart . . . . . . . . . . . . . . . . . . . . 44

8.37 Candlesticks for long and short days . . . . . . . . . . . . . . . . . . 45
8.38 A comparison of an arithmetic and logarithmic scale . . . . . . . . . . 46
8.39 A 10 day moving average applied to a daily bar chart . . . . . . . . . 47
8.40 Exponential moving average versus simple moving average . . . . . . 48
8.41 One moving average as trading signal . . . . . . . . . . . . . . . . . . 49
8.42 Example of the double crossover method . . . . . . . . . . . . . . . . 50
8.43 Example of momentum use . . . . . . . . . . . . . . . . . . . . . . . . 51
8.44 Example of using momentum as indicator . . . . . . . . . . . . . . . . 52
8.45 Example of a histogram using two moving averages . . . . . . . . . . 53
8.46 Example of RSI oscillator . . . . . . . . . . . . . . . . . . . . . . . . 54

9.1 Transaction of cryptocurrency in blockchain . . . . . . . . . . . . . . 56

9.2 Cryptography jargon . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
9.3 Symmetric encryption . . . . . . . . . . . . . . . . . . . . . . . . . . 59
9.4 Asymmetric encryption . . . . . . . . . . . . . . . . . . . . . . . . . . 60
9.5 Hash functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
9.6 KeepKey Hardware Wallet . . . . . . . . . . . . . . . . . . . . . . . . 63
9.7 Cryptocurrency miners . . . . . . . . . . . . . . . . . . . . . . . . . . 66
9.8 Evolution of BTC-USD price . . . . . . . . . . . . . . . . . . . . . . . 69
9.9 Most common coin symbols . . . . . . . . . . . . . . . . . . . . . . . 71

10.1 Neural networks layers’ diagram . . . . . . . . . . . . . . . . . . . . . 80

10.2 LSTM Network diagram . . . . . . . . . . . . . . . . . . . . . . . . . 81
10.3 Modular Neural Network diagram . . . . . . . . . . . . . . . . . . . . 82

12.1 Simulation on random trade closures . . . . . . . . . . . . . . . . . . 91

12.2 ROI vs number of trades for ωp range . . . . . . . . . . . . . . . . . . 94
12.3 ROI vs number of trades for p range and ωp = 0.498 . . . . . . . . . . 95
12.4 ROI vs number of trades for p range and ωp = 0.5008 . . . . . . . . . 95
12.5 ROI vs number of trades for A range . . . . . . . . . . . . . . . . . . 96

13.1 Atom environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

13.2 Windows Command Prompt . . . . . . . . . . . . . . . . . . . . . . . 98

14.1 Diagram of the trading bot . . . . . . . . . . . . . . . . . . . . . . . . 100

14.2 Diagram of the backtesting . . . . . . . . . . . . . . . . . . . . . . . . 110

José Manuel Rodrı́guez Lomeña x

List of Tables

5.1 Electricity costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5.2 Budget of the project . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

9.1 PoW & PoS comparison . . . . . . . . . . . . . . . . . . . . . . . . . 66

9.2 Cryptocurrency with more market cap . . . . . . . . . . . . . . . . . 72

15.1 Backtesting results for a frequency of 15 minutes . . . . . . . . . . . . 122

15.2 Backtesting results for a frequency of 5 minutes . . . . . . . . . . . . 123
15.3 Backtesting results for a frequency of 30 minutes . . . . . . . . . . . . 123

xi
LIST OF TABLES

José Manuel Rodrı́guez Lomeña xii

Nomenclature

The next list describes the symbols used within the body of the document:
ADA Cardano
AI Artificial Intelligence
ANN Artificial Neural Network
ARIMA Autoregressive Integrated Moving-Average
ATH All-Time High
BNB Binance Coin
BTC Bitcoin
CCXT Cryptocurrency Exchange Trading Library
CEX Centralized Cryptocurrency Exchange
CSV Comma Separated Values
dApps Decentralized Applications
DeFi Decentralized Finance
DEX Decentralized Cryptocurrency Exchange
DLT Distributed Ledger Technology
ETH Ethereum
FOMO Fear Of Missing Out
GARCH Generalized Autoregressive Conditional Heteroscedasticity
GUI Graphical User Interface
ICO Initial Coin Offering
IPO Initial Public Offering
KYC Know-Your-Customer
LSTM Long-Short Term Memory
LTC Litecoin

xiii
NOMENCLATURE

MACD Moving Average Convergence/Divergence

ML Machine Learning
NFT Non-Fungible Token
NumPy Numerical Python
PKC Public Key Cryptography
PoS Proof-of-Stake
PoW Proof-of-Work
RNN Recurrent Neural Network
ROI Return Of Investment
RSI Relative Strength Index
SKC Secret Key Cryptography
SNN Simulated Neural Network
USD United States Dollar
USDC USD Coin
USDT Tether
XRP Ripple
XTZ Tezos

José Manuel Rodrı́guez Lomeña xiv

Abstract

As an emerging market and research direction, cryptocurrencies and cryptocurrency

trading have seen considerable progress and a notable upturn in interest and activity,
even entering the market people without experience or sufficient knowledge. In
addition, the tremendous volatility and the fact that this market never closes make
the human factor affect crypto asset trading too much.
Hence, in this project a cryptocurrency trading bot is developed. To be exact, the
algorithm consists of two distinguishable parts: the bot itself and the backtesting.
Notwithstanding that both parts departs from the analysis of financial markets in
general, and cryptocurrencies in particular, both present clear differences in terms of
code and pretext.
On the one hand, the bot’s algorithm is used to trade in reality, specifically through
the Binance exchange. Here the user plays risks their monetary capital. On the other
hand, backtesting consists of verifying the trading strategy based on historical data.
Backtesting serves, then, as validation of the strategy to be followed by the bot.
Thus, all the necessary fundamentals to understand both the general cryptocurrency
context and technical analysis relevant concepts are presented, along with a detailed
explanation of the implemented algorithm and a proper analysis of the obtained
results. Finally, further code improvements and new ideas to develop in the future
are suggested, apart from presenting the code developed.

xv
Chapter 1

Aim

Since their inception, cryptocurrency have been quickly evolving to the point of being
a notorious asset nowadays. However, this breakneck progression is accompanied by
a technical complication for both their understanding and their investment, which
no longer makes them so engrossing for them to be acquired by everyone.
Although the constant evolution of technology has meant that new tools to ease the
cryptocurrency investment have emerged, these tools usually entail large fees for the
small investor. In fact, there are a lot of cases where platforms that provide this
type of tools are only trying to make money through these fees, and they do not
really care about their client’s results. That is why this market is still reluctant for
those who do not fully understand it.
Despite of the above, it is doubtless that this kind of tools are especially helpful,
since they separate the mental factor of the investor from the investment itself, in
addition to the fact that it does not involve having to be constantly aware of it.
With all of the above, the aim of this project is to develop a cryptocurrency trading bot
programmed in Phyton. To achieve this, implementation of an algorithm connected
with Binance is carried out.

1
CHAPTER 1. AIM

José Manuel Rodrı́guez Lomeña 2

Chapter 2

Scope

This chapter consists of a brief explanation of what the lecturer will find (Sec. 2.1 In
Scope) and will not find (Sec. 2.2 Out of Scope) in this project.

2.1 In Scope
The points which this project covers are:
• Budget and environmental impact aspects.
• State of the art of cryptocurrency trading bots.
• Theoretical introduction to technical analysis of the financial markets.
• Theoretical introduction to cryptocurrency.
• Query of information for statistical and Machine Learning techniques.
• Briefly description of used Python libraries.
• Several studies on the bet trade percentage.
• Description of the used software.
• Study of the requirements which allow implementing an adequate algorithm.
• Detailed explanation of the bot and backtesting algorithms.
• Analysis of obtained data results for the backtesting of one specific trading
strategy.
• Conclusions and future work.

2.2 Out of Scope

The points that this project does not cover are:
• Development and implementation of more than one trading strategy.

3
CHAPTER 2. SCOPE

• Real results of the trading bot algorithm (trading in reality).

• Exhaustive study on trading strategy development.
• Analysis or implementation of any kind of strategy to trade in the stocks
market.

José Manuel Rodrı́guez Lomeña 4

Chapter 3

Justification

In recent years, the tendency of the number of financial institutions and private
individuals to include cryptocurrencies in their portfolios has accelerated. Cryptocur-
rency is playing an increasingly important role in reshaping the financial system due
to its growing popular appeal and merchant acceptance. While many people are
making investments in this recent market, the dynamical features, uncertainty, the
predictability of cryptocurrency are still mostly unknown, which dramatically risk
the investments.
A lot of risk factors appear within cryptocurrency trading. Now, there are mainly
five that really affect particular individuals who trade: the illusion of control, social
learning and reinforcement, preoccupation, fear of missing out, and anticipated regret.
Despite the fact that skill and strategy can make a difference to outcomes, cryp-
tocurrency trading offers many opportunities for people to overestimate the role that
applying specific types of knowledge might play in these outcomes and, conversely,
they do not realize that their strategy has a clear slope in the luck and chance
of what the market dictates. The illusion of control, defined as a subjective
overestimation of the objective ability to exert control (Langer, 1975 [44]), is known
to be a common feature of gambling (Wood & Clapham, 2005 [58]). The effect of
this risk is likely to be bolstered by other heuristics and biases, including: biased
or self-serving attributions (outcomes due to personal action rather than external
factors); hindsight bias (the outcome is seen as being hypothesised all along); and, the
hot-hand fallacy (perceptions of predictable momentum shifts or winning periods).
Many of these effects are likely to be strong during favourable market conditions.
As a result, traders will rarely be wrong in their choices and most decisions will be
positively reinforced. Traders will therefore, often falsely, infer contingency between
their actions and positive outcomes, an effect which is known to be stronger when the
probability of reinforcement is high (Blanco, 2011 [4]). As a result, traders may gain
a sense of invincibility or perception that they cannot lose and this may contribute to
greater risk taking, for example: speculation of large amounts in just one speculative
coin; not planning for strategies to exit the market at the right time; or, moving
money from a more balanced portfolio towards purchasing riskier altcoins.

5
CHAPTER 3. JUSTIFICATION

Crypto trading has also emerged during the era of social media. This has led to
the emergence of a strong social media culture of crypto advisors or influencers in
platforms such as Youtube. Searching online quickly shows that it is possible to find
at least one positive endorsement of at least one major coin. Some of these arise
from what appear to be more experienced and well-informed sources, but there are
many others that are entirely speculative, ill-informed, and potentially misleading
because they leave out key information [57].
Although similar promotional information has historically been available concern-
ing conventional shares, the volume of material, the interactivity and superficial
confidentiality of the information (i.e., the sense that one is in the know first, or is
getting some special insights) is much stronger. Promoters of particular coins can
show evidence of how much money they have already earned from buying in very
low, and they can use graphics with great effect to show the anticipated growth.
This can serve to create a sense of urgency and a need for immediate action. It
also encourages a culture of mutual social learning and reinforcement, in which
followers of channels seek to promote their successes, while also reading about the
gains scored by others.
Preoccupation is a recognised feature of most major conceptual models of addiction
(Browne & Rockloff, 2019 [7]). Those who engage excessively in a particular activity
often find it difficult to disengage from the activity. They may continuously think
about the activity (preoccupation) and prioritize the activity ahead of other essential
responsibilities. Crypto trading would appear to be an activity that has the potential
to be highly absorbing. Like day-trading, it involves regular scrutiny of price
movements, news and other online media about coin-related developments, the need
to make regular buy and sell decisions and research into the different coins. However,
because crypto markets operate continuously, it is possible for people to be engaged
with the activity at any hour of the day. This creates the potential for crypto
trading to absorb a considerable amount of time and potentially with greater risk of
disruptions to sleep and other daily commitments.
One of the strongest psychological factors that appears to influence crypto-trading
is the Fear Of Missing Out (FOMO). Traders are confronted with displays of
hundreds of coins. Some of them, they already own; others they do not. If one
which they have purchased is going up rapidly, they may regret having not made a
larger investment. If another unpurchased coin is going up which they had previously
considered, they feel annoyed for having missed out on the opportunity. Perhaps
most problematically of all is the situation, when they observe a ‘green screen of
numbers’. The market is going up and they feel compelled to be part of the action.
They purchase a coin when it has reached a short-term peak, only to watch the
price fall soon afterwards. FOMO also applies to sell decisions. When altcoins, in
particular, have rapidly increased in price, there is always the prospect that the rise
might continue. Instead of taking the profit, the person starts to dream of what they
might purchase if the price increases even more, but is then unprepared when the
price falls when the bull-run ends [57].
Cognitive psychology has recognised for some time that many decisions are based

José Manuel Rodrı́guez Lomeña 6

on the desire to minimise anticipated regret (Miller & Taylor, 1995 [45]). One of
the central findings in this area is that acts of commission (doing something) usually
led to stronger feelings of regret than acts of omission (not doing something). This
asymmetry has been used to explain why people are reluctant to sell shares which
they have held for some time. Acts of commission involving regret in trading would
include decisions where a coin holding is sold, only for it to rapidly appreciate in
value. Acts of omission involve situations where ultimately successful stocks or coins
were forgone (not purchased) for other investment decisions.
It is undoubtedly that both acts are likely to cause significant regret in trading and
that this may potentially be a risk factor that is strongly observed in this activity.
So much so that crypto market allows traders to observe the folly of their decisions
over a long period of time: what they missed out on (a coin that has considerably
increase in value in last six months), and what coins they sold too early. For this
reason, the act of commission and omission effects are both likely to be very strong.
Hence, there is no doubt that cryptocurrency trading carries with it a series of
psychological factors that really affect the individual, and in turn mean that only
the most experienced and skilled end up obtaining a net return when trading.
All of the above leads then to the need of finding an alternative method to trade in the
crypto market. Specifically, the idea is to develop a cryptocurrency trading bot based
on technical analysis and Machine Learning. This would allow operating by leaving
aside psychological human factors, and consequently obtaining net profitability more
easily.
Thus, it would be necessary to study how percentage of the money is invested in
each trade, to select the exchange to trade on and to connect it with the algorithm
developed, to define and implement a trading strategy, and to validate this strategy
by means of historical data.

7 José Manuel Rodrı́guez Lomeña

CHAPTER 3. JUSTIFICATION

José Manuel Rodrı́guez Lomeña 8

Chapter 4

Introduction

This last year master’s degree thesis is an academic project developed in order to get
a tough basis in the field of automated cryptocurrency trading. Hence, the knowledge
obtained is used for the implementation of a bot which would allow trading to be
made in reality.
In this line, it is notorious to remark that regardless of this volume only aims to
satisfy the academic requirements that a last year master’s degree thesis demands,
it is imperative to leave proof of the author’s prior knowledge on the field. This
previous knowledge about cryptocurrencies, the Binance exchange, technical analysis
of financial markets, object-oriented programming and Machine Learning, have
allowed this project to be carried out in the context of a master’s degree thesis.
With that, what follows is a general explanation of what the lecturer will read,
focusing on giving an overall idea of how the project itself is presented.
First of all, Chap. 5 and Chap. 6 are related to the estimation of the budget and the
environmental impact of this project. They serve to give the reader an idea of the
expenses and the carbon footprint generated.
In Chap. 7, a state of the art description is made, focusing on nine different cryptocur-
rency trading bots. They all served as a source of inspiration for the development of
the algorithm implemented in this project.
From Chap. 8 to Chap. 11, both included, the fundamentals in which the algorithm
are based on are presented. Chap. 8 states constancy of the actual techniques and
tools used in technical analysis. Chap. 9 serves as an introduction to cryptocurrencies
and trading bots, and all the concepts that come with them. Later in Chap. 10,
statistical and Machine Learning techniques applied to cryptocurrencies are presented.
At turn, Chap. 11 states constancy of the Python libraries used to develop the trading
bot code.
The proposal of Chap. 12 is to present some relevant studies whose results are needed
to decide what percentage of the total capital is risked in each trade. First of all,
a code based on repetition and on random trade closures is simulated. After that,

9
CHAPTER 4. INTRODUCTION

a mathematical development on that percentage is carried out, focusing on some

variables of interest.
Chap. 13 describes the software used in this project. In turn, Chap. 14 is related
to the software implementation, providing a study of the requirements that the
algorithm must accomplish and detailing its development.
In Chap. 15 are presented the backtesting results, which serves to validate the trading
strategy which would be followed by the bot. The, Chap. 16 presents the conclusions
of the whole project, and Chap. 17 aims to leave proof of some comments and
indications to future developments.
Finally, Appendix A and Appendix B form the appendix of the project. They are
respectively related to the trading bot algorithm, to the code developed in Chap. 12
to carry out the studies on the bet trade percentage.

José Manuel Rodrı́guez Lomeña 10

Chapter 5

Budget

The proposal of this chapter is to collect the expenses generated for the realisation
of the project, in such a way that the budget of the same can be estimated. Before
that, the unit cost of all expenses is calculated, taking into account the depreciation
of physical assets.

5.1 Breakdown of Unit Costs

Due to the depreciation that physical assets suffer over time, the amortisation cost is
calculated as:

VP − VR
a= (5.1)
na

a
rh = (5.2)
nh

Where:
• a: amortisation [e/year]
• VP : purchase value [e]
• VR : residual value at the end of the amortisation period [e]
• na : amortisation period [years]
• rh : hourly rate [e/h]
• nh : working time per year [hours/year]
Hence, for the calculation of the working time per year, 40 hours of weekly work are
considered during the 45 non-holiday weeks of each year, which make a total of:

hours weeks hours

nh = 40 · 45 = 1800
week year year

11
CHAPTER 5. BUDGET

Computer
The personal laptop used during the project is a LENOVO Ideapad330, whose
purchase value VP is 680e, over an amortisation period of 5 years, and a residual
value VR of 20%.

680 − 136
a= = 108.8e/year
5

108.8
rh = = 0.060e/h
1800

Software
Secondly, the license price of the used programs has to be included. All of them are
detailed in the following list.
• Microsoft Office 2010. Computing tool. Used for analyzing data. License price:
126e/year.

126
rh = = 0.070e/h
1800
• GitHub Desktop. Cloud-based repository. Used for code control version.
License price: free.
• Atom. Programming tool. Used for editing code. License price: free.
• MATLAB R2021b. Engineering tool. Used for developing studies. License
price: free (college agreement).
• Mendeley Desktop. Referencing tool. Used for referencing. License price: free
(college agreement).
• LATEX, latex text editor. Used for the writing of the report. License price: free.
• Adobe Reader. Reader. Used for reading articles. License price: free.
• Google Chrome. Web browser. Used for the query of information. License
price: free.

Personnel Cost
The total time dedication to carry out the present project is a total of 300 hours.
Here, an engineer’s salary is considered to be 25e per hour.

José Manuel Rodrı́guez Lomeña 12

 5.2. TOTAL BUDGET

Electricity
It is considered both the electricity needed to charge the laptop and due to the light.
The laptop has been used full time during the development of the project, thus 300
hours. Since its autonomy is 6 hours, 50 charges have been made. Considering that
the computer is fully charged in 3.5 hours, a total of 175 hours is obtained.
On the other hand, it is estimated that the use of the light has been only required
for 75% of the time, i.e. 225 hours.
With that, Table (5.1) presents each of the above costs, considering an electricity
average cost of 0.25e/kW h1 .

Table 5.1: Electricity costs

Concept Power [kW ] Hours [h] Energy [kW h] Cost [e]

Laptop 0.12 175 21 5.25
Light 0.24 225 54 13.5
TOTAL COST 18.75

5.2 Total Budget

In view of the above results, Table (5.2) details the costs in each field and provides
the budget of the project.

Table 5.2: Budget of the project

Concept Period [h] rh [e/h] Cost [e]

Computer 300 0.060 18
Microsoft Office 50 0.070 3.5
Personnel cost 300 25 7,500
Electricity 18.75
TOTAL COST 7,540.25

At that point, it is essential to note that the personnel cost has not had to be assumed,
as it is the author himself who carries out the project. Therefore, the actual spending
has been 40.25e.

1
This price has been obtained from the monthly and daily averages of the development period of
the project, in the place of residence of the author.

13 José Manuel Rodrı́guez Lomeña

CHAPTER 5. BUDGET

José Manuel Rodrı́guez Lomeña 14

Chapter 6

Environmental Impact

The study of environmental impact is a fundamental point in the development of

any project. Specifically, it is presented the carbon footprint of this project. In
other words, it is calculated how many kilos of CO2 are dispersed by the atmosphere
because of this project.
As commented in Chap. 5, this project has been carried out only by means of a
laptop. Although actually its manufacture leaves an environmental footprint, it is not
considered in the impact of this project owing to its use time is negligible in relation
to their useful life. Hence, what does have a clear influence on the environment is
spent electricity.
From Table (5.1), it can be seen that the total electrical energy that has been
expanded during the whole project is 75 kW h. Since the electricity company that
supplies the author’s house is Endesa Energı́a S.A., and according to [10] its impact
carbon factor is of 0.38 kg of CO2 per kW h of electrical energy:

kg CO2
MCO2 = 0.38 · 75 kW h = 28.5 kg
1 kW h
Taking into account that electricity due to the light is of 54 kW h, which supposes a
a 72% of the total, it is clearly that most of the environmental footprint is caused
by the use of light. Hence, it would be captivating to reduce that impact by using
sunlight during the day. To achieve this, the workspace would have to be adapted
and take advantage of the natural light that enters through the windows.

15
CHAPTER 6. ENVIRONMENTAL IMPACT

José Manuel Rodrı́guez Lomeña 16

Chapter 7

State of the Art

In this chapter is presented a brief state of the art description regarding to cryp-
tocurrency trading bots1 . Specifically, nine different trading bots of the main ones
existing in the market are presented. Notwithstanding that the intention is mainly
to illustrate which trading bots operates nowadays, they also serve as a source of
inspiration for the present project.

7.1 Capfolio
Capfolio is a next generation cryptocurrency trading platform with an all-in-one
solution for beginner, intermediate and advanced cryptocurrency traders. It allows
testing out the crypto markets, cloning leaders, or even building a sophisticated
trading algorithm [8].

Capfolio is a secure, low-latency, institutional quality platform, in which a wide range

of tools is available: from data gathering and signal tracking, through backtesting2
and strategy development, to automated trading. This platform was built by crypto
traders for crypto traders of all levels, aiming at offering everyone a straightforward
access to the cryptocurrency markets.

Among the great diversity of tools Capfolio offers, basktesting stands out, since there
is no need for complex programming or in-depth statistical knowledge, which makes
backtesting easier than usual. Furthermore, it is also worth noting that it allows
operating on five different exchanges3 [8].

1
The reader unfamiliar with cryptocurrency trading bots is recommended to read Sec. 9.2
Cryptocurrency Bot.
2
The reader unfamiliar with backtesting is recommended to read Sec. 9.2.1 Backtesting.
3
The reader unfamiliar with cryptocurrency exchanges is recommended to read Sec. 9.1.8
Cryptocurrency Exchange.

17
CHAPTER 7. STATE OF THE ART

7.2 3Commas
3Commas helps users grow their crypto investments using a feature-rich terminal
and proven automated bots that operate on 17 major exchanges [1].
3Commas helps traders win regardless of market conditions. offering always a trading
strategy that can profit from it. 3Commas bots happen to be really good at reducing
average acquisition costs, directly increasing the profit margins from each trade.
Apart from its security, based on the fact that it only works with exchanges using API
keys4 , and an innovative trade automation focused on executing trading strategies
at scale, 3Commas offers the possibility of taking profit and stop-loss orders at the
same time.

7.3 CCXT
The Cryptocurrency Exchange Trading Library (CCXT) is a cryptocurrency trading
system with a unified API out of the box and optional normalized data. It supports
many Bitcoin and altcoin exchange markets and merchant APIs [19].
Expert traders can create a trading strategy based on this data and access public
transactions through the APIs [22]. The CCXT library is used to connect and
trade with cryptocurrency exchanges and payment processing services worldwide. It
provides quick access to market data for storage, analysis, visualization, indicator
development, algorithmic trading, strategy backtesting, automated code generation
and related software engineering.
As commented, it is designed for coders, skilled traders, data scientists and financial
analysts to build trading algorithms. Hence, current CCXT features include [19]:
• Support for many cryptocurrency exchanges.
• Fully implemented public and private APIs.
• Optional normalized data for cross-exchange analysis and arbitrage.
• Out-of-the-box unified API, very easy to integrate.

7.4 Blackbird
Blackbird is a C++ trading system that automatically executes long and short
arbitrage between Bitcoin exchanges. It can generate market-neutral strategies that
do not transfer funds between exchanges [20].
The motivation behind Blackbird is to naturally profit from these temporary price
differences between different exchanges while being market neutral. Unlike other
4
An API key is used as a form of authentication to provide users with authorized access to the
data it returns. The authentication access is provided in the form of a secret token. In simple
words, an API key is akin to a password that lets the API confirm your identity.

José Manuel Rodrı́guez Lomeña 18

 7.5. STOCKSHARP

Bitcoin arbitrage systems, Blackbird does not sell but actually short sells Bitcoin
on the short exchange. This feature offers two important advantages. Firstly, the
strategy is always market agnostic: fluctuations (rising or falling) in the Bitcoin
market will not affect the strategy returns, which at turn eliminates the huge risks
of this strategy. Secondly, this strategy does not require transferring funds between
Bitcoin exchanges. Buy and sell transactions are conducted in parallel on two different
exchanges, which involves in no need to deal with transmission delays [19].

7.5 StockSharp
StockSharp is an open-source trading platform for trading at any market of the
world including 48 cryptocurrency exchanges [27]. It has a free C# library and
free trading charting application. Manual or automatic trading (algorithmic trading
robot, regular or HFT) can be run on this platform. StockSharp consists of five
components that offer different features [19]:
• S#.Designer: Free universal algorithm strategy app, easy to create strategies;
• S#.Data: Free software that can automatically load and store market data;
• S#.Terminal: Free trading chart application (trading terminal);
• S#.Shell: Ready-made graphics framework that can be changed according to
needs and has a fully open source in C#;
• S#.API: A free C# library for programmers using Visual Studio. Any trading
strategies can be created in S#.API.

7.6 Freqtrade
Freqtrade is a free and open-source cryptocurrency trading robot system written
in Python. It is designed to support all major exchanges and is controlled by
Telegram. It contains backtesting, mapping and money management tools, and
strategy optimization through Machine Learning [24]. Thus, Freqtrade has the
following features [19]:
• Persistence: Persistence is achieved through SQLite technology.
• Strategy optimization through Machine Learning: Use Machine Learning
to optimize your trading strategy parameters with real trading data.
• Marginal Position Size: Calculates winning rate, risk-return ratio, optimal
stop-loss and adjusts position size, and then trades positions for each specific
market.
• Telegram management: use telegram to manage the robot.
• Dry run: Run the robot without spending money;

19 José Manuel Rodrı́guez Lomeña

CHAPTER 7. STATE OF THE ART

7.7 CryptoSignal
CryptoSignal is a professional technical analysis cryptocurrency trading system [23].
Investors can track over 500 coins of thirteen different exchanges. Therefore, Cryp-
toSignal presents an automated technical analysis which includes momentum, RSI,
Ichimoku Cloud, MACD, among others; and also a system which gives alerts through
email or Telegram. Additionally, it offers modular code for easy implementation of
trading strategies, and it is easy to install with Docker5 [19].

7.8 Ctubio
Ctubio is a C++ based low latency (high frequency) cryptocurrency trading system
[26]. This trading system can place or cancel orders through supported cryptocurrency
exchanges in less than a few milliseconds. Moreover, it provides a charting system
that can visualize the trading account status including trades completed, or even
target position for fat currency [19].

7.9 Catalyst
Catalyst is a cryptocurrency trading system which makes trading strategies easy to
express and backtest them on historical data (daily and minute resolution), providing
analysis and insights into the performance of specific strategies [21]. Also, it integrates
statistics and Machine Learning libraries (such as matplotlib, scipy, statsmodels and
sklearn) to support the development, analysis and visualization of the latest trading
systems [19].

7.10 Using Machine Learning

Once the nine cryptocurrency trading bots have been presented, it is notorious to
remark the import role of Machine Learning (ML) technology in these current systems,
since it constructs computer algorithms that automatically improve themselves by
finding patterns in existing data without explicit instructions [19].
Machine Learning is increasingly used in the prediction of cryptocurrency returns6 ,
by means of algorithms which solve both classification and regression problems from
a methodological point of view. Indeed, different kind of ML algorithms are present
in the vast majority of trading bots.

5
Docker is an open platform for developing, shipping, and running applications.
6
Further information about Machine Learning in cryptocurrency is presented in Sec. 10.1 Machine
Learning.

José Manuel Rodrı́guez Lomeña 20

Chapter 8

Technical Analysis of the Financial

Markets

It is prominent that financial markets can be analyzed from both the technical and
the fundamental point of view. While technical analysis concentrates on the study
of market action, fundamental analysis focuses on the economic forces of supply
and demand that cause prices to move higher, lower, or stay the same. Hence,
the fundamental approach examines all of the relevant factors affecting the price
of a market in order to determine the intrinsic value of that market, whereas the
important point for the technician is the effect.
Although it is essential having at least a passing awareness of the fundamentals,
trading bots are usually based on technical analysis. For that reason, this section aims
at stating constancy of the actual techniques and tools used in technical analysis, by
means of Technical Analysis of the Financial Markets from John J. Murphy [46]. A
reading of the original work by John J. Murphy is recommended for a more in-depth
treatment.

8.1 Philosophy of Technical Analysis

There are three premises on which the technical approach is based:
1. Market action discounts everything: That means that anything that
can possible affect the price - fundamentally, politically, psychologically, or
otherwise - is actually reflected in the price of the market.
2. Prices move in trends: The identification of trends in early stages of their
development allows trading in the direction of those trends.
3. History repeats itself : Patterns that have worked well in the past are
assumed that they will continue to work well in the future, since they are based
on the study of human psychology, which tends not to change. Future is thus
understood just as a repetition of the past.

21
CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

8.2 Dow Theory

Charles Dow made his reputation as a financial expert late 1890’s, when his writings
formed the basis for the Dow Theory in market analysis. In fact, Dow Theory still
forms the cornerstone of the study of technical analysis, even in the face of today’s
sophisticated computer technology, and the proliferation of newer and supposedly
better technical indicators. Hereinafter, the six basic tenets of Dow Theory are
described [46].

1. The Averages Discount Everything

This idea is based on the fact that the markets reflect every possible knowable factor
that affects overall supply and demand. It is one of the basic premises of technical
theory, and argues that while markets cannot anticipate events such as earthquakes,
they quickly discount such occurrences, and almost instantaneously assimilate their
affects into the price action.

2. The Market Has Three Trends

Before discussing how trends behave, it must be clarified what Dow considered a
trend. He defined an uptrend as a situation in which each successive rally closes
higher than the previous rally high, and each successive rally low also closes higher
than the previous rally low. In other words, an uptrend has a pattern of rising peaks
and troughs. The opposite situation, with successively lower peaks and troughs,
defines a downtrend.

Therefore, Dow considered a trend to have three parts: primary, secondary, and
minor, which he compared to the tide, waves and ripples of the sea. Specifically, the
primary trend represents the tide, the secondary or intermediate trend represents the
waves that make up the tide, and the minor trends behave like ripples on the waves.

An observer can thus determine the direction of the tide by noting the highest point
on the beach reached by successive waves. If each successive wave reaches further
inland than the preceding successive wave recedes, the tide has turned out and is
ebbing. Unlike actual ocean tides, which last a matter of hours, Dow conceived of
market tides as lasting for more than a year, and possible for several years.

The secondary, or intermediate, trend represents corrections in the primary trend and
usually lasts three weeks to three months. These intermediate corrections generally
retrace between one-third and two-thirds of the previous trend movement and most
frequently about half, or 50%, of the previous move.

Lastly, the minor trend often lasts less than three weeks. This trend represents
fluctuations in the intermediate trend.

Greater detail on trend concepts is presented in Sec. 8.3 Basic Concepts of Trends,
obviously by using that same basic concepts and terminology.

José Manuel Rodrı́guez Lomeña 22

 8.2. DOW THEORY

3. Major Trends Have Three Phases

Dow focused his attention on primary or major trends, which he felt usually take
place in three distinct phases: an accumulation phase, a public participation phase,
and a distribution phase.

The accumulation phase represents informed buying by the most astute investors. If
the previous trend was down, then at this point astute investors recognize that the
market has assimilated all the so-called bad news. The public participation phase, in
which most technical trend-followers begin to participate, occurs when prices begin
to advance rapidly and business news improves. Finally, the distribution phase takes
place when bullish stories increasingly appear; when economic news is better than
ever; and when speculative volume and public participation increase. During this
last phase, the same informed investors who began to ”accumulate” near the bear
market bottom, begin to ”distribute” before anyone else starts selling.

4. The Averages Must Confirm Each Other

Dow meant that no important bull or bear market signal could take place unless both
averages gave the same signal, thus confirming each other. He felt that both averages
must exceed a previous secondary peak to confirm the inception or continuation of a
bull market. He did not believe that the signals had to occur simultaneously, but
recognized that a shorter length of time between the two signals provided stronger
information. Hence, when the two averages diverged from one another, Dow assumed
that the prior trend was still maintained.

5. Volume Must Confirm the Trend

Dow recognized volume as a secondary but important factor in confirming price

signals. Simply stated, volume should expand or increase in the direction of the
major trend. In a major uptrend, volume would then increase as prices move higher,
and diminish as prices fall. In a downtrend, volume should increase as prices drop
and diminish as they rally.

6. A Trend Is Assumed to Be in Effect Until It Gives Definite Signals

That It Has Reversed

This tenet forms much of the foundation of modern trend-following approaches. A

number of technical tools are available to traders to assist in the difficult task of
spotting reversal signals, including the study of support and resistance levels, price
patters, trendlines, and moving averages. Some indicators can provide even earlier
warning signals of loss of momentum. All of that not withstanding, the odds generally
favor that the existing trend will continue.

23 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

Figure 8.1: Failure Swing [46] Figure 8.2: Nonfailure Swing [46]

In Fig. 8.1, notice that the rally at poiint C is lower than the previous peak at A.
Price then declines below point B. The presence of these two lower peaks and two
lower troughs gives a clear-cut sell signal at the point where the low at B is broken
(point S). This reversal pattern is sometimes referred to as a failure swing.

At turn, in Fig. 8.2, the rally top at C is higher than the previous peak at A. Then
price declines below point B. Some Dow theorists would see a sell signal at S1, while
other would need to see a lower high at E before turning bearish at S2. This reversal
pattern is referred to as a nonfailure swing, and is much more powerful pattern than
the failure swing.

Fig. 8.3 and Fig. 8.4 show the same scenarios at a market bottom, respectively. In
Fig. 8.3, it can be clearly observed that the buy signal takes place when point B is
exceeded (at B1). At turn, Fig. 8.4 shows the buy signal occurrences at points B1 or
B2.

Figure 8.3: Failure Swing Bottom [46] Figure 8.4: Nonfailure Swing Bottom [46]

José Manuel Rodrı́guez Lomeña 24

 8.3. BASIC CONCEPTS OF TRENDS

8.3 Basic Concepts of Trends

The concept of trend is absolutely essential to the technical approach to market
analysis. All of the tools used by the chartist - support and resistance levels, price
patterns, moving averages, trendlines, etc. - have the sole purpose of helping to
measure the trend of the market for the purpose of participating in that trend.

In a general sense, the trend is simply the direction of the market: which way it is
moving. Even though, at that point it should be considered that markets do not
generally move in a straight line in any direction, but are characterized by a series of
zigzags, which resemble a series of successive waves with fairly obvious peaks and
troughs. Therefore, it is the direction of those peaks and troughs that constitutes
market trend.

Trend Has Three Directions

With all of the above, an uptrend would be defined as a series of successively higher
peaks and troughs, as shown in Fig. 8.5; a downtrend is just the opposite, a series
of declining peaks and troughs (Fig. 8.6); and horizontal peaks and troughs would
identify a sideways price trend, as evidenced in Fig. 8.7.

Figure 8.5: Example of an uptrend with ascending peaks and troughs [46]

Figure 8.6: Example of a downtrend with descending peaks an troughs [46]

25 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

Figure 8.7: Example of a sideways trend with horizontal peaks and troughs [46]

It is absolutely relevant to be aware of the distinction of having a sideways trend

(more commonly referred to as trendless), in which prices move in a flat, horizontal
pattern that is referred to as a trading range. This type of sideways action reflects a
period of equilibrium in the price level where the forces of supply and demand are in
a state of relative balance.

Trend Has Three Classifications

In addition to having three direction, trend is usually broken down into the three
categories mentioned in Sec. 8.2 Dow Theory: major, intermediate, and near term
trends. Hence, each trend becomes a portion of its next larger trend. For example,
in Fig. 8.8, the major trend is up as reflected by the rising peaks and troughs (points
1, 2, 3, 4). The corrective phase (2-3) represents an intermediate correction within
the major uptrend. But notice that the wave 2-3 also breaks down into three smaller
waves (A, B, C). At point C, the analyst would say that the major trend was still
up, but the intermediate and near terms trends was still up, but the intermediate
near terms trends were down. At point 4, all three trends would be up.

Figure 8.8: Example the three degrees of trend: major, secondary, and near term [46]

As a general statement, most trend-following approaches focus on the intermediate

trend, which may last for several months. The near term trend is used primarily for
timing purposes. In an intermediate uptrend, short term setbacks would be used to
initiate long positions.

José Manuel Rodrı́guez Lomeña 26

 8.3. BASIC CONCEPTS OF TRENDS

Support and Resistance

It is eminent that prices move in a series of peaks and troughs, and that the direction
of those peaks and troughs determined the trend of the market. It is precisely from
those peaks and troughs that the concepts of support and resistance arises.

The troughs, or reactions lows, are called support. The term is self-explanatory and
indicates that support is a level or area on the chart under the market where buying
interest is sufficiently strong to overcome selling pressure. As a result, a decline is
halted and prices turn back up again. Usually a support level is identified beforehand
by a previous reaction low. In Fig. 8.9, points 2 and 4 represent support levels in an
uptrend.

On the other hand, resistance is the opposite of support and represents a price level
area over the market where selling pressure overcomes buying pressure and a price
advance is turned back. Usually a resistance level is identified by a previous peak.
In Fig. 8.9, points 1 and are resistance levels.

For an uptrend to continue, each successive low (support level) must be higher than
the one preceding it, apart from each rally high (resistance level) being higher than
the one before it. In case of the corrective dip in an uptrend coming all the way
down to the previous low, it may be an early warning that the uptrend is ending or
at least moving from an uptrend to a sideways trend. If the support level is violated,

Figure 8.9: Support and resistance levels in uptrend [46]

then a trend reversal from up to down is likely. Absolutely the same concept is for a
downtrend to continue, but of course the other way around.

With all of the above, it is important to note that support and resistance levels are
defined by: the amount of time that the price has remained in that level; the volume;
and the closeness in the time of the transaction, being the most recent activity the
most powerful one.

27 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

Trendlines
The basic trendline is one of the simplest of the technical tools employed by the
chartist, but is also one of the most valuable. An up trendline is a straight line
drawn upward to the right along successive reaction lows as shown by the solid line
in Fig. 8.10. On the contrary, a down trendline is drawn downward to the right along
successive rally peaks as shown in Fig. 8.11.

Figure 8.10: Example of an up trendline [46]

Figure 8.11: Example of an down trendline [46]

A trendline becomes very useful in a variety of ways. Once a trend assumes a certain
slope or rate of speed, as identified by the trendline, it will usually maintain the
same slope. The trendline then helps not only to determine the extremities of the
corrective phase, but also to notice when that trend is changing.

José Manuel Rodrı́guez Lomeña 28

 8.3. BASIC CONCEPTS OF TRENDS

In an uptrend, as in Fig. 8.12, the inevitable corrective dip will often touch or come
very close to the up trendline. Because the intent of the trader is to buy dips in an
uptrend, that trendline provides a support boundary under the market that can be
used as a buying area. In the same way, a down trendline can be used as a resistance
area for selling purposes, as can be seen in Fig. 8.13.

Figure 8.12: Breaking of an up trendline [46]

In both Fig. 8.12 and Fig. 8.13, the violation of the trendline at point 9 indicates a
trend change, calling for liquidation of all positions in the direction of the previous
trend. Very often, the breaking of the trendline is one of the best early warnings of a
change in trend.

Figure 8.13: Breaking of a down trendline [46]

Channel Line
The channel line is an useful variation of the trendline technique. Sometimes prices
trend between two parallel lines - the basic trendline and the channel line. Obviously,
when this is the case and when the analyst recognizes that a channel exists, this
knowledge can be used to profitable advantage.
In an uptrend, as in Fig. 8.14, the basic up trendline can be used for the initiation of
new long positions, while the channel line can be used for short term profit taking,
or even to initiate a countertrend short position. Similarly, Fig. 8.15 represents the
opposite situation.

29 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

Figure 8.14: Example of an uptrend channel [46]

Figure 8.15: Example of a downtrend channel [46]

As in the case of the basic trendlines, the longer the channel remains intact and the
more often it is successfully tested, the more important and reliable it becomes.

8.4 Volume
Most of the above discussion of charting theory has mainly concentrated on price
action. Nevertheless, volume plays also an essential role in the forecasting process.

Volume is the number of entities traded during the time period under study. It is
often plotted by a vertical bar at the bottom of the char, just under the price action,
as in Fig. 8.16.

José Manuel Rodrı́guez Lomeña 30

 8.5. CHART PATTERNS

Figure 8.16: Example of volume bars [46]

In Fig. 8.16, notice that volume bars are noticeable larger as prices are rallying (see
circles). That means that volume is confirming the price rise and is bullish.
As a general rule for interpreting volume, if it is increasing, then the current price
trend will probably continue in its present direction (either up or down). If, however,
volume is declining, the action can be viewed as a warning that the current price
trend may be nearing an end. Hence, volume should increase or expand in the
direction of the existing price trend. In an uptrend, volume should be heavier as the
price moves higher, and should decrease or contract on price dips. As long as this
pattern continues, volume is said to be confirming the price trend.

8.5 Chart Patterns

It would be a mistake to assume that most changes in trend are very abrupt affairs.
the fact is that important changes in trend usually require a period of transition.
The point is that these periods of transition do not always signal a trend reversal.
Sometimes these sideways periods just indicate a pause or consolidation in the
existing trend after which the original trend is resumed.
Hence, the study of these transition period and their forecasting implications leads
to the question of price patterns, which are no more than formations that appear on
price charts and can be classified into different categories, and that have predictive
value.
There are two major categories of prices patterns: reversal and continuation. As
these names imply, reversal patterns indicate that a considerable reversal in trend

31 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

is taking place, whereas the continuation patterns suggest that the market is only
pausing for a while, possible to correct a near term overbought or oversold condition,
after which the existing trend will be resumed. The trick then is to distinguish
between the two types of patterns as early as possible during the formation of the
pattern itself.

For the reader to note, almost all the explanation below are justified in an uptrend,
only not to repeat the same, but they are absolutely valid for downtrends too,
obviously with the situation reversed.

8.5.1 Reversal Patterns

This section aims then at examining the three most commonly used major reversal
patterns: the head and shoulders, triple tops and bottoms, and double tops and
bottoms.

But before beginning that discussion of the individual major reversal patterns, there
are a few preliminary points to be considered that are common to all of these reversal
patterns:

1. A prerequisite for any reversal pattern is the existence of a prior trend.

2. The first signal of an impending trend reversal is often the breaking of an

important trendline.

3. The larger the pattern, the greater the subsequent move.

4. Topping patterns are usually shorter in duration and more volatile than bottoms.

5. Bottoms usually have smaller price ranges and take longer to build.

6. Volume is usually more important on the upside.

8.5.1.1 The Head and Shoulders

This is the best known and most reliable of all major reversal patterns. In fact, most
of the other reversal patterns are just variations of the head and shoulders and will
not require as extensive treatment.

Picture a situation in a major uptrend, where a series of ascending peaks and

troughs gradually begin to lose momentum. The uptrend then levels off for a while.
During this time, the forces of supply and demand are in relative balance. Once
this distribution phase has been completed, support levels along the bottom of the
horizontal trading range are broken and a new downtrend has been established. That
new downtrend now has descending peaks and troughs. Fig. 8.17 is presented so as
to better understand the pattern.

José Manuel Rodrı́guez Lomeña 32

 8.5. CHART PATTERNS

Figure 8.17: Example of a head and shoulders top [46]

At point A, the uptrend is proceeding as expected with no signs of a top. Volume

expands on the price move into new highs, which is normal. The corrective dip to
point B is on lighter volume, which is also to be expected. At point C, however, the
alert chartist might notice that the volume on the upside breakout through point
A is a bit lighter than on the previous rally. The change is not in itself of a major
importance, but a little yellow caution light goes on in the back of the analyst’s head.

Prices then begin to decline to point D and something even more disturbing happens.
The decline carries below the top of the previous peak at point A. Remember that, in
an uptrend, a penetrated peak should function as support on subsequent corrections.
The decline well under point A, almost the the previous reaction low at point B, is
another warning that something may be going wrong with the uptrend.

The market rallies again to point E, this time on even lighter volume, and is not able
to reach the top of the previous peak at point C. To continue an uptrend, each high
point must exceed the high point of the rally preceding it. The failure of the rally at
point E to reach the previous peak at point C fulfills half of the requirement for a
new downtrend - namely, descending peaks.

By this time, the major up trendline (line 1) has already been broken, usually at
point D, constituting another danger signal. But, despite all of these warnings, the
only known point is that the trend has shifted from up to sideways. This might be
sufficient cause to liquidate long positions, but not necessarily enough to justify new
short sales.

Also, the accompanying volume pattern plays an important role in the development
of the head and shoulders top, as it does in all price patterns. As a general rule, the
second peak (the head) should take place on lighter volume than the left shoulder.
This is not a requirement, but a strong tendency and an early warning of diminishing
buying pressure. The most relevant volume signal takes place during the third peak
(the right shoulder), where volume should be noticeably lighter than on the previous

33 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

two peaks, and then expand on the breaking of the neckline, decline during the return
move, and finally expand again once the return move is over.
As Fig. 8.18 shows, the inverse head and shoulders is pretty much a mirror image of
the topping pattern. Now, there are three distinct bottoms with the head (middle
trough) a bit lower than either of the two shoulders.

Figure 8.18: Example of an inverse head and shoulders [46]

8.5.1.2 Triple Tops and Bottoms

The triple top and the triple bottom, which are much rarer in occurrence, are just
slight variations of their respective head and shoulder pattern. The main difference
is that the three peaks or troughs are at about the same level, as can be observed in
both Fig. 8.19 and Fig. 8.20.
As far as volume is concerned, it tends to decline with each successive peak at the
top and should increase at the breakdown point. The triple top is not complete
until support levels along both of the intervening lows have been broken. Conversely,
prices must close through the two intervening peaks at the bottom to complete a
triple bottom. Heavy upside volume on the completion of the bottom is also essential.

Figure 8.19: Example of a triple top [46]

José Manuel Rodrı́guez Lomeña 34

 8.5. CHART PATTERNS

Figure 8.20: Example of a triple bottom [46]

8.5.1.3 Double Tops and Bottoms

A much more common reversal pattern is the double top or the double bottom.
Next to the head and shoulders, it is the most frequently seen and the most easily
recognized. Fig. 8.21 and Fig. 8.22 show both the top and bottom variety, respectively.

Figure 8.21: Example of a double top [46]

Figure 8.22: Example of a double bottom [46]

For obvious reasons, the top is often referred to as an M and the bottom as a W. The
general characteristics of these double patterns are similar to that of their respective
head and shoulders and triple patterns, except that only two peaks appear instead
of three. The volume pattern is similar as is the measuring rule.

35 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

8.5.2 Continuation Patterns

These patterns often indicate that the sideways price action on the chart is nothing
more than a pause in the prevailing trend, and that the next move will be in
the same direction as the trend that preceded the formation. In comparison with
reversal patterns, continuation patterns are usually shorter term in duration and
more accurately classified as near term or intermediate patterns.

8.5.2.1 Triangles
One of the most common continuation patterns is the triangle. There are three types
of triangles: symmetrical, ascending, and descending, respectively represented in
Fig. 8.23, Fig. 8.24 and Fig. 8.25.
Although each type of triangle has a slightly different shape and has different
forecasting implications, the minimum requirement for all triangles to be formed is
four reversal points, two per each trendline.

Symmetrical Triangle
In Fig. 8.23, the triangle actually begins at point 1, which is where the consolidation
in the uptrend begins. Prices pull back to point 2 and then rally to point 3. Point 3,
however, is lower than point 1. The upper trendline can only be drawn once prices
have declined from point 3.
Notice then that point 4 is higher than point 2. Only when prices have rallied
from point 4, the lower upslanting line can be drawn. It is at this point that the
formation of the symmetrical triangles arises, because of having four reversal points
and consequently tow converging trendlines.

Figure 8.23: Example of a bullish symmetrical triangle [46]

For this pattern, there is a time limit for its resolution, and that is the point where
the two lines meet - at the apex. As a general rule, prices should break out in the
direction of the prior trend somewhere between two-thirds to three-quarters of the
horizontal width of the triangle. That is, the distance from the vertical base on the
left of the pattern to the apex at the far right. Because two lines must meet at some
point, that time distance can be measured once the two converging lines are drawn.
An upside breakout is signaled by a penetration of the upper trendline. If prices

José Manuel Rodrı́guez Lomeña 36

 8.5. CHART PATTERNS

remain within the triangle beyond the three-quarters point, the triangle begins to
lose its potency, and usually means that prices will continue to drift out to the apex
and beyond.

Ascending Triangle
It is a variation of the symmetrical triangle. Notice that in Fig. 8.24, the upper
trendline is flat, while the lower line is rising. This pattern indicates that buyers are
more aggressive than sellers. It is considered a bullish pattern and is usually resolved
with a breakout to upside.

Figure 8.24: Example of an ascending triangle [46]

The measuring technique for the ascending triangle is relatively simple. Simply
measure the height of the pattern at its widest point and project that vertical
distance from the breakout point. This is just another of using the volatility of a
price pattern to determine a minimum price objective.

Descending Triangle
The descending triangle is just a mirror image of the ascending, and is generally
considered a bearish pattern. Notice in Fig. 8.25 the descending upper line and the
flat lower line. That indicates that sellers are more aggressive than buyers, and is
usually resolved on the downside. The downside signal is registered by a decisive close
under the lower trendline, usually on increased volume. A return move sometimes
occurs which should encounter resistance at the lower trendline.

Figure 8.25: Example of a descending triangle [46]

37 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

The measuring technique is exactly the same as the ascending triangle, in the sense
that the height of the pattern at the base to the left must be measured, and then
that distance must be projected down from the breakdown point.

8.5.2.2 Broadening Formation

This is an usual variation of the triangle and is relatively rare. It is actually an
inverted triangle. As pattern in Fig. 8.26 shows, the trendlines actually diverge in
the broadening formation, creating a picture that looks like an expanding triangle.

Figure 8.26: Example of a broadening top [46]

Here the volume pattern also differs, since the volume tends to expand along with
the wider price swings, representing then a market that is out of control and usually
emotional. That is associated with an unusual amount of public participation, which
most occurs at major market tops. Consequently, this pattern is usually a bearish
formation, and it generally appears near the end of a major bull market.

8.5.2.3 Flags and Pennants

The flag and pennant formations are quite common. They are usually treated
together because of being very similar in appearance, since they both tend to show
up at about the same place in an existing trend, and also have the same volume and
measuring criteria.
The flag and pennant represent brief pauses in a dynamic market move. In fact, one
of the requirements for both is that they be preceded by a sharp and almost straight
line move. They represent situations where a steep advance or decline has gotten
ahead of itself, and where the market pauses briefly to ”catch its breath” before
running off again in the same direction.
Flags and pennants are among the most reliable of continuation patterns and only
rarely produce a trend reversal. Fig. 8.27 and Fig. 8.28 respectively show what these

José Manuel Rodrı́guez Lomeña 38

 8.5. CHART PATTERNS

two patterns look like. To begin with, notice the steep price advance preceding the
formations on heavy volume. Notice also the dramatic drop off in activity as the
consolidation patterns form and then the sudden burst of activity on the upside
breakout.
The flag, as can be observed in Fig. 8.27, resembles a parallelogram or rectangle
marked by two parallel trendlines that tend to slope against the prevailing trend. In
a downtrend, the flag would have a slight upward slope.

Figure 8.27: Example of a bullish flag [46]

At turn, the pennant is identified by two converging trendlines and is more horizontal,
as can be seen in Fig. 8.28. It very closely resembles a small symmetrical triangle.
An important requirement is that volume should dry up noticeably while each of the
patterns is forming.
With that, the most relevant points of both patterns are:
1. They are both preceded by an almost straight line move on heavy volume.
2. Prices then pause for about one to three weeks on very light volume.
3. The trend resumes on a burst of trading activity.
4. Both patterns occur at about the midpoint of the market move.
5. The pennant resembles a small horizontal symmetrical triangle.
6. The flag resembles a small parallelogram that slopes against the prevailing
trend.
7. Both patterns take less time to develop in downtrends.

39 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

8. Both patterns are very common in the financial markets.

Figure 8.28: Example of a bullish pennant [46]

8.5.2.4 Wedge Formation

The wedge formation is similar to a symmetrical triangle both in terms of its shape
and the amount of time it takes to form. Like the symmetrical triangle, it is identified
by two converging trendlines that come together at an apex. In terms of the amount
of time it takes to form, the wedge usually lasts more than one month but not more
than three months, putting it into the intermediate category.

What distinguishes the wedge is its noticeable slant. The wedge pattern has a
noticeable slant either to the upside or the downside. As a rule, the wedge slants
against the prevailing trend. Therefore, a falling wedge is considered bullish and a
rising wedge is bearish, as Fig. 8.29 and Fig. 8.30 respectively show.

Notice that in Fig. 8.29 the bullish wedge slants downward between two converging
trendlines, while in the downtrend in Fig. 8.30, the converging trendlines have an
unmistakable upward slant.

Anyway, both wedges show up most often within the existing trend and usually
constitute continuation patterns.

José Manuel Rodrı́guez Lomeña 40

 8.5. CHART PATTERNS

Figure 8.29: Example of a bullish falling wedge [46]

Figure 8.30: Example of a bearish wedge [46]

8.5.2.5 Rectangle Formation

The rectangle formation represents a pause in the trend during which prices move
sideways between two parallel horizontal lines, as appreciated in Fig. 8.31 and
Fig. 8.32.
The rectangle usually represents just a consolidation period in the existing trend, and
is often resolved in the direction of the market trend that preceded its occurrence.
In terms of forecasting value, it can be viewed as being similar to the symmetrical
triangle but with flat instead of converging trendlines.
A decisive close outside either the upper or lower boundary signals completion of
the rectangle and points the direction of the trend. The market analyst must always
be on the alert, however, that the rectangular consolidation does not turn into a
reversal pattern. In the Fig. 8.31, for example, notice that the three peaks might
initially be viewed as a possible triple top reversal pattern.

41 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

Figure 8.31: Example of a bullish falling rectangle [46]

Figure 8.32: Example of a bearish rectangle [46]

8.5.3 Confirmation and Divergence

The principle of confirmation is one of the common themes running throughout the
entire subject of market analysis, and it is used in conjunction with its counterpart -
the divergence.
Confirmation refers to the comparison of all technical signals and indicators to ensure
that most of those indicators are pointing in the same direction and are confirming
one another.
Divergence is just the opposite of confirmation and refers to a situation where different
technical indicators fail to confirm one another. While it is being used here in a
negative sense, divergence is a valuable concept in market analysis, and one of the
best early warning signals of impending trend reversals.

José Manuel Rodrı́guez Lomeña 42

 8.6. CHART CONSTRUCTION

8.6 Chart Construction

In order to better comprehend the market data and properly analyze it, it is displayed
in charts, such as line charts, bar charts and more recently, candlesticks, which are
discussed in detail in Sec. 8.6.1 Candlestick Charting.
Fig. 8.33 shows a standard daily bar chart. It is called a bar chart because each
day’s range is represented by a vertical bar. Thus, the bar chart shows the open,
high, low, and closing prices. The tic to the right of the vertical bar is the closing
price. And the opening price is the tic to the left of the bar.

Figure 8.33: Example of a bar chart [46]

Fig. 8.34 shows what the same market looks like on a line chart, in which only the
closing price is plotted for each successive day.

Figure 8.34: Example of line chart [46]

43 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

8.6.1 Candlestick Charting

Charting market data in candlestick form uses the same data available for standard
bar charts: open, high, low, and close prices. While using the exact same data,
candlestick charts offer a much more visually appealing chart. Information seems to
jump off the page, and is more easily interpreted and analyzed. The box in Fig. 8.35
is a depiction of a single day of prices showing the difference between the bar (left)
and the candlesticks.

Figure 8.35: Example of candlestick [46]

The rectangle represents the difference between the open and close price for the day,
and is called the body. Notice that the body can be either black or white. A white
body means that the close price was greater (higher) than the open price, whereas
the black body means that the close price was lower than the open price.
Therefore, the open and close prices are given much significance in candlesticks. The
small lines above and below the body are referred to as wicks or hairs or shadows.
They represent the high and low prices for the day and are normally not considered
vital in the analysis.
Fig. 8.36 shows the same data in both the popular bar chart and the candlestick
format. Although initially it takes some getting use to candlesticks, after a while
their use ends up being better, since different shapes for candlesticks have different
meanings.

Figure 8.36: Bar chart versus candlestick chart [46]

José Manuel Rodrı́guez Lomeña 44

 8.6. CHART CONSTRUCTION

Firstly, it is necessary to distinguish between long and short days. Long days are
those in which the difference between the open and close prices is great, whereas
in short days that difference is small. Fig. 8.37 visually represents that difference.
Please note that the latter is only related to the size of the body and no reference is
mode to the high and/or low prices.

Figure 8.37: Candlesticks for long and short days [46]

From the identification of patterns with candlesticks, it is possible to forecast the

trend of the market. Hence, candlestick patterns are psychological depictions of
traders’ mentality at the time. It vividly shows the actions of the traders as time
unfolds in the market. The mere fact that humans react consistently during similar
situations makes candle pattern analysis work.

A lot of candlestick patterns exist and are so useful for trading purposes. Nevertheless,
they will not be collected here, due to the enormous amount of information that this
would entail. The reader interested in candlestick pattern analysis is recommended
to read Chapter 12. Japanese Candlesticks on [46].

8.6.2 Arithmetic versus Logarithmic Scale

Charts can be plotted using arithmetic or logarithmic price scales. For some types
of analysis, particularly for very long range trend analysis, there may be some
advantages to using logarithmic charts. Fig. 8.38 shows what the difference scales
would look like.

On the arithmetic scale, the vertical price scale shown an equal distance for each
price unit of change. Notice in this example that each point on the arithmetic scale
is equidistant. On the log scale, however, note that the percentage increases get
smaller as the price scale increases. The distance from points 1 to 2 is the same as
the distance from points 5 to 10 because they both represent the same doubling in
price.

45 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

Figure 8.38: A comparison of an arithmetic and logarithmic scale [46]

8.6.3 Long Term Charts

The short term bar chart covers a relatively short period of time in the life of
any market. A through trend analysis of a market, however, should include some
consideration of how the daily market price is moving in relation to its long range
trend structure. To accomplish that task, longer range charts must be employed,
because they provide a perspective on the market trend that is impossible to achieve
with the use of short term charts alone.

8.7 Moving Averages

The moving average is one of the most versatile and widely used of all technical
indicators. Because of the way it is constructed and the fact that it can be so easily
quantified and tested, it is the basis for many mechanical trend-following systems in
use today. Moving average rules can easily be programmed into a computer, which
then generates specific buy and sell signals. While two technicians may disagree as
to whether a given price pattern is a triangle or a wedge, or whether the volume
patterns favors the bull or bear side, moving average trend signals are precise and
not open to debate.

A moving average is an average of a certain body of data. For example, if a 10 day

average of closing prices is desired, the prices for the last 10 days are added up and
the total is divided by 10. The term moving is used because only the later 10 days’
prices are used in the calculation. Therefore, the body of data to be averaged (the
last 10 closing prices) moves forward with each new trading day. Fig. 8.39 shows a
10 day moving average applied to a daily bar chart.

José Manuel Rodrı́guez Lomeña 46

 8.7. MOVING AVERAGES

Figure 8.39: A 10 day moving average applied to a daily bar chart [46]

The above example deals with a simple 10 day moving average of closing prices. There
are, however, other types of moving averages that are not that simple. Moreover, it
is also relevant to decide how many days should be averaged, to decide either short
or long term analysis.
Despite the fact that some analysts average a midpoint value of the price during the
specific time span which is wanted to be analyzed, or even others make an average
of the high, low and closing prices, the most common is to use the closing price for
moving average analysis.

8.7.1 The Simple Moving Average

The simple moving average is the type used by most technical analysts. But there
are some who question its usefulness on two points. The first criticism is that only
the period covered by the average is taken into account. The second criticism is
that the simple moving averages gives equal weight to each day’s price. Hence, the
last day receives the same weight as the first day in the calculation. Some analysts
believe that a heavier weighting should be given to the more recent price action.

8.7.2 The Linearly Weighted Moving Average

In an attempt to correct the weighting problem, some analysts employ a linearly
weighted moving average. In this calculation, a more recent closing price has always
more significance than its previous one. For instance, a 10-day linearly weighted
moving average follows:

10xi + 9xi−1 + 8xi−2 + ... + 2xi−8 + xi−9

x¯M = (8.1)
55

where 55 is obtained by adding all the integers from 1 to 10, both included.

47 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

8.7.3 The Exponentially Smoothed Moving Average

This type of average addresses both of the problems associated with the simple
moving average. First, the exponentially smoothed average assigns a greater weight
to the more recent data. Therefore, it is a weighted moving average. But while it
assigns lesser importance to past price data, it does include in its calculation all of
the data in the life of the instrument. In addition, the user is able to adjust the
weighting to give greater or lesser weight to the most recent day’s price. This is
done by assigning a percentage value to the last day’s price, which is added to a
percentage of the previous day’s value. The sum of both percentage values adds up
to 100. For example, the last day’s price could be assigned a value of 10%, which is
added to the previous day’s value of 90%. That gives the last day 10% of the total
weighting. That would be the equivalent of a 20 day average. By giving the last
day’s price a smaller value of 5%, lesser weight is given to the last day’s data and the
average is less sensitive. That would be the equivalent of a 40 day moving average.
All of the above could be appreciated in Fig. 8.40.

Figure 8.40: Exponential moving average is more sensitive than the simple one [46]

8.7.4 Using Two Moving Averages

Some traders use just one moving average to generate trend signals. The moving
average is plotted on the bar chart in its appropriate trading day along with that
day’s price action. When the closing price moves above the moving average, a buy
signal is generated. On the contrary, a sell signal is given when prices move below
the moving average. This can be observed in Fig. 8.41.
Although is quite straightforward to use only one moving average, it has some
drawbacks. If a very short term average is employed (a 5 or 10 day), the average
tracks prices very closely and several crossings occur. This action can be either
good or bad. The use of a very sensitive average produces more trades (with higher

José Manuel Rodrı́guez Lomeña 48

 8.7. MOVING AVERAGES

commission costs) and results in many false signals (whipsaws). If the average is too
sensitive, some of the short term random price movement (or ”noise”) activates bad
trend signals.

Figure 8.41: One moving average as trading signal [46]

While the shorter average generates more false signals, it has the advantage of giving
trend signals earlier in the move. It stands to reason that the more sensitive the
average, the earlier the signals will be. So there is a tradeoff at work here. The
trick is to find the average that is sensitive enough to generate early signals, but
insensitive enough to avoid most of the random noise.

At turn, a longer average performs better while the trend remains in motion, but
it ”gives back” a lot more when trend reverses. The very insensitivity of the longer
average (the fact that it trailed the trend from a greater distance), which kept it
from getting tangled up in short term corrections during the trend, works against
the trader when the trend actually reverses. Therefore, the longer averages work
better as long as the trend remains in force, but a shorter average is better when the
trend is in the process of reversing.

It becomes clearer, thus, that the use of one moving average alone has several
disadvantages. It is usually more advantageous to employ two moving averages.

The technique of using two average to generate signals is called the double crossover
method. This means that a buy signal is produced when the shorter average crosses
above the longer. Specifically, when the shorter average exceeds the longer average,
a buy signal is generated, whereas the opposite case supposes a sell signal.

49 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

Figure 8.42: Example of the double crossover method [46]

8.8 Oscillators
Another alternative to trend-following approaches is the oscillator. It is extremely
useful in non-trending markets where prices fluctuate in a horizontal price band,
or trading range, creating a market situation where most trend-following systems
simply do not work that well. The oscillator provides the technical traders with
a tool that can enable them to profit from these periodic sideways and trendless
market environments.

Nevertheless, the value of the oscillator is not limited to horizontal trading ranges.
Used in conjunction with price charts during trending phases, the oscillator becomes
an extremely valuable ally by alerting the trader to short term market extremes,
commonly referred to as overbought or oversold conditions. The oscillator can also
warn that a trend is losing momentum before that situation becomes evident in the
price action itself. Oscillators can signal that a trend may be nearing completion by
displaying certain divergences.

Hence, there are three situation when the oscillator is more useful.

1. The oscillator is most useful when its value reaches an extreme reading near the
upper or lower end of its boundaries. The market is said to be overbought when
it is near the upper extreme and oversold when it is near the lower extreme.
This warns that the price trend is overextended and vulnerable.

2. A divergence between the oscillator and the price action when the oscillator is
in an extreme position is usually an important warning.

3. The crossing of the zero (or midpoint) line can give important trading signals
in the direction of the price trend.

José Manuel Rodrı́guez Lomeña 50

 8.8. OSCILLATORS

8.8.1 Measuring Momentum

The concept of momentum is the most basic application of oscillator analysis. Mo-
mentum measures the velocity of price changes as opposed to the actual price levels
themselves. Market momentum is measured by continually taking price differences
for a fixed time interval. To construct a x-day momentum line, simply subtract the
closing price N days ago from the last closing price. This positive or negative value
is then plotted around a zero line. The formula for momentum is:

M =V −Vx (8.2)

where V is the latest closing price and V x is the closing price x days ago.

Figure 8.43: A comparison of 10 and 40 day momentum lines [46]

If the latest closing price is greater than that of x days ago (prices have moved
higher), then a positive value would be plotted above the zero line. On the contrary,
if the latest close is below the close x days earlier (prices have declined), then a
negative value is plotted below the zero line, according to Equation (8.2).
In terms of momentum duration, the shorter the time period chosen, a more sensitive
line with more pronounced oscillations is produced. Hence, a longer period of time
results in a much smoother line in which the oscillator swings are less volatile. That
can be observed in Fig. 8.43.
Because of the way it is constructed, the momentum line is always a step ahead of
the price movement. It leads the advance or decline in prices, then levels off while the
current price trend is still in effect. It then begins to move in the opposite direction
as prices begin to level off. For instance, in Fig. 8.44 it can be observed that the value
of the momentum indicator is that it turns sooner than the market itself, making it
a leading indicator.

51 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

Figure 8.44: The trendlines on the momentum chart are broken sooner than those
on the price chart [46]

As expected, the momentum chart has a zero line, whose crossing is generally used to
generate buy and sell signals. A crossing above the zero line would be a buy signal,
and a crossing below the zero line, a sell signal. Nevertheless, oscillator analysis
should not be used as an excuse to trade against the prevailing market trend. For
that reason, buy positions should only be taken on crossing above the zero line if
the market trend is up. In the same way, short position should be taken on crossing
below the zero line only if the price trend is down.

8.8.2 Oscillator Using Two Moving Averages

Sec. 8.7.4 Using Two Moving Averages discussed two moving averages being used to
generate buy and sell signals. The crossing of the shorter average above or below the
longer average registered buy and sell signals, respectively. The point is that these
dual moving average combinations could also be used to construct oscillator charts.
This can be done by plotting the difference between the two averages as a histogram,
whose bars appear as a plus or minus value around a centered zero line. This type of
oscillator has three uses:
1. To help spot divergences.
2. To help identify short term variations from the long term trend, when the
shorter average moves too far above or below the longer average.
3. To pinpoint the crossing of the two moving averages, which occur when the
oscillator crosses the zero line.
The shorter average is divided by the longer. In both cases, however, the shorter
average oscillates around the longer average, which is in effect the zero line. As can
be seen in Fig. 8.45, if the shorter average is above the longer, the oscillator would

José Manuel Rodrı́guez Lomeña 52

 8.8. OSCILLATORS

be positive. A negative reading would be present if the shorter average were under
the longer.

Figure 8.45: Example of a histogram using two moving averages [46]

8.8.3 The Relative Strength Index (RSI)

One of the two major problems in constructing a momentum line (using price
differences) is the erratic movement often caused by sharp changes in the values
being dropped off. A sharp advance or a decline some time ago can cause sudden
shifts in the momentum line even if the current prices show little change. Some
smoothing is therefore necessary to minimize these distortions. The second problem
is that there is the need for a constant range for comparison purposes.
Hence, the RSI formula not only provides the necessary smoothing, but also solves
the latter problem by creating a constant vertical range of 0 to 100. The actual
formula is calculated as follows:

100
RSI = 100 − (8.3)
1 + RS

where:

Average of x days’ up closes

RS = (8.4)
Average of x days’ down closes

From Equation (8.3) can be deduced that the shorter the time period, the more
sensitive the oscillator becomes and the wider its amplitude. RSI works best when
its fluctuations reach the upper and lower extremes. Thus, if the user is trading
on very short term basis and wants the oscillator swings to be more pronounced,

53 José Manuel Rodrı́guez Lomeña

CHAPTER 8. TECHNICAL ANALYSIS OF THE FINANCIAL MARKETS

the time period can be shortened. The time period is then lengthened to make the
oscillator smoother and narrower in amplitude.
RSI is plotted on a vertical scale of 0 to 100. As displayed in Fig. 8.46, movements
above 70 are considered overbought, while an oversold condition would be a move
under 30. Because of shifting that takes place in bull and bear markets, the 80 level
usually becomes the overbought level in bull markets and the 20 level the oversold
level in bear markets.

Figure 8.46: Example of RSI oscillator [46]

8.8.4 The Importance of Trend

The importance of trading in the direction of the major trend cannot be overstated.
The danger in placing too much importance on oscillators by themselves is the
temptation to use divergence as an excuse to initiate trades contrary to the general
trend. This action generally proves a costly and painful exercise. The oscillator, as
useful as it is, is just one tool among many others and must always be used as an
aid, not substitute, for basic trend analysis.

8.8.5 When Oscillators Are Most Useful

There are times when oscillators are more useful than at other. Oscillators are really
useful in sideways trading, since they track the price movement very closely, and the
peaks and troughs on the price chart coincide almost exactly with the peaks and
troughs on the oscillator.
Furthermore, oscillators should be ignored at the beginning of a new trend, but
should be followed when the trend reaches maturity. The reason for this is that in
the early stages of a new trend, following an important breakout, oscillators often
reach extremes very quickly and stay there for awhile.

José Manuel Rodrı́guez Lomeña 54

Chapter 9

Introduction to Cryptocurrency

All the fundamentals related to cryptocurrency required for developing the project
are presented in this chapter. Notwithstanding that it is divided into two different
sections so as to facilitate understanding and explanation, they are not independent
of each other during the subsequent project.

9.1 Cryptocurrency
The proposal of this section is to define what a cryptocurrency is, and to outline
the main concepts related to cryptocurrency, such as blockchain, cryptography,
cryptocurrency wallet, and the difference between Proof-of-Work and Proof-of-Stake.
Moreover, an idea of their history, the types of cryptocurrencies, and cryptocurrency
exchange concept are also presented.

9.1.1 Meaning and Definition

A cryptocurrency is any form of currency that exists digitally or virtually, and
uses cryptography to protect transactions. Cryptocurrencies do not have a central
regulatory or issuing authority; instead, they use a decentralized system to record
transactions and issue new units.
Hence, a cryptocurrency is a digital payment system that does not rely on a third
person to verify transactions. It is a peer-to-peer system that can allow anyone
anywhere to send and receive payments. Instead of physical money that is transported
and exchanged in the real world, cryptocurrency payments exist solely in the form of
digital entries directed to an online database describing specific transactions.
Cryptocurrencies get their name because they use encryption to verify transactions.
This means that advanced encryption is required to store and transmit cryptocurrency
data between wallets and to public ledgers. Therefore, the purpose of encryption is
to provide security, so that when one transfers cryptocurrency funds, the transactions
are recorded in a public ledger, known as blockchain, and the cryptocurrencies
themselves are stored in digital wallets.

55
CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

9.1.2 Blockchain
As commented, cryptocurrencies operate on a distributed public ledger called a
blockchain, a record of all transactions held and updated by coin owners. Therefore,
a blockchain is a type of distributed ledger technology (DLT) that consists of growing
list of records. Specifically, block is referred to a collection of data, and chain to a
public database of these blocks, which are stored as a list.

By using a blockchain, participants can confirm transactions without a need for a cen-
tral clearing authority. That is the main reason why they are used in cryptocurrency
systems, as they allow to maintain a secure and decentralized record of transactions
without the need for a trusted third party [36].

One key difference between a typical database and a blockchain is how the data is
structured. A blockchain collects information together in groups, known as blocks,
that hold sets of information. Blocks have certain storage capacities and, when filled,
are closed and linked to the previously filled block, forming a chain of data known as
the blockchain. All new information that follows that freshly added block is compiled
into a newly formed block that will then also be added to the chain once filled.

Hence, this data structure inherently makes an irreversible timeline of data when
implemented in a decentralized nature. When a block is filled, it is set in stone and
becomes a part of this timeline. Each block in the chain is given an exact timestamp
when it is added to the chain [36].

9.1.2.1 Transaction Process

The goal of blockchain is to allow digital information to be recorded and distributed,
but not edited. In this way, a blockchain is the foundation for immutable ledgers, or
records of transactions that cannot be altered, deleted, or destroyed.

Figure 9.1: Transaction of cryptocurrency in blockchain [18]

José Manuel Rodrı́guez Lomeña 56

 9.1. CRYPTOCURRENCY

No matter what industry the blockchain is used for, each blockchain transaction goes
through the same steps shown in Fig. 9.1:

• A transaction is requested and authenticated from each party.

• A block representing that transaction is created, so that blocks are

added to the blockchain; transactions are retrieved in the previous block, which
is combined with the hash of the preceding block to obtain its hash, and then
the derived hash is stored into the current block [19].

• The block is sent to every node in the network, meaning the data is
replicated and stored instantaneously on each node across the system [5].

• Nodes validate the transaction. Confirmation is a critical concept in

cryptocurrencies; they must be accepted and marked as legitimate. The two
ways through which the validation is carried out are detailed in Sec. 9.1.5
Proof-of-Work against Proof-of-Stake.

• Nodes receive a cryptocurrency reward, as a way to send value for these

transactions.

• The block is added to the existing blockchain. After the transaction is

confirmed, each node must add it to its database. This process of transaction
verification and recording is immediate and permanent [5]. Moreover, each
block contains a code called a hash that is unique to that block. The block
carries its own hash and the hash of the block before it so that users always
know where the block should be located in the chain [54].

• The update is distributed across the network. When a transaction is

recorded in the blockchain, details of the transaction such as price, asset, and
ownership, are recorded, verified and settled within seconds across all nodes. A
verified change registered on any one ledger is also simultaneously registered
on all other copies of the ledger. Since each transaction is transparently and
permanently recorded across all ledgers, open for anyone to see, there is no
need for third-party verification [5].

• The transaction is complete. Blocks can contain many transactions. When

it is completed, it is added to the chain, and the hash that it carries ensures
that it is in proper chronological order [54].

In layman terms, the above process ensure that the transaction has become part of
the blockchain. Hence, blockchain is a platform that drives cryptocurrency and is a
technology that acts as a distributed ledger for the network, which creates a means of
transaction and enables the transfer of value and information. Therefore, this work
is undertaken to obtain cryptocurrency tokens, which can be thought of as tools on
the blockchain, and in some cases can also function as resources or utilities. In other
instances, they are used to digitise the value of assets. In summary, cryptocurrencies
are part of an ecosystem based on blockchain technology [19].

57 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

9.1.3 Cryptography
Within the context of any application-to-application communication, there are some
specific security requirements, including [42]:
• Authentication: The process of proving one’s identity.
• Privacy: Ensuring that no one can read the message except the intended
receiver.
• Integrity: Assuring the receiver that the received message has not been altered
in any way from the original.
• Non-repudiation: A mechanism to prove that the sender really sent this
message.
Cryptography, then, not only protects data from theft or alteration, but can also be
used for user authentication. In simplest terms, cryptography is to make communica-
tions secret, which here means that, even in the presence of an eavesdropper who
can monitor all communications, the intended message can still be delivered to the
receiver while kept secret to others [15], often by means of mathematical techniques
to achieve some desirable properties for secrecy.
As can be seen in Fig. 9.2, cryptography is often associated with the process of an
ordinary plaintext being converted to ciphertext. Ciphertext is text that only the
intended receiver can decode. The process that converts plaintext to ciphertext is
known as encryption, while the process of conversion from ciphertext to plaintext is
called decryption [15].

Figure 9.2: Cryptography jargon [15]

Blockchain and cryptocurrencies use cryptography in multiple ways: for wallets, trans-
actions, security, and privacy-preserving protocols, by means of different subsequent
algorithms.

9.1.3.1 Cryptographic Algorithms

Based on the number of keys that are employed for encryption and decryption,
cryptographic algorithms can be classified [42]:

José Manuel Rodrı́guez Lomeña 58

 9.1. CRYPTOCURRENCY

• Secret Key Cryptography (SKC) - Symmetric Encryption: Uses one

key for encryption and another for decryption.
• Public Key Cryptography (PKC) - Asymmetric Encryption: Uses a
single key for both encryption and decryption.
• Hash Functions: Uses a mathematical transformation to irreversibly ”encrypt”
information.
At that point, it is remarkable to note that each of these algorithms have specific
applications that are irreplaceable to each other, and for that reason they should be
differentiated.

Symmetric Encryption
Symmetric encryption uses a single key for both encryption and decryption, as shown
in Fig. 9.3. The message sender uses the key to encrypt the plaintext and sends the
ciphertext to the receiver. The receiver applies the same key to decrypt the message
and recover the plaintext.

Figure 9.3: Symmetric encryption [15]

Symmetric key systems are faster and simpler (when compared to asymmetric key
systems). However, they do not solve the key exchange problem between sender and
receiver if the key is not known in advance [15].
With this form of cryptography, it is therefore obvious that the key must be known
to both the sender and the receiver; that, in fact, is the secret. The biggest difficulty
with this approach, of course, is the distribution of the key.
Secret key cryptography schemes are generally categorized as being either stream
ciphers or block ciphers. Stream ciphers operate on a single bit (byte or computer
word) at a time and implement some form of feedback mechanism so that the key is
constantly changing. A block cipher is so-called because the scheme encrypts one
block of data at a time using the same key on each block.

59 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

In general, the same plaintext block will always encrypt to the same ciphertext when
using the same key in a block cipher whereas the same plaintext will encrypt to
different ciphertext in a stream cipher.
With that, it is clearly that symmetric encryption is ideally suited to encrypting
messages, since the sender can generate a session key on a per-message basis to
encrypt the message; and the receiver, of course, needs the same session key to
decrypt the message [42].

Asymmetric Encryption
Asymmetric encryption uses two different keys for encryption and decryption. The
key that will need to be kept secret is called the private key, while the key that does
not is called the public key.
For instance, if A wants to send a message to B and ensure that B will be the only
person able to comprehend the message, A can encrypt the message using the public
key such that only B can decrypt the message using the private key [15].

Figure 9.4: Asymmetric encryption [15]

While the public key and private key are different, they are mathematically related.
But the mathematical relationship is only usable upon encryption and decryption,
and others can never derive the private key even if they know the public key
[15]. Specifically, it depends upon the existence of so-called one-way functions, or
mathematical functions that are easy to computer whereas their inverse function is
relatively difficult to compute [42].
Asymmetric encryption employs two keys that are mathematically related although
knowledge of one key does not allow someone to easily determine the other key. One
key is used to encrypt the plaintext (public key) and the other key is used to decrypt
the ciphertext (private key), as can be seen in Fig. 9.4. The important point here is
that it does not matter which key is applied first, but that both keys are required for
the process to work [42].

José Manuel Rodrı́guez Lomeña 60

 9.1. CRYPTOCURRENCY

Asymmetric cryptography algorithms are widely used in cryptocurrencies. For

example, the wallet address is a public key, and only those who have the private key
are able to use the money inside [15].

Hash Functions
Hash functions, also called message digests and one-way encryption, are algorithms
that, in some sense, use no key (Fig. 9.5). Instead, a fixed-length hash value is
computed based upon the plaintext that makes it impossible for either the contents
or length of the plaintext to be recovered [42].

Figure 9.5: Hash functions [15]

Hash algorithms are typically used to provide a digital fingerprint of the contents of
a file, often used to ensure that the file has not been altered by an intruder or virus.
Hash functions are also commonly employed by many operating systems to encrypt
passwords. Hash functions, then, provide a measure of the integrity of a file [42].
Bearing the above in mind, it is essential to take into account that a hash function
should fulfill three security properties [15]:
• Collision Resistant: Nobody can find two different input strings so that if you
apply the hash function to them, they have the same output. Mathematically,
it can be expressed as: x and y do not exist, where H(x) = H(y) and x are not
equal to y.
• Hidden: Given H(x), it is infeasible to find x. This can be explained as if you
are given the hashed version of x, you will not find x.
• Puzzle-Friendly: If someone wants to target a certain hash function, get
some value of y. If part of the input is chosen randomly, it is difficult to find
another value to target the hash function value. Given an output y of the hash
function, if k is chosen from a random distribution, there is no way to find an
x, such that the k|x hash to y: H(k|x)=y.

61 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

Hash functions are well-suited for ensuring data integrity because any change made
to the contents of a message will result in the receiver calculating a different hash
value than the one placed in the transmission by the sender. Since it is highly unlikely
that two different messages will yield the same hash value, data integrity is ensured
to a high degree of confidence [42].

9.1.4 Cryptocurrency Wallet

Once cryptography has been understood, it is time to comprehend how one can make
transactions in a specific blockchain: by means of a cryptocurrency wallet.

Cryptocurrency wallets (crypto-wallets) are key to use blockchain. Every user, who
intend to use blockchain platform for any transaction, has to use crypto-wallet.
Unlike traditional pocket wallets, cryptocurrencies are not stored in crypto-wallets.
In fact, cryptocurrencies are neither stored in any single area nor exist anywhere in
any bodily form, but it exists as data of transactions stored on blockchain [52].

Therefore, wallets are software applications which can be used to view cryptocurrency
balances and make transactions. Each wallet type is a piece of different software,
which provides access to blockchain data and operate on blockchain data. In general,
any given wallet can work with one or more cryptocurrencies and features.

Public addresses are like cryptocurrency precise account numbers, they are used
to acquire a particular kind of cryptocurrency and can be shared publicly. For
example, to access Bitcoin blockchain one needs Bitcoin address, similarly to access
Ethereum blockchain one needs Ethereum address. Thus, a wallet helps to access all
transactions associated with that address on respective blockchain [52].

As commented, wallets store one or more cryptocurrency-unique public addresses. A

public address is a hexadecimal string with a combination of numbers and letters,
using both upper case and lower case. It needs to be shared publicly to receive
cryptocurrency. The software within the cryptocurrency wallet is connected directly
to the blockchain. The connected wallet allows user to read transactions and to
submit transactions to the blockchain. More specifically, wallet is client software to
blockchain platform to facilitate operations on the blockchain.

A secret key associated with a public address known as private key is also stored in
the wallet. A private address is also a hexadecimal string. It is difficult to remember
by the human beings, therefore it is store in wallet software or with third party
vendors in case of online wallets. A user need to know a pin associated with each
private key to access the wallet. A pin is easy to remember small string similar to
ATM pin. In general, a wallet is like an online bank account platform, the public
address is similar to an account number, the blockchain is like the bank’s ledger, and
with custodial wallets, the custodian is a bit like a banker.

In a nutshell, a crypto-wallet interacts with the blockchain with stored public address;
and transactions on blockchain are signed with the private key associated with the
public key [52].

José Manuel Rodrı́guez Lomeña 62

 9.1. CRYPTOCURRENCY

9.1.4.1 Types of Cryptocurrency Wallets

According to user needs, cryptocurrencies are handled using different type of crypto-
wallets [52]:
• Desktop wallet: It is a software that can be downloaded and installed on
a PC or a laptop. It is easy to use from a single computer in which it is
installed. Desktop wallets offer one of the maximum tiers of security. However,
if computer is hacked or receives a virus, there is possibility of loosing keys and
thereby funds.
• Online wallet: It is a software that runs on the cloud and is theoretically
easy to use from any location. Although it is handy to access, such wallets
save your private and public keys on-line and are controlled by a third party.
It subjects to on-line risk of hacking or compromising third party.
• Mobile wallet: It is a software that runs as a mobile application on your
phone. It is more convenient and user friendly to use for transactions and
payments like retail stores. Mobile wallet software has a simple user interface
than desktop wallets.
• Hardware wallet: These wallets differ from software wallets in terms of key
storage and management as keys are saved on a device like the one shown
in Fig. 9.6. Although hardware wallets make transactions online, public and
private keys are saved offline. Hence, it delivers immoderate security, since it
is absolutely a physical copy or printout of public and private keys of the user.
It also refers to a bit of software that is used to securely generate multiple keys
which is then subsequently printed on paper. Use of a paper wallet is relatively
trustworthy and it provides the highest stage of security as compared to above
wallets, despite compromising or loss of paper wallet is still risky.

Figure 9.6: KeepKey Hardware Wallet [2]

Digital wallets are not registered under the law of any country, so it is important to
cautiously choose the cryptocurrency wallet for digitized transactions. It is always
better to choose a universal cryptocurrency wallet that supports more than one
cryptocurrency.

63 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

Thus, when choosing the best cryptocurrency wallets to operate with a bot, it is
highly recommended to make use of two types of wallets: a web-based wallet and a
hardware wallet. On the web-based wallet, the immediate cash that is used to operate
is stored. Whereas on the hardware wallet, it is stored the bulk of cryptocurrency
volume. Hence, via the web-based wallet, it is possible to trade to and from other
exchange based wallets, and then transferring trading profits from the web-wallet
onto the hardware wallet for long term safekeeping [52].

9.1.5 Proof-of-Work against Proof-of-Stake

Proof-of-Work (PoW) and Proof-of-Stake (PoS) are both cryptocurrency consensus
mechanisms1 used to confirm transactions and create new blocks in a blockchain2 , so
that it is secured. Their main difference is how they choose and qualify users to add
transactions.

9.1.5.1 Proof-of-Work
PoW is the original crypto consensus mechanism. Proof-of-Work and mining are
closely related ideas. The reason why it is called “Proof-of-Work” is because the
network requires a huge amount of processing power [11]. The work, in this case,
consists of generating a hash (a long string of characters) that matches the target
hash for the current block. The crypto miner who does this wins the right to add
that block to the blockchain and receive rewards [55], thus making the blockchain
secure and causing it to be verified.
To accomplish this, miners use mining devices that quickly generate computations.
The aim is to be the first miner with the target hash because that miner is the one
who can update the blockchain and receive crypto rewards. Hence, the reason why
PoW in cryptocurrency works satisfactorily is because finding the target hash is
challenging enough to prevent the manipulation of transaction records, but at the
same time, verifying it is not, since once a target hash is found, it is straightforward
for other miners to check it [55].
Proof-of-Work has some powerful advantages. Firstly, it is a proven, robust way
of maintaining a secure decentralized blockchain. Additionally, as the value of a
cryptocurrency grows, more miners are incentivized to join the network, increasing
its power and security.
By contrast, it is an energy-intensive process that can have trouble scaling to
accommodate the vast number of transactions that huge blockchains can generate.
Also, because of the amount of processing power involved, it becomes impractical for
any individual or group to meddle with a valuable cryptocurrency’s blockchain [11].
For that reason, alternatives have been developed, the most popular of which is the
aforementioned Proof-of-Stake.
1
A consensus mechanism is a method for validating entries into a distributed database and
keeping the database secure [41]
2
Please read Sec. 9.1.2.1 Transaction Process for more detailed information about the transaction
process of cryptocurrency in blockchain.

José Manuel Rodrı́guez Lomeña 64

 9.1. CRYPTOCURRENCY

9.1.5.2 Proof-of-Stake
Proof-of-Stake is also a cryptocurrency consensus mechanism for processing transac-
tions and creating new blocks in a blockchain. It was created as an alternative to
PoW, since it reduces the amount of computational work needed to verify blocks and
transactions. Under PoW, blockchain is kept blockchain, while PoS changes the way
blocks are verified using the machines of coin owners, so there does not need to be as
much computational work done. Instead, the owners offer their coins as collateral
(staking) for the chance to validate blocks and then become validators [41].
Validators are selected randomly to confirm transactions and validate block informa-
tion. This system randomizes who gets to collect fees rather than using a competitive
rewards-based mechanism like proof-of-work. Thus, to become a validator, a coin
owner must ”stake” a specific amount of coins [41].
Staking is the process of locking up crypto holdings in order to obtain rewards or
earn interest. Staking is another way to describe validating those transactions on a
blockchain. But achieving that consensus requires participants. That’s what staking
is—investors who actively hold onto, or lock up their crypto holdings in their crypto
wallet are participating in these networks’ consensus-taking processes. Stakers are,
in essence, approving and verifying transactions on the blockchain [50].
Hence, each blockchain has their specific minimum number value of its cryptocurrency
to hold on staking, so as to be able to become a validator. Furthermore, it is relevant
to note that blocks are validated by more than one validator, and when a specific
number of the validators verify that the block is accurate, it is finalized and closed.
With all of the above, it is evidently that PoS is seen as less risky regarding the
potential for an attack on the network, as it structures compensation in a way that
makes an attack less advantageous. Furthermore, cryptocurrency owners validate
block transactions based on the number of staked coins, being the next block writer
on the blockchain selected at random, with higher odds being assigned to nodes with
larger stake positions [41].

9.1.5.3 PoW & PoS Comparison

Both consensus mechanisms help blockchains synchronize data, validate information,
and process transactions. Each method has proven to be successful at maintaining a
blockchain, although each has pros and cons. However, the two algorithms have very
differing approaches.
Under PoS, block creators are called validators. A validator checks transactions,
verifies activity, votes on outcomes, and maintains records. Under PoW, block
creators are called miners. Miners work to solve for the hash, a cryptographic
number, to verify transactions. In return for solving the hash, they are rewarded
with a coin.
To ”buy into” the position of becoming a block creator, you need only own enough
coins or tokens to become a validator on a PoS blockchain. For PoW, miners must

65 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

invest in processing equipment (Fig. 9.7) and incur hefty energy charges to power
the machines attempting to solve the computations.

Figure 9.7: Cryptocurrency miners [32]

The equipment and energy costs under PoW mechanisms are expensive, limiting
access to mining and strengthening the security of the blockchain. PoS blockchains
reduce the amount of processing power needed to validate block information and
transactions. The mechanism also lowers network congestion and removes the
rewards-based incentive PoW blockchains have [11].
Taking all of the above into account, Table (9.1) compares the two main consensus
mechanisms used in blockchains.

Table 9.1: PoW & PoS comparison [11]

Proof-of-Work Proof-of-Stake
Block creators are called miners Block creators are called validators
Participants must buy equipment Participants must own coins
and energy to become a miner or tokens to become a validator
Not energy efficient Energy efficient
Robust security due to expensive
Security through community control
upfront requirement
Validators receive transactions
Miners receive block rewards
fees as rewards

9.1.6 History
Although relatively short, cryptocurrencies have a history of a few decades, which is
worth knowing in order to contextualize them. Hence, the history presented below is
a summary taken from [16].

José Manuel Rodrı́guez Lomeña 66

 9.1. CRYPTOCURRENCY

9.1.6.1 The Idea for Cryptocurrency

The idea for cryptocurrency first emerged in 1983, when American cryptographer
David Chaum published a conference paper outlining an early form of anonymous
cryptographic electronic money. The concept was for a currency that could be sent
untraceably and in a manner that did not require centralized entities (i.e. banks). In
1995, Chaum built on his early ideas and developed a proto-cryptocurrency called
Digicash. It required user software to withdraw funds from a bank and required
specific encrypted keys before said funds could be sent to a recipient.
Bit Gold, often deemed a direct precursor to Bitcoin, was designed in 1998 by Nick
Szabo. It required a participant to dedicate computer power to solving cryptographic
puzzles, and those who solved the puzzle received a reward. Combined with Chaum’s
work, it results in something that comes very close to resembling Bitcoin.
But Szabo could not solve the infamous double-spending problem (digital data
can be copied and pasted) without the use of a central authority. As such, it was
not until a decade later when a mysterious person or group, using the pseudonym
Satoshi Nakamoto, set the history of Bitcoin and later cryptocurrencies in motion, by
publishing a white paper called “Bitcoin – A Peer to Peer Electronic Cash System”
[47].

9.1.6.2 The Beginning (2008-2010)

On 31st October 2008, Satoshi Nakamoto published the Bitcoin (BTC) white paper,
describing the functionality of the Bitcoin blockchain network. Satoshi formally began
work on the bitcoin project on 18th August 2008, when they purchased Bitcoin.org.
The history of Bitcoin was underway. Satoshi Nakamoto mined the first block of
the Bitcoin network on 3rd January 2009. They embedded a headline from The
Times newspaper in this initial block, making a permanent reference to the economic
conditions - involving bank bailouts and a centralized financial system - that Bitcoin
was partly a reaction against.
This first block - which resulted in 50 Bitcoins being mined - is now referred to as
the Genesis Block. Bitcoin had virtually no value at this time, as well as for the first
few months of its existence. Six months after Bitcoin became tradeable, in April
2010, the value of one BTC was just under 14 cents. By early November, the price
surged to 36 cents before settling at around 29 cents.

9.1.6.3 The Market Begins to Form (2010-2014)

While it was not yet worth much, Bitcoin was showing it had real world value. In
February 2011, it rose to $1.06 before coming back down to roughly 87 cents. In the
spring, in part due to a Forbes story on the new “Cryptocurrency,” the price took off.
From early April to the end of May, the value of BTC rose from 86 cents to $8.89.
On 1st June, after Gawker published a story about the appeal of currency in the
online drug dealing community, the price more than tripled in a week, to about $27.

67 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

The market value of BTCs in circulation approached nearly $130 million. However,
by September 2011, the value had dropped back down to around $4.77.
In October of that same year, Litecoin (LTC) appeared, one of many forks (i.e. up-
dated versions) of Bitcoin. Litecoin soon became the second-biggest cryptocurrency
by market cap, with the earliest archive of CoinMarketCap [14] (from May 2013)
showing PPCoin, Namecoin and 10 others trailing in the distance. Such cryptocur-
rencies were quickly named as altcoins, with some forked from Bitcoin and others
based on new code. Further information on the types of cryptocurrencies is detailed
in Sec. 9.1.7 Types of cryptocurrency.
In 2012, Bitcoin prices grew steadily, and in September of that year the Bitcoin
Foundation was established to promote the development and uptake of Bitcoin. Then
known as OpenCoin, Ripple (XRP) was also launched that year, with the project
attracting venture capital the next year.
In 2013, amid federal, criminal, regulatory, and software related issues, the price of
BTC constantly rose and crashed. On 19th November, its price reached $755, only to
crash down to $378 the same day. By 30th November, it was all the way up to $1,163
again. This was, however, the beginning of another long-term crash that ended with
Bitcoin dropping back down to $152 by January 2015.

9.1.6.4 Scams Dominate Headlines (2014-2016)

Though intentional, anonymity and lack of centralized control make digital currency
highly attractive to criminals. In January 2014, Mt.Gox - then the largest BTC
exchange in the world - collapsed and declared bankruptcy, having lost 850,000 BTC.
While it is not known what exactly happened, it is likely that the missing BTCs
were stolen gradually over time, beginning in 2011, and resold on various exchanges
for cash (including Mt.Gox), until one day Mt.Gox checked their wallets and found
they were empty. Its CEO Mark Karpeles was charged with embezzlement in 2017,
but acquitted in 2019, so the destination of the missing BTC remains a mystery.
While the hack was not a singular event, it has served as a cautionary tale, and
security on exchanges is much improved. Though smaller exchanges continue to be
hacked even today, larger platforms now provide more guarantees on their reserve
holdings in case of breaches. This includes the Secure Asset Fund for Users on
Binance, for example, which is an emergency insurance fund.
Crypto traders are advised to use a hardware or software wallet3 to safely store their
cryptocurrency rather than storing them on an exchange. Wallets such as these were
not as accessible during this early period in the history of cryptocurrency.

9.1.6.5 Bitcoin Ascends to Worldwide Phenomenon (2016-2018)

Bitcoin prices rose steadily year over year, going from $434 in January 2016, to $998
in January 2017. In July 2017, a software upgrade to Bitcoin was approved, with
3
Readers interested in further information on hardware wallets are referred to Sec. 9.1.4.1 Types
of Cryptocurrency Wallets.

José Manuel Rodrı́guez Lomeña 68

 9.1. CRYPTOCURRENCY

the aim being to support development of the Lightning Network (a layer-two scaling
solution) as well as improve security.
A week after the upgrade was activated in August, Bitcoin was trading at around
$2,700. By 17th December 2017, Bitcoin reached an astronomical all-time high (ATH)
of just under $20,000.
During this same time, a new blockchain project called Ethereum (ETH) was making
noise in the cryptocurrency sphere, having quickly become the number two cryp-
tocurrency by market cap since launching in July 2015. It brought smart contracts4
to cryptocurrency, opening a wide array of potential use cases and generating over
200,000 different projects (and counting). In contrast to Bitcoin, Ethereum enables
additional platforms to launch and operate on its own chain, each with their own
cryptocurrencies and their own use cases. This was a model widely copied by other
new blockchains, with Cardano (ADA) or Tezos (XTZ) launched during this period.

9.1.6.6 Bust and Recovery, and Bust and Recovery (2018-Present)

Bitcoin was not able to sustain its all-time high of $19,783. Likewise, Ethereum,
which reached its own ATH in January 2018 of around $1,400, was also not able to
maintain its newfound level for long. Financial regulations and security concerns
(due to semi-regular exchange hacks) contributed to the market-wide decline, and by
the end of 2018, BTC had dropped down to around $3,700.
However, prices did not stay down for too long. Starting from late 2020, BTC enjoyed
something of a renaissance, beginning with the announcement of Microstrategy in
August that it had bought BTC worth $250 million. This kicked off a bull market
that was joined by the rest of the market, with prices boosted further by the purchase
of Tesla of $1.5 billion in BTC in early 2021. It was in November of that year that
BTC reached its current record high of $69,000, as it is displayed in Fig. 9.8.

Figure 9.8: Evolution of BTC-USD price [28]

4
A smart contract, like any contract, establishes the terms of an agreement. But unlike a
traditional contract, terms of a smart contract are executed as code running on a blockchain, and
allow developers to build apps that take advantage of blockchain security, reliability, and accessibility
while offering sophisticated peer-to-peer functionality - everything from loans and insurance to
logistics and gaming [12].

69 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

The market has fallen once again since this high, dragged down by macroeconomic
concerns resulting from high inflation, rising interest rates and the specter of war.
That said, with global stock markets also falling in late 2021 and 2022, parallel fall
of cryptocurrencies shows that the sector is becoming increasingly entwined with
traditional financial markets.
And while the volatility of cryptocurrencies is both attractive and potentially dev-
astating, the underlying technology behind them all, blockchain, has the power to
change many sectors of our society. Whether it is providing accessible and affordable
financial exchange options, securing your own funds so that no one but you can access
them, or providing accurate data for your insurance quote, blockchain technology
has the potential to be used in almost every area of the economy.
As the market becomes more stable with increased knowledge, and with the in-
troduction of new areas such as stablecoins and decentralized finance (DeFi), it is
simple to get excited about cryptocurrency and its potential from an investment and
technological perspective.

9.1.7 Types of cryptocurrency

The term cryptocurrency is often used interchangeably with token and coin, regardless
of not being the same. Thus, cryptocurrencies generally fall into one of these two
categories [51]:
• Coins: Programmable assets that live within the blockchain of a their own
platform.
• Tokens: Programmable assets that live within the blockchain of a given
platform.

9.1.7.1 Coins
Crypto coins are strings of computer code that can represent an asset, concept,
or project - whether tangible, virtual, or digital - intended for various uses and
with varying valuations. Originally, these coins were meant to function as a type of
currency [51].
Coins are built on their own blockchain and were originally intended as a form
of currency. Generally, any blockchain-based cryptocurrency that is not Bitcoin
is referred to as an altcoin. Thus, an altcoin is an alternative digital currency to
Bitcoin.
A digital coin is created on its own blockchain and acts much like fiat (traditional
money). Coins can be used to store value and as a means of exchange between two
parties doing business with each other [51]. Examples of coins include Bitcoin and
Litecoin, which is an altcoin, among some of those shown in Fig. 9.9.
In addition to altcoins, it is worth mentioning stablecoins. Stablecoins are coins
whose value is pegged, or tied, to that of another currency, commodity, or financial
instrument. They aim to provide an alternative to the high volatility of the most

José Manuel Rodrı́guez Lomeña 70

 9.1. CRYPTOCURRENCY

popular coins, which has made crypto investments less suitable for common transac-
tions [39]. Most common stablecoins are Tether (USDT) and USD Coin (USDC),
which both replicate to the real USD.

Figure 9.9: Most common coin symbols [17]

9.1.7.2 Tokens
Tokens are the digital representations of a particular asset or utility in a blockchain.
All tokens can be termed altcoins, but they are differentiated by residing on top of
another blockchain and not being native to the blockchain on which they reside.

Tokens are usually created and distributed through an initial coin offering (ICO),
much like an initial public offering (IPO) for stock. They can be represented as value
tokens, security tokens or utility tokens.

Therefore, tokens are created on an existing blockchain (not their own), and can
function in many more ways than acting as currency. Instead of representing an
exchange of value, tokens are considered programmable assets on which you may
create and execute unique smart contracts. These contracts can establish ownership
of assets outside the blockchain network.

Tokens can represent units of value - including real-world items like electricity, money,
points, coins, digital assets, and more - and can be sent and received. They can be
used as part of a software application - such as granting access to an app, verifying
identity, or tracking products moving through a supply chain. They can also represent
digital art - as with non-fungible tokens5 (NFTs) [51].

5
Non-fungible tokens (NFTs) are cryptographic assets on a blockchain with unique identification
codes and metadata that distinguish them from each other. Unlike cryptocurrencies, they cannot be
traded or exchanged at equivalency. This differs from fungible tokens like cryptocurrencies, which
are identical to each other and, therefore, can serve as a medium for commercial transactions [38].

71 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

9.1.7.3 Largest Cryptocurrencies by Market Cap

Once understood the existing types of cryptocurrency, Table (9.2) is presented so as
to classify the 15 first cryptocurrencies by more market cap.

Table 9.2: Cryptocurrency with more market cap on 09/12/2022 at 11 am [14]

# Name Price Market Cap Circulating Supply

1 Bitcoin (BTC) $17,208.72 $331.11B 1,213,714 BTC
2 Ethereum (ETH) $1,281.16 $156.78B 5,002,449 ETH
3 Tether (USDT) $1.00 $65.71B 26,103,185,150 USDT
4 Binance Coin (BNB) $290.49 $46.47B 2,133,753 BNB
5 USD Coin (USDC) $1.00 $42.79B 2,210,474,750 USDC
6 Binance USD (BUSD) $1.00 $22.06B 5,887,924,841 BUSD
7 XRP (XRP) $0.3914 $19.67B 1,930,489,911 XRP
8 Dogecoin (DOGE) $0.09826 $13.04B 4,465,554,116 DOGE
9 Cardano (ADA) $0.3146 $10.85B 459,814,823 ADA
10 Polygon (MATIC) $0.9238 $8.07B 285,644,523 MATIC
11 Polkadot (DOT) $5.40 $6.19B 27,277,787 DOT
12 Dai (DAI) $1.00 $5.87B 157,076,833 DAI
13 Litecoin (LTC) $78.12 $5.61B 6,960,013 LTC
14 Shiba Inu (SHIB) $0.000009312 $5.11B 9,014,342,996,273 SHIB
15 Solana (SOL) $13.87 $5.08B 13,756,547 SOL

At first glance, it is prominent that both Bitcoin and Ethereum encompass the largest
amount of capital within the cryptocurrency market, with a 38.5% and 18.3% of the
total market cap, respectively. This reflects the dominance of both cryptos, which
are undoubtedly the most established ones.
Indeed, such is the importance of Bitcoin that traders used BTC dominance to figure
out whether altcoins are trending bullish or bearish against Bitcoin. For example,
one popular theory is that if altcoins are going up, the crypto market is headed for
a bull market. In 2017, for example, a major drop in BTC dominance indicated a
sharp rise in altcoin prices (not a fall in BTC price), which coincided with the entire
market entering a bullish phase.

Bitcoin
As commented, Bitcoin is the clear leader in the crypto sector. It is also the very
first cryptocurrency. Bitcoin launched in 2009; created by a person (or possibly a
group) that goes by the pseudonym Satoshi Nakamoto. As of June 2022, there are
slightly more than 19 million Bitcoin tokens in circulation, against a capped limit of
21 million. Almost a thousand new bitcoins are mined each day, bringing Bitcoin
ever closer to its maximum finite number [51].

Ethereum
Like Bitcoin, Ethereum is a blockchain network. But Ethereum was designed as
a programmable blockchain - meaning it was not created to support a currency,

José Manuel Rodrı́guez Lomeña 72

 9.1. CRYPTOCURRENCY

but rather to enable the network’s users to create, publish, monetize, and deploy
decentralized applications (dApps). ETH was developed as a form of payment on
the Ethereum platform. It might be helpful to think of ETH as a kind of fuel that
powers the Ethereum blockchain. Ethereum has helped to launch many initial coin
offerings because many ICOs are built on the Ethereum blockchain. Ethereum has
also been the blockchain behind the boom in non-fungible tokens (NFTs).
As the two most widely known blockchains and cryptocurrencies, many people often
directly compare Ethereum and Bitcoin against each other. In reality, Bitcoin and
Ethereum are designed to achieve different goals, and in many ways can be regarded
as complementary forces. Bitcoin is a peer-to-peer digital cash network, which
facilitates transactions without the need for a central authority. At turn, Ethereum
is often referred to as the world computer, since it iterates on the technology of
Bitcoin while introducing smart contracts, which allow for building dApps that span
a broad range of crowdfunding platforms, financial instruments, digital games and
collectibles, and decentralized marketplaces [51].

Tether
Tether was the first cryptocurrency marketed as a stablecoin - a breed of crypto
known as fiat-collateralized stablecoins. The value of the Tether is pegged to a fiat
currency - in this case, the U.S. dollar. Tether is the largest stablecoin in the world;
in 2022, the majority of cryptocurrencies are traded using Tether.
Like other stablecoins, Tether is designed to offer stability, transparency, and lower
transaction fees to users. In fact, Tether was originally used to avoid the extreme
volatility of the crypto market [51].

Binance Coin
Binance is one of the biggest cryptocurrency exchanges in the world6 . The Binance
Coin (BNB) was created as a utility token for use as a medium of exchange on Binance.
It was initially built on the Ethereum blockchain, but now lives on Binance’s own
blockchain platform. Originally, BNB allowed traders to get discounts on trading
fees on Binance, but now it also can be used for payments, to book travel, for
entertainment, online services, and financial services.
As one of the top five cryptocurrencies by market cap in 2022, BNB has developed a
wide range of use cases and real-world applications. But, as with other digital assets,
this crypto platform has also faced regulatory hurdles here and abroad.
BNB was created with a maximum of 200 million tokens, about half of which were
made available to investors during its ICO. Every quarter, to drive demand, Binance
buys back and then burns - permanently destroys, or removes from circulation - some
of the coins it holds, aiming at reducing the overall supply and thus increasing the
value of the remaining tokens, as assets tend to rise in price whenever the circulating
supply falls, and they become more scarce [51].
6
Please read Sec. 9.1.8 Cryptocurrency Exchange for more information about Binance.

73 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

9.1.8 Cryptocurrency Exchange

Cryptocurrency exchanges are privately-owned platforms that facilitate the trading
of cryptocurrencies for other crypto assets, including digital and fiat currencies and
NFTs [9].
Basically, there are two kind of exchanges: Centralized Cryptocurrency Exchanges
(CEX) and Decentralized Cryptocurrency Exchanges (DEX), each of which offers
certain advantages and disadvantages.

9.1.8.1 Centralized Cryptocurrency Exchange

Centralized cryptocurrency exchanges act as an intermediary between a buyer and a
seller and make money through commissions and transaction fees. A CEX can be
imagined to be similar to a stock exchange but for digital assets [9].
Most popular crypto exchanges are Binance, Coinbase, Kraken and KuCoin. Much
like stock trading websites or apps, these exchanges allow cryptocurrency investors
to buy and sell digital assets at the prevailing price, called spot, or to leave orders
that get executed when the asset gets to the desired price target, called limit orders.
CEXs operate using an order book system, which means that buy and sell orders
are listed and sorted by the intended buy or sell price. The matching engine of the
exchange then matches buyers and sellers based on the best executable price given
the desired lot size. Hence, the price of a digital asset will depend on the supply and
demand of that asset versus another, whether it be fiat currency or cryptocurrency.
Hence, CEXs decide which digital asset is allowed trading in, which provides a small
measure of comfort that unscrupulous digital assets may be excluded from the CEX.

Binance
According to traffic, liquidity7 and trading volumes, Binance is the best CEX as
of December 2022 [14]. It was founded in 2017 in Hong Kong, and features a
strong focus on altcoin trading, offering crypto-to-crypto trading in more than 600
cryptocurrencies and virtual tokens.
The Binance exchange has among the lowest transaction fees for cryptocurrency
exchanges. It also has high liquidity and offers discounts to users who pay in the native
BNB cryptocurrency tokens. Moreover, with high standards of safety and security
and multi-tier and multi-clustered architecture, Binance delivers high processing
throughput with the capacity to process around 1.4 million orders per second.
To start trading, users have to complete the necessary KYC requirements. Upon
successful trading account creation, users can add cryptocurrency funds to their
public wallet address, provided by Binance, to start trading [35].
7
Liquidity is the efficiency or ease with which an asset or security can be converted into ready
cash without affecting its market price [40].

José Manuel Rodrı́guez Lomeña 74

 9.1. CRYPTOCURRENCY

9.1.8.2 Decentralized Cryptocurrency Exchange

A decentralized exchange is another type of exchange that allows peer-to-peer
transactions directly from your digital wallet without going through an intermediary.
Examples of DEXs include Uniswap, PancakeSwap, dYdX, and Kyber.
These decentralized exchanges rely on smart contracts, self-executing pieces of code
on a blockchain. These smart contracts allow for more privacy and less slippage
(another term for transaction costs) than a centralized cryptocurrency exchange.
On the other hand, even though smart contracts are rules-based, the lack of an
intermediary third party means that the user is left to their own, so DEXs are meant
for sophisticated investors [9].

9.1.8.3 CEX against DEX

Once understood the difference between CEXs and DEXs, it is pertinent to analyze
the advantages and disadvantages of each type of crypto exchange [9].

Advantages of CEX
• User-friendly: Centralized exchanges offer beginner investors a familiar,
friendly way of trading and investing in cryptocurrencies. As opposed to using
crypto wallets and peer-to-peer transactions, which can be complex, users of
centralized exchanges can log into their accounts, view their account balances,
and make transactions through applications and websites.
• Reliable: Centralized exchanges offer an extra layer of security and reliability
when it comes to transactions and trading. By facilitating the transaction
through a developed, centralized platform, centralized exchanges offer higher
levels of comfort.
• Leverage: One of the other benefits of certain CEXs is the option to leverage
your investments using borrowed money from the exchange, called margin
trading. It allows investors to reap higher returns, but losses can also be
amplified.

Disadvantages of CEX
• Hacking risk: Centralized exchanges are operated by companies that are
responsible for the holdings of their customers. Large exchanges usually hold
billions of dollars worth of Bitcoin, making them a target for hackers and theft.
• Transaction fees: Unlike peer-to-peer transactions, centralized exchanges
often charge high transaction fees for their services and convenience, which can
be especially high when trading in large amounts.
• Custody of digital assets and risk of fraud: Most CEXs will hold digital
assets as a custodian in their own digital wallet rather than allow users to
store their private keys on their own digital wallet. While more convenient

75 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

when wanting to trade, there are drawbacks, namely the risk of the centralized
cryptocurrency exchange failing and fraud.

Advantages of DEX

• Custody: Users of decentralized exchanges do not need to transfer their assets

to a third party. Therefore, there is no risk of a company or organization being
hacked, and users are assured of greater safety from hacking, failure, fraud, or
theft.

• Preventing market manipulation: Due to their nature of allowing for the

peer-to-peer exchange of cryptocurrencies, decentralized exchanges prevent
market manipulation, protecting users from fake trading and wash trading.

• Less censorship: Decentralized exchanges do not require customers to fill out

know-your-customer (KYC) forms, offering privacy and anonymity to users.
Since DEXs do not exercise censorship, more cryptocurrencies and digital assets
are available than through a CEX. As a matter of fact, many altcoins are only
available on DEXs.

Disadvantages of DEX

• Complexity: Users of decentralized exchanges must remember the keys and

passwords to their crypto wallets, or their assets are lost forever and cannot be
recovered. They require the user to learn and get familiar with the platform
and the process, unlike centralized exchanges, which offer a more convenient
and user-friendly process.

• Lack of fiat payments: DEXs are best for investors looking to switch from
one digital asset to another and not well suited for someone looking to buy or
sell digital assets with fiat currency, called on and off-ramping. It makes them
less convenient for users that do not already hold cryptocurrencies.

• Liquidity struggles: Some 99% of crypto transactions are facilitated by

centralized exchanges, which suggests that they are accountable for the majority
of the trading volume. Due to the lack of volume, decentralized exchanges often
lack liquidity, and it can be difficult to find buyers and sellers when trading
volumes are low.

With all of the above, it is undoubtedly that both type of exchanges can perfectly
coexist, and in fact they are complementary to each other, since they offer the user
flexibility, as they adapt to the specific needs and knowledge of the user.

Thus, the exchange that is used will depend on the required application, and that is
why it is vitally crucial to actively search for possible exchanges and select the one
that best suits the way that the user wants to operate with cryptocurrencies.

José Manuel Rodrı́guez Lomeña 76

 9.2. CRYPTOCURRENCY BOT

9.2 Cryptocurrency Bot

A cryptocurrency bot is a piece of software that looks for opportunities in the market
and capitalizes on them to generate profit. In other words, they allow automatically
trading with cryptocurrency. Crypto bots are more nimble and agile than a person
could ever be alone, and also can watch various different crypto pairs at once and
spot different opportunities in an instant.
As commented, crypto bots are basically programs that are executed to yield specific
results in trades. This implies that a technical analysis strategy is required to follow,
in order to provide the bot with a set of rules to execute. At that point, it is
mandatory to make sure that the theoretical strategy work in reality before deploying
it in normal trading. This process is called backtesting, and is developed in depth in
Sec. 9.2.1 Backtesting.
In addition to the above, it is essential to note that crypto bots usually relies on
historical data in order to both operate and backtest. For that reason, it is crucial to
have a financial database to obtain the appropriate values for the cryptocurrencies
with which you intend to operate. With that in mind, Yahoo! Finance database is
presented in Sec. 9.2.2 Yahoo! Finance.

9.2.1 Backtesting
Backtesting is the general method for seeing how well a strategy or model would have
worked. Backtesting assesses the viability of a trading strategy by discovering how it
would play out using historical data. If backtesting works, traders and analysts may
have the confidence to employ it going forward [34].
Hence, it is clearly that backtesting allows a trader to simulate a trading strategy
using historical data to generate results and analyze risk and profitability before
risking any actual capital. A well-conducted backtest that yields positive results
assures traders that the strategy is fundamentally sound and is likely to yield profits
when implemented in reality. In contrast, a well-conducted backtest that yields
suboptimal results will prompt traders to alter or reject the strategy.
As long as a trading idea can be quantified, it can be backtested. Some traders
and investors may seek the expertise of a qualified programmer to develop the idea
into a testable form. Typically, this involves a programmer coding the idea into the
proprietary language hosted by the trading platform.
The ideal backtest chooses sample data from a relevant time period of a duration
that reflects a variety of market conditions. In this way, one can better judge whether
the results of the backtest represent a fluke or sound trading.
A backtest should consider all trading costs, however insignificant, as these can add
up over the course of the backtesting period and drastically affect the appearance
of a strategy’s profitability. Traders should ensure that their backtesting software
accounts for these costs.
Out-of-sample testing and forward performance testing provide further confirmation

77 José Manuel Rodrı́guez Lomeña

CHAPTER 9. INTRODUCTION TO CRYPTOCURRENCY

regarding a system’s effectiveness and can show a system’s true colors before real
cash is on the line. A strong correlation between backtesting, out-of-sample, and
forward performance testing results is vital for determining the viability of a trading
system [34].
In order the backtesting results to be analyzed, the use of reward and risk metrics
arises. As reward metrics, it is common to work with the annualized return, the
average take profit/stop-loss, which is the relationship between the average take
profits divided by the average stop-losses obtained in the total of trades, and the
profitability, which is the percentage of winning trades. On the other hand, risk
metrics basically consist of the annualized volatility, understood as the standard
deviation of daily returns of the model in a year, and the maximum drawdown, which
is the maximum negative percentage difference in total portfolio value in a day.
Lastly, two more ratios are of interest when analyzing the results of backtesting.
They combine view of the returns and risk in one metric. Specifically, they are the
Sharpe and Sortino ratios, which are calculated as:

Annualized Net Return

Sharpe Ratio = (9.1)
Annualized Volatility

Annualized Net Return

Sortino Ratio = (9.2)
Annualized Volatility of Negative Returns

As expected, it is convenient that these ratios achieve the higher possible values,
since a higher annualized return will mean earning more money, and less volatility
will mean operating with less risk. Hence, when they exceed a value of 2, the strategy
is considered to operate with a correct return/risk ratio. Additionally, as a rule
of thumb, it is better to have a Sharpe ratio greater than the Sortino ratio, since
it is important that the volatility of the winning trades be lower so as to ensure
consistency when trading.

9.2.2 Yahoo! Finance

Yahoo! Finance is a media property that is part of the Yahoo! network. It provides
financial news, data and commentary including stock quotes, press releases, financial
reports, and original content. It also offers some online tools for personal finance
management. In 2017, Yahoo! Finance added the feature to look at news surrounding
cryptocurrency. It lists over 9,000 unique coins including Bitcoin and Ethereum [59].

José Manuel Rodrı́guez Lomeña 78

Chapter 10

Machine Learning and Statistics

The emergent price prediction schemes for cryptocurrency include strategies that are
based on statistical and ML technologies.
Generally, the statistical models use mathematical equations to encode information
extracted from the data, being thus able to quickly provide adequate models to
estimate and predict the values of various economic variables.
On the other side, with the advancement of big data technology and artificial
intelligence, numerous research studies have applied ML models to classification and
prediction problems. Hence, many researchers have focused their efforts on applying
these new techniques on financial markets.
With that, the aim of this chapter is to present some basic theoretical foundations on
Machine Learning and statistics, which serve to better understand the implementation
of the bot.

10.1 Machine Learning

Before delving into the field of ML, it is essential to know what Artificial Intelligence
is. Artificial Intelligence (AI) is the simulation of human intelligence processes by
machines, especially computer systems, which perform complex tasks in a way that
is similar to how humans solve problems.
Hence, Machine Learning is a subfield of AI. ML is broadly defined as the capability
of a machine to imitate intelligent human behaviour. Indeed, it is the science that
makes computers learn from data. Instead of programming, step by step, each specific
solution for each need raised, as is done in the conventional programming approach,
the ML area is dedicated to the development of generic algorithms that can extract
patterns from different types of data [6].
In order to be able to address any specific task, there are different types of classification
and models. ML is divided into mainly four types: supervised ML, unsupervised ML,
semi-supervised ML, and reinforcement ML. Despite the relevance of these kind of
methods, this section is intended to focus on neural networks, that precisely one of

79
CHAPTER 10. MACHINE LEARNING AND STATISTICS

its methods is the one on which the strategy implemented for the backtesting of this
project is based.

10.1.1 Neural Networks

Neural networks, also known as Artificial Neural Networks (ANNs) or Simulated
Neural Networks (SNNs), are a subset of machine learning and are at the heart of
deep learning algorithms. Their name and structure are inspired by the human brain,
mimicking the way that biological neurons signal to one another.
ANNs are comprised of a node layers, containing an input layer, one or more hidden
layers, and an output layer. Each node, or artificial neuron, connects to another and
has an associated weight and threshold. If the output of any individual node is above
the specified threshold value, that node is activated, sending data to the next layer
of the network. Otherwise, no data is passed along to the next layer of the network.

Figure 10.1: Neural networks layers’ diagram [56]

Neural networks rely on training data to learn and improve their accuracy over
time. However, once these learning algorithms are fine-tuned for accuracy, they are
powerful tools in computer science and AI, allowing us to classify and cluster data at
a high velocity [30].
At a basic level, a neural network is comprised of four main components: inputs,
weights, a bias or threshold, and an output, which must contain a non-linear activation
function. Hence, the algebraic formula is:

m
wi xi + bias = w1 x1 + w2 x2 + w3 x3 + bias (10.1)
X

i=1

With that, neural networks are also divided into nine different types: Perceptron,
Feed Forward Neural Network, Multilayer Perceptron, Convolutional Neural Network,

José Manuel Rodrı́guez Lomeña 80

 10.1. MACHINE LEARNING

Radial Basis Functional Neural Network, Sequence to Sequence Models, Modular

Neural Network, Long Short-Term Memory Network, and Recurrent Neural Network.
Hence, only the last three are discussed below, since are the more relevant ones for
trading bot applications.

10.1.2 Recurrent Neural Network

A Recurrent Neural Network (RNN) is a type of ANN which uses sequential data or
time series data to learn. They are distinguished by their “memory” as they take
information from prior inputs to influence the current input and output. While
traditional deep neural networks assume that inputs and outputs are independent of
each other, the output of RNNs depend on the prior elements within the sequence,
as can be seen in Fig. 10.1. While future events would also be helpful in determining
the output of a given sequence, unidirectional RNNs cannot account for these events
in their predictions.
Another distinguishing characteristic of recurrent networks is that they share pa-
rameters across each layer of the network. Specifically, they share the same weight
parameter within each layer of the network. That said, these weights are still adjusted
in the through the processes of backpropagation and gradient descent to facilitate
Reinforcement Learning [31].

10.1.3 Long Short-Term Memory Network

Long Short-Term Memory (LSTM) networks are a type of RNN that uses special
units in addition to standard units. LSTM units include a ‘memory cell’ that can
maintain information in memory for long periods of time. A set of gates is used
to control when information enters the memory when it is output, and when it is
forgotten, as can be observed in Fig. 10.2.

Figure 10.2: LSTM Network diagram [29]

There are three types of gates: input gate, output gate and forget gate. The input

81 José Manuel Rodrı́guez Lomeña

CHAPTER 10. MACHINE LEARNING AND STATISTICS

gate decides how many information from the last sample will be kept in memory; the
output gate regulates the amount of data passed to the next layer; and the forget
gate controls the tearing rate of memory stored. This architecture allows them to
learn longer-term dependencies.

10.1.4 Modular Neural Network

A modular neural network has a number of different networks that function indepen-
dently and perform sub-tasks. The different networks do not really interact with or
signal each other during the computation process. They work independently towards
achieving the output, as can be seen in Fig. 10.3.

Figure 10.3: Modular Neural Network diagram [29]

As a result, a large and complex computational process are done significantly faster
by breaking it down into independent components. The computation speed increases
because the networks are not interacting with or even connected to each other [29].

10.2 Statistical Concepts

This section is intended to clarify some important statistical concepts for this project.
It is assumed that the reader has an advanced level of statistics, and therefore basic
concepts of statistics are not explained, but instead focuses on those that may be
somewhat more specific.

Firstly, a statistical model used for time-series analysis is discussed. Later, a brief
explanation of what is a combination without repetition is presented.

José Manuel Rodrı́guez Lomeña 82

 10.2. STATISTICAL CONCEPTS

10.2.1 ARIMA model

Before explaining the model itself, it is mandatory to make it clear that a linear
statistical-model-based approach evaluates the linear relationship between prices and
an explanatory variable. Hence, if multiple explanatory variables exist, it is possible
to model the linear relationship between explanatory variables (independent) and
response variables (dependent) with the help of a multiple linear model [43].
The commonly used linear statistical model for time-series analysis is the AutoRe-
gressive Integrated Moving-Average (ARIMA) model, which uses time series data to
either better understand the data set or to predict future trends.
An ARIMA model is a form of regression analysis that gauges the strength of one
dependent variable relative to other changing variables. In order it to be understood,
each of its components are outlined [33]:
• AutoRegression (AR): refers o a model that shows a changing variable that
regresses on its own lagged, or prior, values.
• Integrated (I): represents the differencing1 of raw observations to allow the
time series to become stationary (i.e., data values are replaced by the difference
between the data values and the previous values).
• Moving Average (MA): incorporates the dependency between an observation
and a residual error from a moving average model applied to lagged observations.
Therefore, each component in ARIMA functions has a parameter with a standard
notation. For ARIMA models, a standard notation would be ARIMA with p, d, and
q, where integer values substitute for the parameters to indicate the type of ARIMA
model used. The parameters can be defined as:
• p: the number of lag observations in the model, also known as the lag order.
• d: the number of times the raw observations are differenced; also known as the
degree of differencing.
• q: the size of the moving average window, also known as the order of the
moving average.
In an ARIMA model, the data is differenced in order to make it stationary. A model
that shows stationarity2 is one that shows there is constancy to the data over time.
Most economic and market data show trends, so the purpose of differencing is to
remove any trends or seasonal structures. At that point, it is also relevant to note
that ARIMA models are complicated and work best on very large data sets, so that
computer algorithms and ML techniques are used to compute them.
With all of the above, ARIMA models have strong points and are good at forecasting
1
Differencing is a method of transforming a time series dataset. It can be used to remove the
series dependence on time, so-called temporal dependence. This includes structures like trends and
seasonality.
2
Stationarity means that the statistical properties of a process generating a time series do not
change over time.

83 José Manuel Rodrı́guez Lomeña

CHAPTER 10. MACHINE LEARNING AND STATISTICS

based on past circumstances, but there are reasons to be cautious when using ARIMA.
On the one hand, this model is good for short-term forecasting, and apart from
only needing historical data, it models non-stationary data. On the other hand, it
is not built for long-term forecasting, the prediction of turning points is poor, and
additionally it is computationally expensive.

10.2.2 Combination Without Repetition

A combination is a method used in statistics, which consists of finding the possible
ways a data set can be ordered. Hence, a combination without repetition of n
elements taken k is the different groups of k elements that can be formed by these n
elements, so that two groups differ only if they have different elements [49]. They
are represented as Cn,k , and follow that:
!
n n!
Cn,k = = (10.2)
k k!(n − k)!

José Manuel Rodrı́guez Lomeña 84

Chapter 11

Python Libraries

A Python library is a collection of related modules. It contains bundles of code that

can be used repeatedly in different programs. It makes Python Programming simpler
and convenient for the programmer, thus eliminating the need for writing codes from
scratch. Hence, Python libraries play a very vital role in fields of Machine Learning,
Data Science, and Data Visualization, among others.
Bearing that in mind, the aim of this chapter is to state constancy of the libraries
used to develop the trading bot code, in order to provide the reader with their
knowledge, and assume their full understanding in the description of the algorithm
in Chap. 13.
Therefore, the libraries are divided into five different categories: general libraries;
data analysis libraries; visualization libraries; trading libraries; and Machine Learning
libraries. All the information detailed below related to these libraries is extracted
from the Python Documentation website [48].

11.1 General Libraries

csv
The so-called CSV (Comma Separated Values) format is the most common import
and export format for spreadsheets and databases.
The csv module implements classes to read and write tabular data in CSV format.
It allows programmers to say, “write this data in the format preferred by Excel,” or
“read data from this file which was generated by Excel,” without knowing the precise
details of the CSV format used by Excel. Programmers can also describe the CSV
formats understood by other applications or define their own special-purpose CSV
formats.
The csv module’s reader and writer objects read and write sequences. Programmers
can also read and write data in dictionary form using the DictReader and DictWriter
classes.

85
CHAPTER 11. PYTHON LIBRARIES

math
This module provides access to the standard mathematical constants and functions
defined by the C standard. Programmers can use that module to perform vari-
ous mathematical calculations, such as numeric, trigonometric, logarithmic, and
exponential calculations.

os
This module provides the facility to establish the interaction between the user and
the operating system. It offers many useful OS functions that are used to perform
OS-based tasks and get related information about operating system. The OS comes
under Python’s standard utility modules.

random
The random module is a built-in module to generate the pseudo-random variables. It
can be used perform some action randomly such as to get a random number, selecting
a random elements from a list, or shuffling elements randomly.

sys
This module provides access to some variables used or maintained by the interpreter
and to functions that interact strongly with the interpreter. It is always available.

11.2 Data Analysis Libraries

pandas
It is a Python package that provides fast, flexible, and expressive data structures
designed to make working with ”relational” or ”labeled” data both easy and intuitive.
It aims to be the fundamental high-level building block for doing practical, real world
data analysis in Python. Additionally, it has the broader goal of becoming the most
powerful and flexible open source data analysis / manipulation tool available in any
language. It is already well on its way towards this goal.
Here are the main characteristics that pandas presents:
• Easy handling of missing data in both floating and non-floating point data.
• Size mutability: Columns can be inserted and deleted from DataFrame and
higher dimensional objects.
• Automatic and explicit data alignment: Objects can be explicitly aligned to a
set of labels, or the user can simply ignore the labels and let Series, DataFrame,
etc. automatically align the data for you in computations.
• Powerful, flexible group by functionality to perform split-apply-combine
operations on data sets, for both aggregating and transforming data.

José Manuel Rodrı́guez Lomeña 86

 11.3. VISUALIZATION LIBRARIES

• Make it easy to convert ragged, differently-indexed data in other Python

and NumPy data structures into DataFrame objects.
• Intelligent label-based slicing, fancy indexing, and subsetting of large data
sets.
• Intuitive merging and joining data sets.
• Flexible reshaping and pivoting of data sets.
• Hierarchical labeling of axes (possible to have multiple labels per tick).
• Robust IO tools for loading data from flat files (CSV and delimited), Excel
files, databases, and saving/loading data from the ultrafast HDF5 format.
• Time series-specific functionality: date range generation and frequency con-
version, moving window statistics, date shifting and lagging.

NumPy
NumPy (Numerical Python) is the fundamental package for scientific computing
with Python. It is used for working with arrays, and also has function for working in
domain of linear algebra, Fourier transform, and matrices.

datetime
The datetime module is one of the most common modules to manipulate date and
time object data. It supplies classes for manipulating dates and times, looping
through a range of dates, parsing and format date strings, and more.
While date and time arithmetic is supported, the focus of the implementation is on
efficient attribute extraction for output formatting and manipulation.

11.3 Visualization Libraries

matplotlib
Matplotlib is a comprehensive library for creating static, animated, and interactive
visualizations in Python. It makes easy things easier and hard things possible. For
instance, it allows:
• Creating publication quality plots.
• Making interactive figures that can zoom, pan, update.
• Customizing visual style and layout.
• Exporting to many file formats.
• Using a rich array of third-party packages built on Matplotlib.

87 José Manuel Rodrı́guez Lomeña

CHAPTER 11. PYTHON LIBRARIES

11.4 Trading Libraries

backtrader
It is a feature-rich Python framework for backtesting and trading. It allows focusing
on writing reusable trading strategies, indicators and analyzers instead of having to
spend time building infrastructure. In general, every complex component of ordinary
backtesting can be created with a single line of code by calling special functions.

talib
TA-Lib is widely used by trading software developers requiring to perform technical
analysis of financial market data. Among its main features, it stands out:
• Includes 150+ indicators such as ADX, MACD, RSI, Stochastic, Bollinger
Bands, etc.
• Candlestick pattern recognition.
• Open-source API for C/C++, Java, Perl, and Python.

yfinance
yfinance is a popular open source library as a means to access the financial data
available on Yahoo! Finance, which offers an excellent range of market data on
stocks, bonds, currencies and cryptocurrencies. For further information, reader is
encouraged to read Sec. 9.2.2 Yahoo! Finance.

binance
This package aims at connecting Python with Binance exchange, through the previous
generation of an API key in that exchange. It allows obtaining and reading specific
user data, in order the trading bot to operate in Binance.

11.5 Machine Learning Libraries

scikit-learn
scikit-learn is probably the most useful library for ML in Python, since it contains a
lot of efficient tools for ML and statistical modeling including classification, regression,
clustering and dimensionality reduction.

keras & tensorflow

keras is a high-level, deep learning API developed by Google for implementing neural
networks, and belongs to TensorFlow library. It is written in Python and is used to
make the implementation of neural networks easy. It also supports multiple back-end
neural network computation.

José Manuel Rodrı́guez Lomeña 88

Chapter 12

Studies

This chapter gathers studies related to this project that do not directly form its main
body, but are a fundamental part of its understanding. Specifically, some studies to
decide what percentage of the total capital is risked in each trade are carried out.

Apart from fixing a particular trading strategy, one of the most relevant features when
implementing the bot is knowing how much money is bet on each trade. Obviously,
that bet trade percentage will affect the Return Of Investment (ROI), also according
to the percentage of winning trades that the bot itself has.

Bearing the above in mind, some ideas to to analyze the relationship between these
variables arise. First of all, a code based on repetition and on random trade closures
is simulated. After that, a mathematical development on that percentage is carried
out, focusing on some variables of interest.

12.1 Simulation on Random Trade Closures

Before mathematically analyzing the described problem, a Python code is developed
in order to get a first insight in the expected results. Before proceeding with its
explanation, it is worth mentioning that this code is presented in Sec. B.1 Simulation
on Random Trade Closures.

The code consists of a trading simulator in which that closes trades randomly,
obviously always within a certain specified limits. First of all, it can be defined both
the minimum and the maximum percentage of winning trades that the bot executes.
Also, it is notorious to remark that the simulation can be carried out by deciding if
commissions are taken into account or not and, if proceed, select the percentage of
that commission. Moreover, number of trades to perform in each iteration of the
simulation and also the number of iterations can be chosen. Lastly, the limits for
both take profit (for winning trades) and stop-loss (for losing trades) are defined.

Therefore, the main core of that code consists of a for structure in which the
simulation is carried out for the entire range of percentage of winning trades. At

89
CHAPTER 12. STUDIES

that point, a combination without repetition1 function is also coded, as shown in

Sec. B.2 Combination Without Repetition. This combination is used for generating
all the possible index winning combinations for the given number of trades. In order
the reader to better understand the latter, a specific example with four trades and a
percentage winning trades of 50% is given:

[W, W, L, L]
[W, L, W, L]
[W, L, L, W ]
[L, W, W, L]
[L, W, L, W ]
[L, L, W, W ]

where W refers to won trades and L to lost trades. For the above example, there are
six different combinations, which is in accordance with Eq. (10.2). With n = 4 and
k = 0.5n = 2, Eq. (10.2) anticipates these six possible combinations.
The fact of performing different iterations for each combination is because either the
take profit or the stop-loss (depending on whether it is a winning or losing trade,
respectively) is a random number less than a specific chosen value. For instance,
random values up to +10% for take profit and up to -10% for stop-loss.
With all of the above, the trades are simulated for every percentage of winning
trades, traversing the specified range for the bet trade percentage. From that, some
numerical results are extracted and graphically displayed. In this case, it is of interest
to represent the ROI as a function of the bet trade percentage, for the whole range
of percentage of winning trades.
Having understood the purpose of the study and the operation of the code, the
results of a case with specific values of interest are shown and analyzed.
The percentage of winning trades is fixed from 48 to 52%. No commissions are
applied to the trades2 . The maximum value for take profit and stop-loss is 10%. The
number of trades is set to 100, and a total of 10 iterations are performed for each
winning combination. With that, the results shown in Fig. 12.1 are obtained.
At first instance, it can be observed that percentages of winning trades below 50%
clearly lead to a negative ROI, while percentages above 50% lead to a positive ROI.
At the midpoint, 50% assumes mainly negative ROI values, according to the average
obtained for each bet trade percentage. It can be said, therefore, that the 50% line
marks winning or losing money, as long as commissions are not charged on the trades.
Without a doubt, it would be expected that the fact of applying commissions would
1
The reader not familiar with combination without repetition is recommended to read Sec. 10.2.2
Combination Without Repetition.
2
As will be seen later, the idea is to trade Bitcoin on Binance, an exchange that does not charge
commissions for trading with said cryptocurrency.

José Manuel Rodrı́guez Lomeña 90

 12.1. SIMULATION ON RANDOM TRADE CLOSURES

lower the percentage of winning trades that would make money (positive ROI). In
other words, the higher the commissions, the higher percentage of winning trades
required to maintain the same ROI.
Regardless of the percentage, it is also concluded that as the bet trade percentage
increases, the greater the dispersion of the ROI obtained. From a mathematical
sense, the above makes perfect sense, since the take profit and stop loss are generated
randomly within each of the combinations without repetition to iterate.

Figure 12.1: Simulation on random trade closures

Looking at Fig. 12.1, there is no doubt that as long as it is ensured that the bot
operates with an effectiveness greater than 50%, or even 51% to be sure, the higher
the bet trade percentage, the higher the ROI obtained. However, here one must take
into account the fact that a losing streak can lead to the total bankruptcy of the
estate. In view of this and the clear dispersion that is presented in Fig. 12.1, it is
concluded that a good range in which to operate would be for a bet trade percentage
of between 2% and 4%.
Furthermore, it is worth mentioning that the fact of randomizing the closing of the
trades is not faithfully close to reality. Here, the trades would have to be closed
according to a specific strategy. However, it is intended to carry out a generic study
for the case presented.
Assessing all of the above, it was decided to mathematically analyze the above
variables, to relate them to each other and to be able to draw more general conclusions.

91 José Manuel Rodrı́guez Lomeña

CHAPTER 12. STUDIES

12.2 Analysis of Strategy Variables

As commented, a mathematical relationship is found for the proposed variables. To
do this, it is essential to make the definition of each variable clear:
• b: bet trade percentage
• t: take profit
• s: stop-loss
• p=b·t
• g =b·s
• A= g
p
= s
t
(A: stop-loss vs take profit factor)
• n: number of trades
• ω: number of won trades
• ωp = ω
n
(ωp : percentage of won trades)
With all these variables defined, the definition of winning and losing trades itself
leads to:

1 + ROI = (1 + p)ω · (1 − g)n−ω (12.1)

In Eq. (12.1), (1 + p)ω represents the ω won trades, while (1 − g)n−ω represents the
n − ω lost trades. To eliminate some variables, Eq. (12.1) can become:

1 + ROI = (1 + p)ω · (1 − Ap)n(1−ωp ) (12.2)

To eliminate the exponents, apply logarithms to each side of Eq. (12.2).

log(1 + ROI) = n [ωp · log(1 + p) + (1 − ωp ) · log(1 − Ap)] (12.3)

To simplify Eq. (12.3), the terms are rearranged:

log(1 + ROI)
= ωp [log(1 + p) − log(1 − Ap)] + log(1 − Ap) (12.4)
n

And continuing with the simplification:

log(1+ROI)
− log(1 − Ap)
ωp = n
(12.5)
log(1 + p) − log(1 − Ap)

Eq. (12.5) represents a relationship between five variables of interest: the percentage
of winning trades ωp , the return of investment ROI, the number of trades n, the

José Manuel Rodrı́guez Lomeña 92

 12.2. ANALYSIS OF STRATEGY VARIABLES

stop-loss vs take profit factor A, and the variable p, which is no more than the bet
trade percentage b multiplied by the take profit t.

As far as factor A is concerned, there is no doubt that it is a positive value. When

it takes values less than one, take profit becomes greater than stop-loss. This does
not only mean that take profit gives more money than the stop-loss removes, but
also that it is more probable to close a trade because of the stop-loss, by clearly
having less margin in which the price can fall. When A is exactly one, take profit and
stop-loss values are equal. And in case of taking values greater than one, stop-loss
becomes greater than take profit, and consequently, in this case it is more probable
to close a trade because of the take profit.

As for variable p, it takes values from 5 · 10−4 to 5 · 10−3 . This is because it is assumed
that b varies between 0.01 and 0.05, and that t comes from 0.05 to 0.1, all limits
included.

Lastly, n is limited to 40,000 trades, assuming that the bot can operate with the daily
number of trades it wishes, and therefore the time range of the study will depend on
the frequency with which the bot operates, obviously according to the implemented
strategy.

Once the variables are completely fixed, it is time to better analyze their relationship.
It is decided to represent the ROI versus the number of trades by fixing two of the
others three variables, and traversing a specific range for the other one. To achieve
that, Matlab code is implemented, and the three plots in Fig. 12.2, Fig. 12.3 and
Fig. 12.5 are obtained, from respectively the codes shown in Sec. B.3 ROI vs Number
of Trades for ωp Range, Sec. B.4 ROI vs Number of Trades for p Range and Sec. B.5
ROI vs Number of Trades for A Range.

Firstly, both A and p are fixed. ωp is varied from 0.48 to 0.52, since a lower value
than 0.48 would suppose that the strategy could be improved, and a higher value
than 0.52 would almost ensure a positive ROI.

At that point, it is of vital relevance to assess that the latter is true as long as factor
A takes values close to unity, since for values much greater than one, the benefit of
each won trade would be much greater than the loss of a trade lost, and therefore
more trades could be lost than won and still making solid money. And the same
would happen the other way around, in case of having a factor A that is much smaller
than unity, much more money would be lost with a stop-loss than the one that would
be gained with a take profit, and consequently, money could be lost even having a
ωp clearly higher than 50%.

Despite taking into account the above reasoning, the analysis is also carried out in
the indicated ωp range, obtaining then the plot in Fig. 12.2.

93 José Manuel Rodrı́guez Lomeña

CHAPTER 12. STUDIES

ROI vs number of trades [A = 1, p = 0.0005]

6

wp = 0.48
5 wp = 0.49
wp = 0.50

4 wp = 0.51
wp = 0.52

3
ROI

-1
0 0.5 1 1.5 2 2.5 3 3.5 4
Number of trades (n) 4
10

Figure 12.2: ROI vs number of trades for ωp range

With A factor set to one, the ROI becomes positive as long as ωp is above 50%, and
turns negative when ωp is less than 50%. When that percentage is just that value,
the ROI remains almost zero. Hence, it can be concluded that the greater ωp , the
better the ROI is, always taking into account that the zero-ROI-line (where hardly
any money is lost or earned) is affected by the A factor.
Now, it is intended to analyze what is the direct influence of variable p. It should be
expected that the higher the p value, greater tendency exists to win or lose money.
In other words, in a positive ROI tendency, a higher value of p is expected to lead
to even a greater positive tendency, and the same with a negative ROI tendency.
Indeed, the latter is displayed in Fig. 12.3. It is clearly observed that for a situation
in which ROI has to be negative (A = 1 and ωp = 0.498), an increase in p leads to
greater monetary losses.
Nevertheless, this does not happen for the entire ranges of ωp , as can be seen in
Fig. 12.4, where not only the rule of operating with a higher p value leads to a greater
tendency is broken, but also the tendency changes completely according to that value
of p.
It is worth taking a look at the plots obtained. There are three which presents a
positive ROI tendency, while one reverts to negative. The first three plots corresponds
with p values of 5 · 10−4 (blue), 1.5 · 10−3 (orange) and 2.5 · 10−3 (yellow). The point

José Manuel Rodrı́guez Lomeña 94

 12.2. ANALYSIS OF STRATEGY VARIABLES

ROI vs number of trades [A = 1, wp = 0.498]

0

-0.1

-0.2

-0.3

-0.4
ROI

-0.5

-0.6

-0.7
p = 0.0005
p = 0.0015
-0.8
p = 0.0025
p = 0.0050
-0.9

-1
0 0.5 1 1.5 2 2.5 3 3.5 4
Number of trades (n) 4
10

Figure 12.3: ROI vs number of trades for p range and ωp = 0.498

ROI vs number of trades [A = 1, wp = 0.5008]

0.15

0.1

0.05

-0.05
ROI

-0.1

-0.15

-0.2
p = 0.0005
p = 0.0015
-0.25
p = 0.0025
p = 0.0050
-0.3

-0.35
0 0.5 1 1.5 2 2.5 3 3.5 4
Number of trades (n) 4
10

Figure 12.4: ROI vs number of trades for p range and ωp = 0.5008

95 José Manuel Rodrı́guez Lomeña

CHAPTER 12. STUDIES

is that yellow and orange cases presents better results on ROI than blue case, but
orange case is better than yellow one. This means that, with A = 1 and the chosen
ωp = 0.5008, more money is earned by following a strategy with a p value of 1.5 · 10−3
rather than a value of 2.5 · 10−3 .
The explanation behind this is based on the nature of Eq. (12.5). It consists of a
fraction with different logarithms, some of whose arguments also contain more than
one variable. Hence, it is in fact expected that some non-linear behaviour arises,
at least in specific numerical ranges for any of the variables. Having observed the
above, it is clear that said equation should be studied mathematically in depth, as it
is commented in Chap. 17.
Lastly, it is intended to study the behavior of the ROI for different values of A, from
0.98 to 1.02, both included. With p = 5 · 10−4 and ωp = 0.5, Fig. 12.5 shows that
for values of A above 1, the ROI is expected to be negative, since the stop-loss is
greater than the take profit, and consequently more money is lost than earned. On
the contrary, for values of A below 1, with which take profit is greater than stop-loss
value, a positive ROI is obtained.

ROI vs number of trades [p = 0.0005, wp = 0.50]

0.6

0.5 A = 0.98
A = 0.99
0.4 A=1
A = 1.01
0.3 A = 1.02

0.2
ROI

0.1

-0.1

-0.2

-0.3

-0.4
0 0.5 1 1.5 2 2.5 3 3.5 4
Number of trades (n) 4
10

Figure 12.5: ROI vs number of trades for A range

In view of all the above, it is concluded that the most appropriate option is to
perform that analysis with the specific trading strategy, which would help to fix some
values and extract more particular conclusions.

José Manuel Rodrı́guez Lomeña 96

Chapter 13

Software Description

Software used for the development of the project is briefly described in this chapter.
Specifically, the less common software is explained, and it is understood that programs
such as Excel or Matlab are known to the reader.

13.1 Atom
Atom is a free and open-source text and source code editor for macOS, Linux, and
Microsoft Windows (Fig. 13.1). It supports plug-ins written in JavaScript, and
is embedded with GitHub. Atom enables users to install third-party packages to
customize the features and looks of the editor. Moreover, Atom’s default packages
can apply syntax highlighting for multiple programming languages and file formats.

Figure 13.1: The bot has been programmed in Atom environment

97
CHAPTER 13. SOFTWARE DESCRIPTION

13.2 GitHub Desktop

GitHub Desktop is an application that enables the interaction with GitHub using a
Graphical User Interface (GUI) instead of the command line or a web browser. It
allows completing most Git commands from the desktop with visual confirmation
of changes. One can push to, pull from, and clone remote repositories, and uses
collaborative tools such as attributing commits and creating pull requests.

13.3 Command Prompt

In Windows operating systems, the Command Prompt is a program that emulates
the input field in a text-based user interface screen with the Windows GUI. It can be
used to execute entered commands and perform advanced administrative functions.
Despite not being a specific software, requires to be flagged for code development, as
shown in Fig. 13.2.

Figure 13.2: Code has been run by using Command Prompt

José Manuel Rodrı́guez Lomeña 98

Chapter 14

Software Implementation

The proposal of this chapter is to detail the development of the implementation of the
algorithm proposed for the development of the trading bot. Firstly, its requirements
are stated and justified. And then, a further description of its development is detailed.
Please note that this chapter is not related to the description of neither the language
programming nor the explanation of how using Atom.

14.1 Requirements of the Algorithm

Apart from the method design itself, an essential aspect to take into account for
developing the algorithm is its requirements. The next list outlines them:

• The algorithm must be developed to trade through Binance exchange.

• It must use Yahoo! Finance data.

• It must be totally flexible. Variables must be easy to be changed. Different

strategies could be used at the same time.

• It must ensure the backtesting of strategies.

• It must be delivered before January 12th , 2023.

The algorithm is thereby implemented according to the first four features, and taking
into account the temporary imposition of the last requirement. At all times, it is
a priority that the algorithm is properly integrated into Binance exchange, since it
would not be useful from a practical point of view. Further, downloaded data must
be obtained from Yahoo! Finance platform. The bot must allow the user to easily
swap the value of the input variables, such as the bet percentage on each trade, the
pair being traded, or the time frequency of the trade. Lastly, the backtesting of any
developed strategy must be carried out before its own use in reality.

Studied the demanded requirements, the algorithm can already be implemented.

99
CHAPTER 14. SOFTWARE IMPLEMENTATION

14.2 Algorithm
The implemented algorithm is clearly separated into two different parts: one for the
trading bot itself; and another one for the backtesting. Moreover, each part has
been implemented from the development of several files, which facilitate both the
programming and the understanding of the algorithm.
Next, the description of the implementation of said files is presented, taking into
account that the reader has perfectly understood what was detailed in the previous
chapters. Before that, it is also notorious to remark that the code that is taught in
this chapter is not complete, since only the essential content for its explanation and
understanding is shown. The entire code is presented in Appendix A.

14.2.1 Bot
This part of the algorithm consists of the trading bot itself. Specifically, three
different files are implemented: botData, botCore, and botMain. The relationship that
these files have within the algorithm is the one observed in Fig. 14.1.

Figure 14.1: Diagram of the trading bot

Both botData and botCore files are implemented in order to ease the development
of the botMain file. On the one hand, botData sets all the necessary data, which
is chosen by the user, to make the bot run. On the other hand, botCore consists
of a class in which functions to properly implemented the bot are defined. These
functions include from the data downloading from Binance and its treatment, to the
creation of orders to execute the trading.
All the above functions are imported into the botMain file, in which the trading bot
itself is implemented. The bot is thus properly connected with Binance, and some
forward market orders has been created so as to verify the functionality1 . However,
1
More comments on that are presented in Sec. 14.2.1.3 botMain.

José Manuel Rodrı́guez Lomeña 100

 14.2. ALGORITHM

and as Chap. 2 progressed, the bot has not yet worked with any strategy in reality.
In other words, it has not yet really faced the financial market of cryptocurrencies,
since no real trade has been completed yet.

Therefore, it can be said that the bot has everything ready to work, despite the fact
that it has not done so in reality. This topic is covered in depth in Chap. 17, where
comments on the future work of this project are presented.

Having introduced the general structure of that part of the algorithm, it is time to
analyze each file, in order to better understand the algorithm itself.

14.2.1.1 botData
This class aims at setting the data to run the bot. The user should specify their
public and secret keys of the API created in their own Binance account. Furthermore,
the cryptocurrency pair being traded must also be selected. Also, start and end
dates of the period during which it is intended to trade, and the frequency of the
candlesticks to be traded.

Code Listing 14.1: The user can choose the settings to run the bot
# This class sets the data to run the bot .

# Python libraries .
import datetime as dt
from datetime import datetime

# Binance public and private keys .

BINANCE = {
" APIKey " : " XXXXXXX " ,
" SecretKey " : " XXXXXXX "
}

# Cryptocurrrency pair .
COIN = {
" Crypto " : " BTC " ,
" Fiat " : " USDT "
}

# Start and end dates ( year , month , day , hours , minutes , seconds ) .
DATE = {
" StartDate " : " 2022 - 05 - 01 10 : 00 : 00 " ,
" EndDate " : " 2022 - 07 - 15 00 : 00 : 00 "
# datetime . now ()
}

# Selected frequency .
FREQUENCY = " 15m "

Please note that both the real API and secret keys are not displayed in Code Listing
14.1 (XXXXXXX is shown) because of privacy reasons, as the author does not intend
to share his passwords with anyone else.

101 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

14.2.1.2 botCore
This file consists of a class called Bot BinanceClass, which at turn contains different
functions that are used to implement the bot. Despite the fact that are presented in
separated code listings, all the functions are within the aforementioned class.
Firstly, the constructor of the class2 is displayed. It assigns and stores all the
information necessary to trade, such as public and private keys, cryptocurrency pair
data, frequency to operate with, price data, commission, or balance.

Code Listing 14.2: Constructor of the class in botCore

# Class with Binance data and functions .
class Bot_BinanceClass :

def __init__ ( self , APIKey , secretKey , crypto , fiat , frequency ,

dataElements ) :

self . APIKey = APIKey

self . secretKey = SecretKey

self . crypto = crypto

self . fiat = fiat

self . frequency = frequency

self . dataElements = dataElements

self . apiBinance = Client ( self . APIKey , self . secretKey )

# Variables on purchase and sale operations .

self . BUY = SIDE_BUY
self . SELL = SIDE_SELL

self . pairInfo = self . apiBinance . get_symbol_info ( self . crypto

+ self . fiat )

self . cryptoDecimals = int ( self . pairInfo [ " b as eAs se tP re ci si on

"])
self . fiatDecimals = int ( self . pairInfo [ " quotePrecision " ] )

self . filtersData = self . pairInfo [ " filters " ]

self . PRICE_FILTER = self . filtersData [ 0 ]

self . LOT_SIZE = self . filtersData [ 2 ]
self . MIN_NOTIONAL = self . filtersData [ 3 ]

self . minPrice = float ( self . PRICE_FILTER [ " minPrice " ] )

self . maxPrice = float ( self . PRICE_FILTER [ " maxPrice " ] )
self . tickSize = float ( self . PRICE_FILTER [ " tickSize " ] )

self . minQty = float ( self . LOT_SIZE [ " minQty " ] )

self . maxQty = float ( self . LOT_SIZE [ " maxQty " ] )

2
A constructor is a special method of a class or structure in object-oriented programming that
initializes a newly created object of that type. Whenever an object is created, the constructor is
called automatically [53].

José Manuel Rodrı́guez Lomeña 102

 14.2. ALGORITHM

self . stepSize = float ( self . LOT_SIZE [ " stepSize " ] )

self . minNotional = float ( self . MIN_NOTIONAL [ " minNotional " ] )

self . pairFees = self . apiBinance . get_trade_fee ( symbol = self

. crypto + self . fiat ) [ 0 ]

self . makerCommission = float ( self . pairFees [ " makerCommission

"])
self . takerCommission = float ( self . pairFees [ " takerCommission
"])

self . cryptoBalance = self . apBinance . g et_ass et_ba lance ( self .

crypto ) [ " free " ]
self . fiatBalance = self . apiBinance . g et_ass et_ba lance ( self .
fiat ) [ " free " ]

The function dataFunction is used for filtering the required data. Pandas library is
used to transform the input data into a data frame, which is properly distributed
into the necessary columns: time, price related items, volume and trades. Then, this
data frame is transformed into date reference time, to finally being the output of the
function itself.
Code Listing 14.3: The data downloaded from Binance is filtered
def dataFilter ( Data ) :
df = pd . DataFrame ( Data )
df = df . drop ( [6 , 7 , 9 , 10 , 11 ] , axis = 1 )

columns = [ " time " , " open " , " high " , " low " , " close " , " volume "
, " trades " ]
df . columns = columns

for i in columns :
df [ i ] = df [ i ] . astype ( float )

df [ " start " ] = pd . to_datetime ( df [ " time " ] * 1000000 )

return df

The function updateData aims at updating the data of the last candle available. To
achieve that, the information of the new candle is got from Binance. Then, the data
of the first candle stored (in index 0) is removed, and the rest are moved one index
down, so that the new candle can be stored in the last position.
Code Listing 14.4: The data of the last candle registered by Binance is updated
def updateData ( self ) :
newCandle = self . apiBinance . get_klines ( symbol = self . crypto +
self . fiat , interval = self .
frequency , limit = 1 )
# limit = 1 -> the data of the last candle is downloaded
# interval -> interval of that candle ( frequency )
self . df . drop ( index = 0 , inplace = True ) # The first candle
is removed and the rest

103 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

are moved 1 index down

self . df = self . df . append ( self . dataFilter ( newCandle ) ,
ignore_index = True )
self . df . index = list ( range ( self . DataElements ) )

The function updateAccountBalance is used for updating the crypto and fiat balances.
The process is simple, since every time both balances are obtained from Binance.
Code Listing 14.5: Crypto and fiat balances are updated
def u p d a t e A c c o u n t B a l a n c e ( self ) :
self . cryptoBalance = self . apiBinance . g et_as set_ba lance ( self
. crypto ) [ " free " ]
self . fiatBalance = self . apiBinance . g et_as set_ba lance ( self .
fiat ) [ " free " ]

The function marketOrder aims at creating a market order to be executed. A try-

except structure is used. The function always tries to create the order, which needs
to know in which direction of the market is intended to trade (buy or sell) and the
quantity to be risked. In case of not being possible of creating the order, a Binance
API exception is printed.
Code Listing 14.6: Creation of a generic market order
def marketOrder ( self , operationSide , quantity ) :
try :
self . OrderName = self . apiBinance . create_order (
symbol = self . Crypto + self . Fiat ,
side = operationSide ,
type = " MARKET " ,
qty = quantity
)

except B i na n c eA P I Ex c e pt i o n as e :
print ( e )

The function limitOrder works similar to the function marketOrder in terms of

structure, but in that case time and price limit values are also demanded to create
the order.
Code Listing 14.7: Creation of a limit market order
def limitOrder ( self , operationSide , quantity , price ) :
try :
self . orderName = self . apiBinance . create_order (
symbol = self . crypto + self . fiat ,
side = operationSide ,
type = " LIMIT " ,
timeInForce = TIME_IN_FORCE_GTC ,
qty = " { :. { } f } " . format ( quantity , self .
cryptoDecimals ) ,
price = " { :. { } f } " . format ( price , self . cryptoDecimals
)
)

José Manuel Rodrı́guez Lomeña 104

 14.2. ALGORITHM

except B i na n c eA P I Ex c e pt i o n as e :
print ( e )

The function notifyOrder is used for notifying orders that has been executed, taking
into account that an order could be completed with more than one purchase. For
instance, when a market order is created, the bot will try to buy at the lower possible
price, and because of price and time delays, it is possible that all quantity may not
be purchased at the same price. For that reason, a for structure which considers all
the different possible purchases is needed. Here, the commissions are also calculated.
Lastly, and by means of the function orderRegister, the data of the order is registered.

Code Listing 14.8: An order that hast just been filled is notified
def notifyOrder ( self ) :
totalPrice = 0
totalQuantity = 0
averagePrice = 0
totalComissions = 0

for i in self . newOrder [ " fills " ] : # It is necessary to

calculate the mean of the
order
totalComissions + = float ( i [ " commission " ] )
totalQuantity + = float ( i [ " qty " ] )
totalPrice + = float ( i [ " price " ] ) * float ( i [ " qty " ] )

averagePrice = totalPrice / totalQuantity

# Registering the order data

self . orderRegister ( self . newOrder [ " type " ] , self . newOrder [ "
side " ] , totalComissions ,
self . newOrder [ "
commissionAsset " ] ,
averagePrice ,
totalQuantity , totalPrice
)

The function orderRegister aims at stating constancy of all the orders executed, both
purchases and sales. All of them are written into a text file, following a specific
structure for the data. It is relevant to note that in order to be able to write into
the file, it firstly should be opened, and closed once the order has been registered.

Code Listing 14.9: All the purchase and sale orders are registered
def orderRegister ( self , orderType , side , comission , coinComission ,
avgPrice , qtyCrypto , qtyFiat ) :
f1 = open ( " OrdersData / Registro_ { } _ { } _BOT . txt " . format ( self .
crypto + self . fiat , self .
frequency ) , " a + " )

if os . stat ( " OrdersData / Registro_ { } _ { } _BOT . txt " . format ( self .

crypto + self . fiat , self .
frequency ) ) . st_size = = 0 :

105 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

f1 . write ( " DATE \ t \ t \ t ORDER_TYPE \ t ORDER \ t PRICE \ t

CRYPTO_QUANTITY \ t
FIAT_QUANTITY \ t
COMISSION \ t
COIN_COMISSION \ n " )

f1 . write ( " { } \ t { } \ t { } \ t \ t { } \ t \ t \ t \ t { } \ t \ t \ t { } \ t \ t
{ } \ t \ t { } \ n " . format ( str
( dt . datetime . now () ) ,
orderType , side , str (
avgPrice ) , str ( qtyCrypto )
, str ( qtyFiat ) , comission
, coinComission ) )

f1 . close ()

The function getCandleData is used for reading the downloaded candlestick data
from the .csv file.
Code Listing 14.10: Reading the downloaded candlestick data
def getCandleData ( self ) :
self . candleData = pd . read_csv ( " MarketData / { } { } / Freq_ { } . csv "
. format ( self . crypto , self
. fiat , self . frequency ) )
self . dataCandles = len ( self . candleData )

All the functions displayed in Code Listing 14.11 consists of the calculation of specific
indicators that can be later used in the trading strategy implementation. Note
that the numerical calculations are obtained from talib library and stored in the
candlestick data.
Code Listing 14.11: Calculation of some different indicators
# Calculation of the SMA Indicator
def getSMA ( self , days ) :
self . candleData [ " SMA_ { } " . format ( days ) ] = ta . SMA ( self . candleData
[ " close " ] , days )

# Calculation of the EMA Indicator

def getEMA ( self , days ) :
self . candleData [ " EMA_ { } " . format ( days ) ] = ta . EMA ( self . candleData
[ " close " ] , days )

# Calculation of Stochastic Indicator

def getStochastic ( self , days , pfast , pslow ) :
STO = ta . stochastic ( self . candleData , period = days , pfast =
pfast , pslow = pslow )
self . candleData [ " STO_k_ { } _ { } _ { } " . format ( days , pfast , pslow ) ] =
STO . k
self . candleData [ " STO_d_ { } _ { } _ { } " . format ( days , pfast , pslow ) ] =
STO . d

José Manuel Rodrı́guez Lomeña 106

 14.2. ALGORITHM

14.2.1.3 botMain
The trading bot itself is implemented in this file. As has been commented, the bot is
ready to work in reality, but has not done it yet. It is clearly that this is one of the
next imminent steps to work on, as it is discussed in Chap. 17.
Anyway, the bot is connected with Binance and is able to operate, since it downloads
data price, updates it, and creates orders to start the trading. Obviously, these orders
have to be automatically created from the based on a trading strategy previously
validated in backtesting.
Having clarified the above, it is time to understand the code thoroughly. Firstly,
the correspondent libraries, options and data are imported. Binance client library is
imported to get and store both API and secret keys. Also, botData and botCore files
are imported, and data related to available frequencies to trade with in Binance is
generated.

Code Listing 14.12: Importing libraries and data

# Binance libraries .
from binance . client import Client

# Imported files .
sys . path . insert (0 , os . path . abspath ( ’ ../../.. ’) )
from BinanceKeys import BINANCE_KEYS
from botData import BINANCE , COIN , DATE , FREQUENCY
import botCore

# API data .
apiBinance = Client ( BINANCE [ " APIKey " ] , BINANCE [ " SecretKey " ] )

# Available frequency in Binance .

fr eq ue nc yA va il ab le = [ " 1m " , " 3m " , " 5m " , " 15m " , " 30m " , " 1h " , " 2h " , "
4h " , " 6h " , " 8h " , " 12h " , " 1d " , " 3d
" , " 1w " , " 1M " ]

The function downloadCandlesData aims at downloading the candlestick data from

Binance. According to the selected time range in which to trade (start and end
dates from botData), the total time range in seconds is obtained. Then, as selected
frequency to trade with gives the time each candle lasts (for that frequency), the
total number of candles is calculated as the division of the total time by the time
of each candle. At that point, it is remarkable to note that the time duration of
the candle is obtained by an if-elif structure in which all the frequency cases are
considered, and for each one the correspondent time is calculated.
Before returning the number of candles to download, the subsequent data is already
downloaded from Binance, taking the frequency as the interval and this number of
candles as limit to stop downloading. This data is filtered and stored in a csv file.

Code Listing 14.13: Downloading candlestick data from Binance

def d ow n l o ad C a nd l e sD a t a ( Coin , Fiat , Frequency , StartDate , EndDate ) :

107 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

# Calculating the total number of candles to download according

to the selected time range
FI = dt . datetime ( int ( StartDate [ 0 : 4 ] ) , int ( StartDate [ 5 : 7 ] ) , int (
StartDate [ 8 : 10 ] ) , int (
StartDate [ 11 : 13 ] ) ,
int ( StartDate [ 14 : 16 ] ) , int ( StartDate [ 17 : 19 ] ) )
FF = dt . datetime ( int ( EndDate [ 0 : 4 ] ) , int ( EndDate [ 5 : 7 ] ) , int (
EndDate [ 8 : 10 ] ) , int ( EndDate [
11 : 13 ] ) ,
int ( EndDate [ 14 : 16 ] ) , int ( EndDate [ 17 : 19 ] ) )
# EndDate
Total_Time = math . floor (( FF - FI ) . total_seconds () )

if " m " in Frequency :

if ( len ( Frequency ) ) = = 2 :
CandleSeconds = 60 * float ( Frequency [ 0 ] )
CandleMinutes = int ( Frequency [ 0 ] )
else :
CandleSeconds = 60 * float ( Frequency [ : 2 ] )
CandleMinutes = int ( Frequency [ : 2 ] )
elif " h " in Frequency :
if ( len ( Frequency ) ) = = 2 :
CandleSeconds = 3600 * float ( Frequency [ 0 ] )
CandleMinutes = 60 * int ( Frequency [ 0 ] )
else :
CandleSeconds = 3600 * float ( Frequency [ : 2 ] )
CandleMinutes = 60 * int ( Frequency [ : 2 ] )
elif " d " in Frequency :
CandleSeconds = 86400 * float ( Frequency [ 0 ] )
CandleMinutes = 24 * 60 * int ( Frequency [ 0 ] )
elif " w " in Frequency :
CandleSeconds = 7 * 86400 * float ( Frequency [ 0 ] )
CandleMinutes = 7 * 24 * 60 * int ( Frequency [ 0 ] )
elif " M " in Frequency :
CandleSeconds = 30 * 86400 * float ( Frequency [ 0 ] )
CandleMinutes = 30 * 24 * 60 * int ( Frequency [ 0 ] )

DataElements = math . floor ( Total_Time / CandleSeconds )

Header = [ " time " , " open " , " high " , " low " , " close " , " volume " , "
trades " , " start " ]
Data = apiBinance . get_klines ( symbol = Coin + Fiat , interval =
Frequency , limit =
DataElements )
Data = in itialD ataFi lter ( Data )

if not os . path . isdir ( " MarketData / { } { } / " . format ( Coin , Fiat ) ) :

os . mkdir ( " MarketData / { } { } / " . format ( Coin , Fiat ) )

if os . path . exists ( " MarketData / { } { } / Freq_ { } . csv " . format ( Coin ,

Fiat , Frequency ) ) :
os . remove ( " MarketData / { } { } / Freq_ { } . csv " . format ( Coin , Fiat ,
Frequency ) )

Data . to_csv ( " MarketData / { } { } / Freq_ { } . csv " . format ( Coin , Fiat ,

José Manuel Rodrı́guez Lomeña 108

 14.2. ALGORITHM

Frequency ) )

return DataElements

Code Listing 14.14 consists of the code lines which finally implements the bot. First
of all, it is checked if the selected frequency to trade with is available in Binance, and
if so, the data is downloaded, filtered and stored, according to all the parameters
defined by the user in botData. In case of the frequency not being available, an error
message is printed and the bot does nothing. Secondly, the bot variable is created
from the class of botCore file, by simply calling it.
At that point, the strategy would have to be called and from there start trading
from the creation of orders that appear thanks to the intelligence developed for the
strategy.

Code Listing 14.14: Trading bot implementation

if checkFrequency ( frequencyAvailable , FREQUENCY ) :
DataElements = d o wn l o ad C a nd l e sD a t a ( COIN [ " Crypto " ] , COIN [ " Fiat " ]
, FREQUENCY , DATE [ " StartDate "
] , DATE [ " EndDate " ] )
else :
print ( " Error for the selected frequency ( { } ) \ n " . format (
FREQUENCY ) )

Bot = botCore . Bot_BinanceClass ( BINANCE [ " APIKey " ] , BINANCE [ "

SecretKey " ] , COIN [ " Crypto " ] , COIN
[ " Fiat " ] , FREQUENCY , DataElements
)

14.2.2 Backtesting
This part of the algorithm aims at validating the implemented trading strategies
before using them in reality by the bot algorithm. Hence, it is a necessary prior step
to verify that the strategy to follow is really going to generate a positive ROI. As
progressed before, this validation is carried out by using cryptocurrency data from
Yahoo! Finance, since this platform allows downloading data for a lot of coins and
for their entire time range.
Here, six different files are developed: settings, sizers, downloadData, strategy, back-
testingCore, backtestingMain. Their relationship within the algorithm is shown in
Fig. 14.2.
The backtestingMain file collects the functions of all the other files to end up imple-
menting the backtesting. Both settings and sizers files aims at setting specific data
and variables. Later, the downloadData file is used to download the correspondent
market information data from Yahoo. The trading strategy to be backtested is
defined in the strategy file. And lastly, the backtestingCore aims at defining functions
to be used for the backtesting. It is relevant to remark that this file also calls three
other files: settings, strategy and downloadData, as represented in Fig. 14.2.

109 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

Figure 14.2: Diagram of the backtesting

Having introduced the general structure of that part of the algorithm, it is time to
analyze each file, in order to better understand the algorithm itself.

14.2.2.1 settings
This class aims at setting the data for the backtesting. The user must specify the
cryptocurrency pair (present on the Yahoo! Finance list), the start and end dates
and the frequencies which with the backtesting is intended to be performed.
Please note that this trading bot is intended to not only trade cryptocurrencies, but
also stocks in the future. For this reason, a stock (present on the Yahoo! Finance
list) must also be indicated and the type of asset to be tested: 0 for cryptocurrencies,
1 for stocks.
Code Listing 14.15: The user can choose the settings to run the backtesting
# Python libraries
import datetime as dt
from datetime import datetime

# Cryptocurrency pair ( must be in Yahoo Finance list ) .

PAIR = {
" Base " : " BTC " ,
" Quote " : " USD "
}

# Stocks ( must be in Yahoo Finance list ) .

STOCK = {
" Ticker " : " TSLA "
}

José Manuel Rodrı́guez Lomeña 110

 14.2. ALGORITHM

# Defining asset type : 0 for cryptocurrency , 1 for stocks .

ASSET = {
" AssetType " : 0
}

# Start and end dates ( year - month - day ) for the backtesting (
datetime . now () for current date ) .
DATE = {
" StartDate " : " 2022 - 01 - 01 " ,
" EndDate " : " 2022 - 07 - 15 "
}

# Frequencies available for the candlestick data ( must be in Yahoo

Finance list ) .
FREQUENCIES = [ " 1m " , " 2m " , " 5m " , " 15m " , " 30m " , " 60m " , " 90m " , " 1h " ,
" 1d " , " 5d " , " 1wk " , " 1mo " , " 3mo " ]

14.2.2.2 sizers
The sizer of the bet trade percentage is set in this class, by importing the percent
sizer from the backtrader library. Although this is a simple setting, it is separated
into a specific code file because of the possibility of adding new sizers in the future.

Code Listing 14.16: Sizers for the backtesting are defined

# Python libraries
from backtrader . sizers import PercentSizer

class FullMoney ( PercentSizer ) :

params = (
( " percents " , 3 ) ,
)

From the analysis developed in Chap. 12, it has been decided to use a bet trade
percentage of 3%, since with a take profit/stop-loss of a 5%, it is in a range for p
without expected high dispersion and with a value that seems to maximize the return
for a strategy that wins a bit more than half the total trades, which is the aim to
reach.

14.2.2.3 downloadData
This file defined the functions for extraction the datasets containing the information
of the market (candles) from Yahoo! Finance. Specifically, two different functions
are presented3 . The first one checks if the frequency selected is on Yahoo, and the
second one downloads the requested data.
The function checkFrequency initializes a boolean named check being false. The list
of frequency available in Yahoo is traversed by means of a for instruction. Only in
3
In fact, three different functions are coded in the file, but the third one is the same as the one
for downloading data for crypto, but for stocks. The entire code is presented in Appendix A.

111 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

case that the selected frequency is found in that list, the boolean check becomes true.
After that, in case of that boolean is still false, a warning message is printed to the
user, and the available frequencies are displayed. The function returns the value of
the boolean check.

Code Listing 14.17: Checking if the selected frequency is on Yahoo

def checkFrequency ( FrequencyAva : list , Frequency : str ) :

check = False
for i in range ( len ( FrequencyAva ) ) :
if Frequency = = FrequencyAva [ i ] :
check = True

if check = = False :
print ( " The selected frequency of { } is not available .
Select one of the
following frequencies : \
n " . format ( Frequency ) )
for i in range ( len ( FrequencyAva ) ) :
print ( " { } \ t " . format ( FrequencyAva [ i ] ) )
exit ()

return check

As commented, the second function is for downloading the candlestick data into .csv
files. Firstly, note that the download is performed only if the selected frequency
exists (if not, again an error message appears). Obviously, this checking is carried
out by means of the checkFrequency function detailed above.

Code Listing 14.18: Downloading candlestick data

def dataCrypto ( FrequencyAva : list , Frequency : str , Pair : tuple ,
StartDate : str , EndDate : str ) :

# Downloading in case the frequency exists

if checkFrequency ( FrequencyAva , Frequency ) :

# Downloading data from Yahoo Finance

Data = yf . download ( ’{ } -{ } ’. format ( Pair [ " Base " ] , Pair [ " Quote
" ] ) , start = StartDate ,
end = EndDate ,
index_as_date = False ,
interval = Frequency )
Data [ ’ time ’] = Data . index
Data [ ’ time ’] = Data [ ’ time ’] . dt . strftime ( ’%Y - %m - % d % H : % M : % S ’
)

# Generating the directory of the data

if not os . path . isdir ( " MarketData / Crypto / { } { } / " . format ( Pair [
" Base " ] , Pair [ " Quote " ] ) ) :
os . mkdir ( " MarketData / Crypto / { } { } / " . format ( Pair [ " Base " ] ,
Pair [ " Quote " ] ) )

# Deleting the name in case it already exists

José Manuel Rodrı́guez Lomeña 112

 14.2. ALGORITHM

if os . path . exists ( " MarketData / Crypto / { } { } / Freq_ { } . csv " .

format ( Pair [ " Base " ] , Pair
[ " Quote " ] , Frequency ) ) :
os . remove ( " MarketData / Crypto / { } { } / Freq_ { } . csv " . format (
Pair [ " Base " ] , Pair [ "
Quote " ] , Frequency ) )

# Generating . csv file from DataFrame

Data . to_csv ( " MarketData / Crypto / { } { } / Freq_ { } . csv " . format (
Pair [ " Base " ] , Pair [ " Quote
" ] , Frequency ) )

else :
print ( " Error for the selected frequency ( { } ) \ n " . format (
Frequency ) )

14.2.2.4 strategy
This file aims at creating a model on which the trading operations are based. Specifi-
cally, a Recurrent Neural Network through an ARIMA model4 is computed. Hence,
it is essential to note that here only the operation of the algorithm is described, and
that no reference is made to the parameters required by the strategy to carry out
the backtesting. All this is developed in Chap. 15, where the results obtained by
applying the entire backtesting algorithm are also discussed.
The file basically consists of a class called Strategy, whose constructor develops the
aforementioned neural network.
Firstly, the data used for training the neural network is obtained by means of down-
loadData file. This data is preprocessed from numpy and keras libraries. Specifically,
the data is converted into a training set, which is adequately reshaped according to
its own length, to later be scaled using the MinMaxScaler() function. Lastly, the set
is divided into a dependent and a independent variable to follow the implementation
of an ARIMA model.
With that, the RNN is initialized, and the subsequent layers are added. This is done
by using Sequential() class, which is a framework for creating neural network models.
That model is fitted with the data sets from the ARIMA model. This means that
the model itself learns how to operate from the provided historical data.
Once the model is trained, a test model is created so as to be able to backtest the
effectiveness of the neural network. Obviously, the dates with which the model is
tested cannot coincide with the period during the model has been trained. More
comments on that are detailed in Chap. 15.
Code Listing 14.19: Constructor of the strategy file
class Strategy ( bt . Strategy ) :

def __init__ ( self ) :

4
Please read Chap. 10 carefully, where it is detailed what a RNN is and the ARIMA model.

113 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

# Downloading training data

self . trainingData = downloadData . dataCrypto ( FREQUENCIES ,
dataFrequency , PAIR , DATE
[ " StartDate " ] , DATE [ "
EndDate " ] )

# Data preprocessing
self . trainingSet = self . trainingData . values
self . trainingSet = np . reshape ( self . trainingSet , ( len ( self .
trainingSet ) , 1 ) )
self . trainingSet = MinMaxScaler () . fit_transform ( self .
trainingSet )
self . lengthTS = len ( self . trainingSet )
self . trainingX = self . trainingSet [ 0 : self . lengthTS - 1 ]
self . trainingX = np . reshape ( self . trainingX , ( len ( self .
trainingX ) , 1 , 1 ) )
self . trainingY = self . trainingSet [ 1 : self . lengthTS ]

# Initializing the RNN

self . RNN = Sequential ()

# Output layer is added

self . RNN . add ( Dense ( units = 1 ) )

# Input layer and LSTM layer are added

self . RNN . add ( LSTM ( units = 4 , activation = ’ sigmoid ’ ,
input_shape = ( None , 1 ) ) )

# Compiling the RNN

self . RNN . compile ( optimizer = ’ adam ’ , loss = ’
me an _s qu ar ed _e rr or ’)

# Fitting the RNN to the training data set

self . RNN . fit ( self . trainingX , self . trainingY , batch_size = 5
, epochs = 100 )

# Creating a test model

self . testingData = downloadData . dataCrypto ( FREQUENCIES ,
dataFrequency , PAIR , DATE
[ " StartDate " ] , DATE [ "
EndDate " ] )
self . testSet = self . testingData . values
self . input = np . reshape ( self . testSet , len ( self . testSet ) , 1 )
self . input = MinMaxScaler () . transform ( self . input )
self . input = np . reshape ( self . input , ( len ( self . input ) , 1 , 1 )
)
self . testModel = self . RNN . predict ( self . input )
self . testModel = MinMaxScaler () . i nvers e_tran sform ( self .
testModel )

14.2.2.5 backtestingCore
This file consists of a class called BacktestingClass, which at turn contains different
functions that are used to develop the backtesting. Despite the fact that are presented

José Manuel Rodrı́guez Lomeña 114

 14.2. ALGORITHM

in separated code listings, all the functions are within the aforementioned class.
Firstly, the constructor of the class is displayed. It sets all the information for the
backtesting and downloads the subsequent candlestick data, which at turn is entered
into the strategy class.
The most relevant feature is the import of the Cerebro class5 , which allows gathering
all the data, the strategy and the analyzers while ensuring a correct functioning at
any moment. Furthermore, it allows the backtesting to be executed, returning the
results and giving access to the plotting facilities.
It is also outstanding to note that this constructor is designed to operate with both
cryptocurrencies and stocks, despite the fact that this project is focused exclusively
on the first type of asset.
Code Listing 14.20: Constructor of the BacktestingClass
class BacktestingClass :

def __init__ ( self , assetType : int , pair : dict , stock : str , date
: dict , dataFrequency : list ) :

self . pair = pair

self . stock = stock

self . dtStartDate = dt . datetime ( int ( date [ " StartDate " ] [ 0 : 4 ] ) ,

int ( date [ " StartDate " ] [ 5 :
7 ] ) , int ( date [ " StartDate "
] [ 8 : 10 ] ) )
self . dtEndDate = dt . datetime ( int ( date [ " EndDate " ] [ 0 : 4 ] ) , int
( date [ " EndDate " ] [ 5 : 7 ] ) ,
int ( date [ " EndDate " ] [ 8 : 10 ]
))
self . cerebro = bt . Cerebro ()

for freq in dataFrequency :

if assetType = = 0 :
# Downloading data for cryptocurrency
downloadData . dataCrypto ( FREQUENCIES , freq , self .
pair , date [ "
StartDate " ] , date
[ " EndDate " ] )

# Cryptocurrency data feeded into the backtesting

data = bt . feeds . Y ah o o Fi n a nc e C SV D a t a (
dataname = ( " MarketData / Crypto / { } { } / Freq_ { } . csv
" . format ( self
. pair [ " Base " ]
, self . pair [ "
Quote " ] , freq
)),
fromdate = self . dtStartDate ,
5
The Cerebro class is an open class which belongs to the backtrader library. Please read Sec. 11.4
Trading Libraries to know more about this library.

115 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

todate = self . dtEndDate ,

reverse = False )

elif assetType = = 1 :
# Downloading data for stocks
downloadData . dataStocks ( FREQUENCIES , freq , self .
stock , date [ "
StartDate " ] , date [
" EndDate " ] )

# Stocks data feeded into the backtesting

data = bt . feeds . Y ah o o Fi n a n ce C S VD a t a (
dataname = ( " MarketData / Stocks / { } / Freq_ { } . csv " .
format ( self .
stock , freq ) )
,
fromdate = self . dtStartDate ,
todate = self . dtEndDate ,
reverse = False )

self . cerebro . adddata ( data )

The function setInitialMoney is used for setting the initial cash with which is intended
to trade.
Code Listing 14.21: Setting initial cash of the portfolio
def setInitialMoney ( self , money ) :

self . cerebro . broker . setcash ( money )

self . InitialMoney = money

The function printCurrentMoney displays on screen which is the portfolio value at

the requested time.
Code Listing 14.22: Printing the current portfolio value
def pri ntCurr entMon ey ( self ) :

print ( " Current Portfolio Value : % . 2f " % self . cerebro . broker

. getvalue () )

The function setCommissions aims at setting the commission applied in each trade
of the specific cryptocurrency which is intended to trade with on Binance.
Code Listing 14.23: Setting Binance commissions
def setComissions ( self , comissions ) :

self . cerebro . broker . setcommission ( commission = comissions )

The function setSizers is used for setting the correspondent sizers for trading. In
this case, the bet percentage for each trade is the only sizer which is set.
Code Listing 14.24: Setting sizers

José Manuel Rodrı́guez Lomeña 116

 14.2. ALGORITHM

def setSizers ( self , sizer ) :

self . cerebro . addsizer ( sizer )

The function setStrategy simply sets the strategy to be backtesting, adding it to

cerebro.
Code Listing 14.25: Setting backtesting strategy
def setStrategy ( self , strategy ) :

self . cerebro . addstrategy ( strategy )

The function runStrategy runs the implemented strategy by means of cerebro. Addi-
tionally, the results of that strategy are analyzed by the talib library.
Code Listing 14.26: Running the implemented strategy
def runStrategy ( self ) :

self . strats = self . cerebro . run ()

self . strat = self . strats [ 0 ]
self . results = self . strat . analyzers . ta . get_analysis ()

The function plotBacktestingResults is used for plotting backtesting results. Specifi-

cally, it displays the market analyzers and backtesting plots by calling the printMar-
ketAnalyzers, which will be explained later, and the plot functionality of cerebro.
Code Listing 14.27: Plotting backtesting results
def p l o t B a c k t e s t i n g R e s u l t s ( self ) :

self . p r i n t M a r k e t A n a l y z e r s ()
self . cerebro . plot ()

The function setBotAnalyzers sets the bot analyzers. Hence, the family of Analyzer
objects aims at provide an analysis of what has happened or even of what is actually
happening with the implemented trading system, both in terms of profit and risk.
To deepen the understanding and use of these analyzers, the reading of [3] is highly
recommended.
Code Listing 14.28: Setting bot analyzers
def setBotAnalyzers ( self ) :

self . cerebro . addanalyzer ( bt . analyzers . TradeAnalyzer , _name =

" ta " )

The function printMarketAnalyzers is used for displaying the market analyzers plots.
If there are no trades, this function warns the user.
Code Listing 14.29: Printing market analyzers plot
def p r i n t M a r k e t A n a l y z e r s ( self ) :

117 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

results = self . strat . analyzers . ta . get_analysis ()

if results . total . total = = 0 :

print ( " There have been no trades for the current
backtesting . " )
return

else :
closedTrades = results . total . closed
wonTrades = results . won . total
winRatio = wonTrades / closedTrades
netProfit = results . pnl . net . total
totalReturn = ( netProfit / self . InitialMoney ) * 100

totalDays = ( self . dtEndDate - self . dtStartDate ) . days

monthlyReturn = totalReturn * 30 . 0 / totalDays

fig , ax = plt . subplots (1 , 1 )

column = [ " Results " ]

rows = [ " Closed trades " ,

" Won trades " ,
" Win ratio " ,
" Net profit " ,
" Total return " ,
" Start period " ,
" End period " ,
" Monthly return " ]

results = [ [ " { } " . format ( closedTrades ) ] ,

[ " { } " . format ( wonTrades ) ] ,
[ " { :. 2f } % " . format ( 100 * winRatio ) ] ,
[ " { :. 2f } { } " . format ( netProfit , self . pair [
" Quote "
])],
[ " { :. 2f } % " . format ( totalReturn ) ] ,
[ " { } " . format ( self . dtStartDate ) ] ,
[ " { } " . format ( self . dtEndDate ) ] ,
[ " { :. 2f } % " . format ( monthlyReturn ) ] ]

df = pd . DataFrame ( results , columns = column )

ax . axis ( ’ tight ’)
ax . axis ( ’ off ’)
ax . table = plt . table ( cellText = df . values ,
rowLabels = rows ,
colLabels = column ,
colWidths = None ,
loc = " center " )
ax . table . set_fontsize ( 20 )
ax . table . scale ( 1 .8 , 1 . 8 )

for ( row , col ) , cell in ax . table . get_celld () . items () :

if ( row = = 0 ) or ( col = = - 1 ) :

José Manuel Rodrı́guez Lomeña 118

 14.2. ALGORITHM

cell . set_text_props ( fontproperties =

FontProperties
( weight = ’ bold
’) )

ax . table . a u t o _ s e t _ c o l u m n _ w i d t h ( col = list ( range ( len ( df .

columns ) ) ) )
plt . title ( " Backtesting Results " , fontsize = 30 ,
fontweight = " bold " )

14.2.2.6 backtestingMain
This file is formed by calling functions of all the others files, causing the backtesting
to be carried out. Firstly, the backtesting strategy variable is created from calling
the class of the backtestingCore file, which needs some data from settings and strategy
as input variables. With that, several functions from the backtestingCore file are
properly called. The initial money is arbitrarily set to $1,000, and the bet trade
percentage from sizers is loaded to the backtesting data information. Then, the
analyzers to trade and the strategy are also set. And lastly, the strategy is run to be
backtested, and the subsequent results are displayed.
It is also notorious that although there is a line which sets the commissions that apply
when trading, it is indeed commented. This is why the strategy to be backtested
trades with Bitcoin, cryptocurrency whose trading is commission free on Binance.
When trading with commissions, this line would have to be uncommented and would
simply subtract the corresponding commissions from the total capital. In this case,
since there are no commissions, it is decided to eliminate the line, since despite
the fact that the code would work the same and would not take into account any
commission (applying a 0), thus the algorithm does not have to directly go through
a line that does not does nothing for this specific case.
Code Listing 14.30: Strategy is backtested
backStrategy = backtestingCore . BacktestingClass ( ASSET [ " AssetType " ] ,
PAIR , STOCK [ " Ticker " ] , DATE ,
strat . dataFrequency )
# Bot_BackTest . setComissions ( Bot . makerCommission )
backStrategy . setInitialMoney ( 1000 . 0 )
backStrategy . setSizers ( sizers . FullMoney )
backStrategy . setBotAnalyzers ()
backStrategy . setStrategy ( strat . Strategy )
backStrategy . runStrategy ()
backStrategy . p rintCu rrentM oney ()
backStrategy . p l o t B a c k t e s t i n g R e s u l t s ()

As mentioned, backtesting results are obtained for the implemented strategy. This is
undoubtedly the next step, which is performed in Chap. 15.

119 José Manuel Rodrı́guez Lomeña

CHAPTER 14. SOFTWARE IMPLEMENTATION

José Manuel Rodrı́guez Lomeña 120

Chapter 15

Results

This chapter aims to state constancy of the results obtained for the backtesting of
the implemented strategy, which serves to verify that the bot can actually generate
money. Specifically, numerical values of the parameters that shape the neural network
of the trading strategy are discussed, in addition to the results when said strategy is
backtested.
Before commenting on all of the above, it is relevant to note that compliance with
all algorithm requirements is achieved. These requirements are detailed in Sec. 14.1
Requirements of the Algorithm, and are justified throughout the explanation of the
algorithm in Sec. 14.2 Algorithm.
As previously explained in Sec. 14.2.2.4 strategy, the trading strategy consists
of a RNN through an ARIMA model, which has been trained to trade Bitcoin.
This training has been carried out based on BTC data from September 17, 2014
to December 31, 2022, both included1 . The training data has been limited to
2020 to have data with which to backtest the model, since it is obvious that the
backtesting data must be different from the training data, otherwise the bot would
not be encountering a real trading situation, but rather it would face already known
operations.
Hence, in order the backtesting to be performed, data from January 1, 2021 to
November 30, 2022 is selected. Furthermore, some other relevant parameters should
also be defined. As concluded from the studies developed in Chap. 12, a good
combination of parameters would be to set the bet trade percentage to 3% and the
take profit and stop-loss values to 5%. Obviously, as Bitcoin is the cryptocurrency
to be traded, the pair is BTC/USDT and the asset type2 is cryptocurrency.
As far as the frequency of the candlesticks is concerned, in this case it is convenient
to select a frequency of 15 minutes, since it is intended to make dozens of trades a
day, so fast intraday trading is required. For this reason, it is also decided to train
the model and test it for frequencies of 5 and 30 minutes.
1
BTC data is available in Yahoo! Finance from September 17, 2014
2
Remember that the algorithm is designed to also operate in stocks in the future, so asset type
is also necessary to be fixed.

121
CHAPTER 15. RESULTS

With all of the above, the backtesting is run, obtaining the results shown below. At
that point, it should be remembered that no commissions are applied in Binance
when trading with BTC, and also that backtesting parameters are explained in
Sec. 9.2.1 Backtesting.

15.1 Backtesting for a Frequency of 15 Minutes

The implemented trading strategy is firstly backtested for a frequency range of 15
minutes. The results with that frequency are then shown in Table (15.1).

Table 15.1: Backtesting results for a frequency of 15 minutes

Metric Value
Annualized Net Return 7.3675%
Profitability 50.3621%
Average Take Profit/Stop-Loss 1.2397
Maximum Drawdown 1.1838%
Annualized Volatility 0.5876%
Annualized Volatility Negative Returns 0.7664%
Sharpe Ratio 12.5383
Sortino Ratio 9.6131

From Table (15.1), it is undoubtedly that the implemented trading strategy is really
sound and ready to be executed in reality with a candlestick frequency of 15 minutes.
Firstly, the net return is above 7%, due to a profitability slightly higher than 50%
for the total of 7,236 trades during the two-year period. This also implies operating
with a positive average return in percentage per trade, since the total take profit is
higher than the total stop-loss.
As for the risk metrics, despite having a considerable higher maximum drawdown (in
comparison with the net return) the analyzed volatilities are low enough to ensure
that the risk offered is medium. In line with that, the return-risk ratios are clearly
above 2, which indicates that the expected return clearly compensates for the risk
taken.

15.2 Backtesting for a Frequency of 5 Minutes

Table (15.2) presents the obtained results for the backtesting for a frequency of 5
minutes.
Despite the fact that all the metrics worsen compared to operating with a frequency
of 15 minutes, it is clear that operating for a frequency of 5 minutes would not be a
mistake, as consistent profits are also generated.
In fact, the return is quite similar, only 0.14% lower. However, it should be considered
that the decrease of the profitability from 50.3621% to 50.2144% would have implied

José Manuel Rodrı́guez Lomeña 122

 15.3. BACKTESTING FOR A FREQUENCY OF 30 MINUTES

a lower return than that 7.2223% obtained. The point here is that by decreasing the
frequency range, the number of total trades has increased to 13,146 during the two
years, which obviously allows obtaining a higher return, as more winning trades are
closed.
In view of the above, it is mandatory to comment on commissions when trading,
since it is obvious that if trading with commissions, the strategy would be clearly
diminished when operating with a frequency of 5 minutes, because the number of
trades made is clearly greater. Undoubtedly, this must be considered when designing
a trading strategy, and without a doubt, a way to reduce commissions should always
be tried to find, since commissions is return that is lost directly.

Table 15.2: Backtesting results for a frequency of 5 minutes

Metric Value
Annualized Net Return 7.2223%
Profitability 50.2144%
Average Take Profit/Stop-Loss 1.1387
Maximum Drawdown 1.3704%
Annualized Volatility 0.6854%
Annualized Volatility Negative Returns 0.8559%
Sharpe Ratio 10.5373
Sortino Ratio 8.4372

The analysis of the metrics can be extrapolated to that carried out for the frequency
of 15 minutes, since the values obtained are very similar. In a generic way, it can
be concluded that the return metrics are once again acceptable and the return-risk
ratio is still high.

15.3 Backtesting for a Frequency of 30 Minutes

Lastly, the results for a frequency of 30 minutes are displayed in Table (15.3).

Table 15.3: Backtesting results for a frequency of 30 minutes

Metric Value
Annualized Net Return 2.6817%
Profitability 50.2389%
Average Take Profit/Stop-Loss 1.0533
Maximum Drawdown 2.5566%
Annualized Volatility 0.9276%
Annualized Volatility Negative Returns 1.0157%
Sharpe Ratio 2.8910
Sortino Ratio 2.6403

123 José Manuel Rodrı́guez Lomeña

CHAPTER 15. RESULTS

Even though the profitability when operating with a frequency of 30 minutes is

almost the same than that for the frequency of 5 minutes, again the total number of
trades takes a vital role for the return of the strategy. Now, 4,378 trades has been
performed during the two-year period, which involves in closing less winning trades
and, consequently, in reducing benefits.
Apart from that, the maximum drawdown has increased considerably, becoming
almost as large as the expected return, a fact that certainly weakens the strategy.
Moreover, and in accordance with the above, the return-risk ratios have decreased to
just above 2.
It cannot be said that trading with this 30 minute candles is not a winning strategy,
but it is clearly worse than trading with 5 or 15 minute candles. In addition, it would
be necessary to see if the fact of operating with commissions would mean ending up
losing money with this frequency.

15.4 Strategy Analysis

The trading strategy implemented from the RNN has passed the backtesting, and
it clearly works better when operation for a frequency of 15 minutes, despite the
fact that a similar profitability is obtained for all the candlestick frequencies studied.
Hence, the strategy offers a clearly positive return, with not a high risk, and above
all, with a high return-risk ratio. Without a doubt, the model is ready to operate in
reality.
However, perhaps it would be convenient to evaluate it with more data (from different
years), but because the cryptocurrency market is so recent (data from around a
decade is only available), there is not enough data available to train the model and
be able to test it over a long period of time.
At that point, the idea of operating in the stock market reappears3 , since a quantity
of data from decades would be available, which would allow both training the model
for a longer period of time, and also making a more reliable backtesting.

3
It is recalled that the bot is designed to be able to operate in the stock market, and it would
only be necessary to connect to a platform that operates with stocks, since Binance does not.

José Manuel Rodrı́guez Lomeña 124

Chapter 16

Conclusions

In this project, a cryptocurrency trading bot has been developed. Its algorithm is
divided into two differentiated parts: one for the bot itself, and the other for the
backtesting of trading strategies.
The part for the bot itself is only coded, and the work here consisted of verifying that
the connectivity with Binance is correct and the algorithm is capable of executing
automatic trades. Hence, all the functions to run the strategy have been developed
and their proper functioning verified.
As far as the backtesting part is concerned, its subsequent algorithm is also developed,
obtaining the cryptocurrency data from Yahoo! and ensuring that any strategy could
be backtested by simply calling its specific strategy file.
For this project, one trading strategy has been implemented and backtested, achieving
results with a consistent positive return of 7.36% and an acceptable risk to operate.
This means that this strategy can be coupled to the algorithm of the bot and forms
the basis for trading on Binance.
A fundamental fact to keep in mind when trading is that the number of trades
and, therefore, the frequency of the candlesticks totally affect the return obtained,
obviously in addition to the profitability and possible commissions. It has been
possible to see that for a strategy without commissions, for similar profitabilities, as
the number of trades increases, the return increases. This agrees with the results
obtained in the studies of Chap. 12.
Thus, it makes sense that the number of trades is greater for frequencies of minor
candles. Here, therefore, it must be assessed that the fact of changing the frequency
also affects profitability and, consequently, depending on the strategy followed, it
will not always be better to operate with lower frequencies, not only because of the
return, but also for the risk.
It is also outstanding to note that the backtesting part has been designed in different
files to be able to backtest different strategies with some ease, simply by changing
settings and sizers parameters. In this sense, the bot algorithm would also allow

125
CHAPTER 16. CONCLUSIONS

trading with more than one strategy at the same time by simply creating another
class of bot with a different strategy.
With all of the above, this project can be considered an absolute success, since
the entire base of an algorithm for a trading bot has been developed, guaranteeing
compliance with all the initial requirements. The results achieved are really promising,
and this leads to the expectation of generating money automatically once the bot
operates in reality.
Finally, these last lines are intended to record that this Master’s Degree Thesis is
much more than a college work. It has meant getting fully into what could be a
large-scale engineering project, since not only has the world of cryptocurrencies been
deepened, but it has also gone through the technical analysis of the market, the
powerful field of Machine Learning, and object-oriented programming through Python.
Undoubtedly, specializing in these fields provides the author with a multitude of new
tools with great practical-technical application.

José Manuel Rodrı́guez Lomeña 126

Chapter 17

Future Work

Despite the fact that good results have been achieved, this project needs to continue
to be developed. This chapter is then related to expose what should be done in the
near future, basing on everything stated in the previous chapters.
After verifying through backtesting that the implemented strategy achieves a good
percentage of winning trades and that, therefore, everything seems to indicate that
the bot can generate money, the first imminent step to develop is to make the bot
trade in reality through Binance. To do this, the strategy file should be simply called
within the main of the bot, implementing the relevant buy and sell orders.
Without a doubt, this will be a critical step in determining the success of the bot,
not only because of the effectiveness that the strategy in question can offer, but also
because of making sure that the bot operates as expected. Any glitches, no matter
how minuscule, will be revealed when the trades are executed in reality.
At this point, it will be necessary to carry out an analysis of the bot’s operation,
with the aim of improving it. For instance, optimizing its response time to a possible
trade, or even modifying its algorithmic structure for better performance.
Once there is no doubt that the bot works optimally, another possible improvement
would be the implementation of new trading strategies. For this, it would be necessary
to deepen both Machine Learning and time series analysis techniques, which allow a
more severe bot training. In turn, the fundamentals in technical analysis could be of
great help to finish setting a more robust strategy in terms of reliability.
Parallel to the above, it is essential to place more emphasis on the study of the bet
trade percentage. It would be necessary to analyze the proposed variables in greater
depth from a mathematical point of view, taking into account that each specific
strategy can work in certain ranges for the relevant variables.
Despite the fact that not paying commissions when trading with Bitcoin is really
advantageous, it would be convenient to check the effectiveness of the strategies for
other cryptocurrencies, since perhaps a higher return can be achieved.
In this sense, the development of the bot could also be extended to the stock market,

127
CHAPTER 17. FUTURE WORK

also taking advantage of the fact that the algorithm has already been programmed
with this intention. Here, however, the bot would have to be connected to some
stock platform. Clearly, it would be ideal to separate this part of connecting with
the exchange, so as not to overlap lines of code and simplify the development of the
algorithm.
Lastly, and as a more long-term step, it could be considered using a personal cloud,
in which the large amount of data generated could be stored, and all files would
remain protected.

José Manuel Rodrı́guez Lomeña 128

Appendix A

Bot Code

In this appendix, it is presented the full developed code for the trading bot algorithm
explained in Chap. 14, both the bot and backtesting part.

A.1 botData
Code Listing A.1: botData file code
# This class sets the data to run the bot .

# Python libraries .
import datetime as dt
from datetime import datetime

# Binance public and private keys .

BINANCE = {
" APIKey " : " XXXXXXX " ,
" SecretKey " : " XXXXXXX "
}

# Cryptocurrrency pair .
COIN = {
" Crypto " : " BTC " ,
" Fiat " : " USDT "
}

# Start and end dates ( year , month , day , hours , minutes , seconds ) .
DATE = {
" StartDate " : " 2022 - 05 - 01 10 : 00 : 00 " ,
" EndDate " : " 2022 - 07 - 15 00 : 00 : 00 "
# datetime . now ()
}

# Selected frequency .
FREQUENCY = " 15m "

129
APPENDIX A. BOT CODE

A.2 botCore
Code Listing A.2: botCore file code
# TRADING BOT CLASS FILE : functions defined to implement the bot .

# Python libraries .
import pandas as pd
import numpy as np
import datetime as dt
import math
import os
import talib as ta

# Binance libraries .
from binance . client import Client
from binance . exceptions import B i na n c eA P I Ex c e pt i o n
from binance . enums import *

# Class with Binance data and functions .

class Bot_BinanceClass :

def __init__ ( self , APIKey , secretKey , crypto , fiat , frequency ,

dataElements ) :

self . APIKey = APIKey

self . secretKey = SecretKey

self . crypto = crypto

self . fiat = fiat

self . frequency = frequency

self . dataElements = dataElements

self . apiBinance = Client ( self . APIKey , self . secretKey )

# Variables on purchase and sale operations .

self . BUY = SIDE_BUY
self . SELL = SIDE_SELL

self . pairInfo = self . apiBinance . get_symbol_info ( self . crypto

+ self . fiat )

self . cryptoDecimals = int ( self . pairInfo [ " b as eAs se tP re ci si on

"])
self . fiatDecimals = int ( self . pairInfo [ " quotePrecision " ] )

self . filtersData = self . pairInfo [ " filters " ]

self . PRICE_FILTER = self . filtersData [ 0 ]

self . LOT_SIZE = self . filtersData [ 2 ]
self . MIN_NOTIONAL = self . filtersData [ 3 ]

self . minPrice = float ( self . PRICE_FILTER [ " minPrice " ] )

self . maxPrice = float ( self . PRICE_FILTER [ " maxPrice " ] )
self . tickSize = float ( self . PRICE_FILTER [ " tickSize " ] )

José Manuel Rodrı́guez Lomeña 130

 A.2. BOTCORE

self . minQty = float ( self . LOT_SIZE [ " minQty " ] )

self . maxQty = float ( self . LOT_SIZE [ " maxQty " ] )
self . stepSize = float ( self . LOT_SIZE [ " stepSize " ] )

self . minNotional = float ( self . MIN_NOTIONAL [ " minNotional " ] )

self . pairFees = self . apiBinance . get_trade_fee ( symbol = self

. crypto + self . fiat ) [ 0 ]

self . makerCommission = float ( self . pairFees [ " makerCommission

"])
self . takerCommission = float ( self . pairFees [ " takerCommission
"])

self . cryptoBalance = self . apBinance . g et_ass et_ba lance ( self .

crypto ) [ " free " ]
self . fiatBalance = self . apiBinance . g et_ass et_ba lance ( self .
fiat ) [ " free " ]

# Filtering of data downloaded from Binance

def dataFilter ( Data ) :
df = pd . DataFrame ( Data )
df = df . drop ( [6 , 7 , 9 , 10 , 11 ] , axis = 1 )

columns = [ " time " , " open " , " high " , " low " , " close " , " volume "
, " trades " ]
df . columns = columns

for i in columns :
df [ i ] = df [ i ] . astype ( float )

df [ " start " ] = pd . to_datetime ( df [ " time " ] * 1000000 )

return df

# Updating the data of the last candle registered by Binance

def updateData ( self ) :
newCandle = self . apiBinance . get_klines ( symbol = self . crypto +
self . fiat , interval = self .
frequency , limit = 1 )
# limit = 1 -> the data of the last candle is downloaded
# interval -> interval of that candle ( frequency )
self . df . drop ( index = 0 , inplace = True ) # The first candle
is removed and the rest
are moved 1 index down
self . df = self . df . append ( self . dataFilter ( newCandle ) ,
ignore_index = True )
self . df . index = list ( range ( self . DataElements ) )

# Updating the data on the number of coins of the selected

asset
def u p d a t e A c c o u n t B a l a n c e ( self ) :

131 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

self . cryptoBalance = self . apiBinance . g et_as set_ba lance ( self

. crypto ) [ " free " ]
self . fiatBalance = self . apiBinance . g et_as set_ba lance ( self .
fiat ) [ " free " ]

# Creating a generic market order ( usually Market )

def marketOrder ( self , operationSide , quantity ) :
try :
self . OrderName = self . apiBinance . create_order (
symbol = self . Crypto + self . Fiat ,
side = operationSide ,
type = " MARKET " ,
qty = quantity
)

except B i na n c eA P I Ex c e pt i o n as e :
print ( e )

# Creating a market LIMIT order

def limitOrder ( self , operationSide , quantity , price ) :
try :
self . orderName = self . apiBinance . create_order (
symbol = self . crypto + self . fiat ,
side = operationSide ,
type = " LIMIT " ,
timeInForce = TIME_IN_FORCE_GTC ,
qty = " { :. { } f } " . format ( quantity , self .
cryptoDecimals ) ,
price = " { :. { } f } " . format ( price , self . cryptoDecimals
)
)

except B i na n c eA P I Ex c e pt i o n as e :
print ( e )

# Creating a test order to verify that it works

def testOrder ( self , operationSide , quantity ) :
try :
self . newOrder = self . apiBinance . c reate _test_ order (
symbol = self . crypto + self . fiat ,
side = operationSide ,
type = " MARKET " ,
qty = " { :. { } f } " . format ( quantity , self .
cryptoDecimals ) ,
)

except B i na n c eA P I Ex c e pt i o n as e :
print ( e )

# Notifying that an order hast just been filled

def notifyOrder ( self ) :
totalPrice = 0
totalQuantity = 0
averagePrice = 0
totalComissions = 0

José Manuel Rodrı́guez Lomeña 132

 A.2. BOTCORE

for i in self . newOrder [ " fills " ] : # It is necessary to

calculate the mean of the
order
totalComissions + = float ( i [ " commission " ] )
totalQuantity + = float ( i [ " qty " ] )
totalPrice + = float ( i [ " price " ] ) * float ( i [ " qty " ] )

averagePrice = totalPrice / totalQuantity

# Registering the order data

self . orderRegister ( self . newOrder [ " type " ] , self . newOrder [ "
side " ] , totalComissions ,
self . newOrder [ "
commissionAsset " ] ,
averagePrice ,
totalQuantity , totalPrice
)

# Registering all the purchase and sale orders made by the bot
def orderRegister ( self , orderType , side , comission ,
coinComission , avgPrice ,
qtyCrypto , qtyFiat ) :
f1 = open ( " OrdersData / Registro_ { } _ { } _BOT . txt " . format ( self .
crypto + self . fiat , self .
frequency ) , " a + " )

if os . stat ( " OrdersData / Registro_ { } _ { } _BOT . txt " . format ( self .

crypto + self . fiat , self .
frequency ) ) . st_size = = 0 :
f1 . write ( " DATE \ t \ t \ t ORDER_TYPE \ t ORDER \ t PRICE \ t
CRYPTO_QUANTITY \ t
FIAT_QUANTITY \ t
COMISSION \ t
COIN_COMISSION \ n " )

f1 . write ( " { } \ t { } \ t { } \ t \ t { } \ t \ t \ t \ t { } \ t \ t \ t { } \ t \ t
{ } \ t \ t { } \ n " . format ( str
( dt . datetime . now () ) ,
orderType , side , str (
avgPrice ) , str ( qtyCrypto )
, str ( qtyFiat ) , comission
, coinComission ) )

f1 . close ()

# Reading the downloaded candlestick data

def getCandleData ( self ) :
self . candleData = pd . read_csv ( " MarketData / { } { } / Freq_ { } . csv "
. format ( self . crypto , self
. fiat , self . frequency ) )
self . dataCandles = len ( self . candleData )

# Calculation of the SMA Indicator

def getSMA ( self , days ) :

133 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

self . candleData [ " SMA_ { } " . format ( days ) ] = ta . SMA ( self .

candleData [ " close " ] , days
)

# Calculation of the EMA Indicator

def getEMA ( self , days ) :
self . candleData [ " EMA_ { } " . format ( days ) ] = ta . EMA ( self .
candleData [ " close " ] , days
)

# Calculation of Stochastic Indicator

def getStochastic ( self , days , pfast , pslow ) :
STO = ta . stochastic ( self . candleData , period = days , pfast =
pfast , pslow = pslow )
self . candleData [ " STO_k_ { } _ { } _ { } " . format ( days , pfast , pslow )
] = STO . k
self . candleData [ " STO_d_ { } _ { } _ { } " . format ( days , pfast , pslow )
] = STO . d

A.3 botMain
Code Listing A.3: botMain file code
# TRADING BOT MAIN CODE : Binance trading bot code .

# Python libraries .
import pandas as pd
import numpy as np
import csv
import datetime as dt
import math
import os
import random
import sys

# Python libraries options .

pd . options . mode . ch ain ed _a ss ig nm en t = None # default = ’ warn ’

# Binance libraries .
from binance . client import Client

# Imported files .
sys . path . insert (0 , os . path . abspath ( ’ ../../.. ’) )
from BinanceKeys import BINANCE_KEYS
from botData import BINANCE , COIN , DATE , FREQUENCY
import botCore

# API data .
apiBinance = Client ( BINANCE [ " APIKey " ] , BINANCE [ " SecretKey " ] )

# Available frequency in Binance .

f req ue nc yA va il ab le = [ " 1m " , " 3m " , " 5m " , " 15m " , " 30m " , " 1h " , " 2h " , "
4h " , " 6h " , " 8h " , " 12h " , " 1d " , " 3d
" , " 1w " , " 1M " ]

José Manuel Rodrı́guez Lomeña 134

 A.3. BOTMAIN

# Checking for a correct frequency .

def checkFrequency ( freqAva : list , Frequency : str ) :
check = False
for i in range ( len ( freqAva ) ) :
if Frequency = = freqAva [ i ] :
check = True

if check = = True :
print ( " The selected frequency of { } is correct . \ n " . format (
Frequency ) )
else :
print ( " The selected frequency of { } is NOT correct . \ n " .
format ( Frequency ) )

return check

# Filtering of downloaded data .

def init ialDa taFilt er ( Data ) :
df = pd . DataFrame ( Data )
df = df . drop ( [6 , 7 , 9 , 10 , 11 ] , axis = 1 )

columns = [ " time " , " open " , " high " , " low " , " close " , " volume " , "
trades " ]
df . columns = columns

for i in columns :
df [ i ] = df [ i ] . astype ( float )

df [ " start " ] = pd . to_datetime ( df [ " time " ] * 1000000 , format = ’%Y - %m -
%d %H:%M:%S’ )

return df

def d ow n l o ad C a nd l e sD a t a ( Coin , Fiat , Frequency , StartDate , EndDate ) :

# Calculating the total number of candles to download according

to the selected time range
FI = dt . datetime ( int ( StartDate [ 0 : 4 ] ) , int ( StartDate [ 5 : 7 ] ) , int (
StartDate [ 8 : 10 ] ) , int (
StartDate [ 11 : 13 ] ) ,
int ( StartDate [ 14 : 16 ] ) , int ( StartDate [ 17 : 19 ] ) )
FF = dt . datetime ( int ( EndDate [ 0 : 4 ] ) , int ( EndDate [ 5 : 7 ] ) , int (
EndDate [ 8 : 10 ] ) , int ( EndDate [
11 : 13 ] ) ,
int ( EndDate [ 14 : 16 ] ) , int ( EndDate [ 17 : 19 ] ) )
# EndDate
Total_Time = math . floor (( FF - FI ) . total_seconds () )

if " m " in Frequency :

if ( len ( Frequency ) ) = = 2 :
CandleSeconds = 60 * float ( Frequency [ 0 ] )
CandleMinutes = int ( Frequency [ 0 ] )
else :
CandleSeconds = 60 * float ( Frequency [ : 2 ] )

135 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

CandleMinutes = int ( Frequency [ : 2 ] )

elif " h " in Frequency :
if ( len ( Frequency ) ) = = 2 :
CandleSeconds = 3600 * float ( Frequency [ 0 ] )
CandleMinutes = 60 * int ( Frequency [ 0 ] )
else :
CandleSeconds = 3600 * float ( Frequency [ : 2 ] )
CandleMinutes = 60 * int ( Frequency [ : 2 ] )
elif " d " in Frequency :
CandleSeconds = 86400 * float ( Frequency [ 0 ] )
CandleMinutes = 24 * 60 * int ( Frequency [ 0 ] )
elif " w " in Frequency :
CandleSeconds = 7 * 86400 * float ( Frequency [ 0 ] )
CandleMinutes = 7 * 24 * 60 * int ( Frequency [ 0 ] )
elif " M " in Frequency :
CandleSeconds = 30 * 86400 * float ( Frequency [ 0 ] )
CandleMinutes = 30 * 24 * 60 * int ( Frequency [ 0 ] )

DataElements = math . floor ( Total_Time / CandleSeconds )

Header = [ " time " , " open " , " high " , " low " , " close " , " volume " , "
trades " , " start " ]
Data = apiBinance . get_klines ( symbol = Coin + Fiat , interval =
Frequency , limit =
DataElements )
Data = in itialD ataFi lter ( Data )

if not os . path . isdir ( " MarketData / { } { } / " . format ( Coin , Fiat ) ) :

os . mkdir ( " MarketData / { } { } / " . format ( Coin , Fiat ) )

if os . path . exists ( " MarketData / { } { } / Freq_ { } . csv " . format ( Coin ,

Fiat , Frequency ) ) :
os . remove ( " MarketData / { } { } / Freq_ { } . csv " . format ( Coin , Fiat ,
Frequency ) )

Data . to_csv ( " MarketData / { } { } / Freq_ { } . csv " . format ( Coin , Fiat ,
Frequency ) )

return DataElements

# TRADING BOT IMPLEMENTATION

if checkFrequency ( frequencyAvailable , FREQUENCY ) :
DataElements = d ow n l o ad C a nd l e sD a t a ( COIN [ " Crypto " ] , COIN [ " Fiat " ]
, FREQUENCY , DATE [ " StartDate "
] , DATE [ " EndDate " ] )
else :
print ( " Error for the selected frequency ( { } ) \ n " . format (
FREQUENCY ) )

Bot = botCore . Bot_BinanceClass ( BINANCE [ " APIKey " ] , BINANCE [ "

SecretKey " ] , COIN [ " Crypto " ] , COIN
[ " Fiat " ] , FREQUENCY , DataElements
)

José Manuel Rodrı́guez Lomeña 136

 A.4. SETTINGS

A.4 settings
Code Listing A.4: settings file code
# This class sets the data for the backtesting .

# Python libraries
import datetime as dt
from datetime import datetime

# Cryptocurrency pair ( must be in Yahoo Finance list ) .

PAIR = {
" Base " : " BTC " ,
" Quote " : " USD "
}

# Stocks ( must be in Yahoo Finance list ) .

STOCK = {
" Ticker " : " TSLA "
}

# Defining asset type : 0 for cryptocurrency , 1 for stocks .

ASSET = {
" AssetType " : 0
}

# Start and end dates ( year - month - day ) for the backtesting (
datetime . now () for current date ) .
DATE = {
" StartDate " : " 2022 - 01 - 01 " ,
" EndDate " : " 2022 - 07 - 15 "
}

# Frequencies available for the candlestick data ( must be in Yahoo

Finance list ) .
FREQUENCIES = [ " 1m " , " 2m " , " 5m " , " 15m " , " 30m " , " 60m " , " 90m " , " 1h " ,
" 1d " , " 5d " , " 1wk " , " 1mo " , " 3mo " ]

A.5 sizers
Code Listing A.5: sizers file code
# This class defines the sizers for the backtesting .

# Python libraries
from backtrader . sizers import PercentSizer

class FullMoney ( PercentSizer ) :

’’’
Setting the percentage of money to invest in a trade .

@param PercentSizer : percentage of money to be invested .

137 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

@return none
’’’

params = (
( " percents " , 3 ) ,
)

A.6 downloadData
Code Listing A.6: downloadData file code
# BACKTESTING DATA EXTRACTION : User defined functions for
extracting the datasets
containing the information of the
market ( candles ) from Yahoo
Finance .

# Python Libraries
import pandas as pd
import numpy as np
import csv
import datetime as dt
import math
import os
import random
import yfinance as yf

def checkFrequency ( FrequencyAva : list , Frequency : str ) :

’’’
Checking for a correct frequency : Check if the frequency of the
data selected by the user
exists on the repository of
Yahoo Finance .
Otherwise , prints an error message suggesting the available
frequencies .

@param FrequencyAva : List of frequencies available from

Yahoo Finance .
@param Frequency : User desired frequency .

@return check : Boleean variable that is true if Frequency

exists in FrequencyAva .
’’’

check = False
for i in range ( len ( FrequencyAva ) ) :
if Frequency = = FrequencyAva [ i ] :
check = True

if check = = False :
print ( " The selected frequency of { } is not available .
Select one of the
following frequencies : \

José Manuel Rodrı́guez Lomeña 138

 A.6. DOWNLOADDATA

n " . format ( Frequency ) )

for i in range ( len ( FrequencyAva ) ) :
print ( " { } \ t " . format ( FrequencyAva [ i ] ) )
exit ()

return check

def dataCrypto ( FrequencyAva : list , Frequency : str , Pair : tuple ,

StartDate : str , EndDate : str ) :

’’’
Downloading the candlestick data into . csv files for
cryptocurrencies .

@param FrequencyAva : List of frequencies avaliable from

Yahoo Finance .
@param Frequency : User desired frequency .
@param Pair : Coin pair operated .
@param StartDate : Starting date of the downloaded data .
@param EndDate : Ending date of the downloaded data .

@return None
’’’

# Downloading in case the frequency exists

if checkFrequency ( FrequencyAva , Frequency ) :

# Downloading data from Yahoo Finance

Data = yf . download ( ’{ } -{ } ’. format ( Pair [ " Base " ] , Pair [ " Quote
" ] ) , start = StartDate ,
end = EndDate ,
index_as_date = False ,
interval = Frequency )
Data [ ’ time ’] = Data . index
Data [ ’ time ’] = Data [ ’ time ’] . dt . strftime ( ’%Y - %m - % d % H : % M : % S ’
)

# Generating the directory of the data

if not os . path . isdir ( " MarketData / Crypto / { } { } / " . format ( Pair [
" Base " ] , Pair [ " Quote " ] ) ) :
os . mkdir ( " MarketData / Crypto / { } { } / " . format ( Pair [ " Base " ] ,
Pair [ " Quote " ] ) )

# Deleting the name in case it already exists

if os . path . exists ( " MarketData / Crypto / { } { } / Freq_ { } . csv " .
format ( Pair [ " Base " ] , Pair
[ " Quote " ] , Frequency ) ) :
os . remove ( " MarketData / Crypto / { } { } / Freq_ { } . csv " . format (
Pair [ " Base " ] , Pair [ "
Quote " ] , Frequency ) )

# Generating . csv file from DataFrame

Data . to_csv ( " MarketData / Crypto / { } { } / Freq_ { } . csv " . format (
Pair [ " Base " ] , Pair [ " Quote
" ] , Frequency ) )

139 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

else :
print ( " Error for the selected frequency ( { } ) \ n " . format (
Frequency ) )

def dataStocks ( FrequencyAva : list , Frequency : str , Ticker : str ,

StartDate : str , EndDate : str ) :

’’’
Downloading the candlestick data into . csv files for stocks .

@param FrequencyAva : List of frequencies avaliable from

Yahoo Finance .
@param Frequency : User desired frequency .
@param Pair : Coin pair operated .
@param StartDate : Starting date of the downloaded data .
@param EndDate : Ending date of the downloaded data .

@return None
’’’

# Downloading in case the frequency exists

if checkFrequency ( FrequencyAva , Frequency ) :

# Downloading data from Yahoo Finance

Data = yf . get_data ( Ticker , start_date = StartDate , end_date
= EndDate , index_as_date
= True , interval =
Frequency )
Data [ ’ time ’] = Data . index
Data [ ’ time ’] = Data [ ’ time ’] . dt . strftime ( ’%Y - %m - % d % H : % M : % S ’
)

# Generating the directory of the data

if not os . path . isdir ( " MarketData / Stocks / { } / " . format ( Ticker )
):
os . mkdir ( " MarketData / Stocks / { } / " . format ( Ticker ) )

# Deleting the name in case it already exists

if os . path . exists ( " MarketData / Stocks / { } / Freq_ { } . csv " . format
( Ticker , Frequency ) ) :
os . remove ( " MarketData / Stocks / { } / Freq_ { } . csv " . format (
Ticker , Frequency ) )

# Generating . csv file from DataFrame

Data . to_csv ( " MarketData / Stocks / { } / Freq_ { } . csv " . format (
Ticker , Frequency ) )

else :
print ( " Error for the selected frequency ( { } ) \ n " . format (
Frequency ) )

José Manuel Rodrı́guez Lomeña 140

 A.7. STRATEGY

A.7 strategy
Code Listing A.7: strategy file code
# Python libraries
import numpy as np
from sklearn . preprocessing import MinMaxScaler
from matplotlib import pyplot as plt
from keras . models import Sequential
from keras . layers import Dense
from keras . layers import LSTM

# Imported files
from settings import PAIR , STOCK , ASSET , DATE , FREQUENCIES
import downloadData

# Frequency of the data

dataFrequency = [ ’ 15m ’]

class Strategy ( bt . Strategy ) :

def __init__ ( self ) :

# Downloading training data

self . trainingData = downloadData . dataCrypto ( FREQUENCIES ,
dataFrequency , PAIR , DATE
[ " StartDate " ] , DATE [ "
EndDate " ] )

# Data preprocessing
self . trainingSet = self . trainingData . values
self . trainingSet = np . reshape ( self . trainingSet , ( len ( self .
trainingSet ) , 1 ) )
self . trainingSet = MinMaxScaler () . fit_transform ( self .
trainingSet )
self . lengthTS = len ( self . trainingSet )
self . trainingX = self . trainingSet [ 0 : self . lengthTS - 1 ]
self . trainingX = np . reshape ( self . trainingX , ( len ( self .
trainingX ) , 1 , 1 ) )
self . trainingY = self . trainingSet [ 1 : self . lengthTS ]

# Initializing the RNN

self . RNN = Sequential ()

# Output layer is added

self . RNN . add ( Dense ( units = 1 ) )

# Input layer and LSTM layer are added

self . RNN . add ( LSTM ( units = 4 , activation = ’ sigmoid ’ ,
input_shape = ( None , 1 ) ) )

# Compiling the RNN

self . RNN . compile ( optimizer = ’ adam ’ , loss = ’
m ean _s qu ar ed _e rr or ’)

# Fitting the RNN to the training data set

141 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

self . RNN . fit ( self . trainingX , self . trainingY , batch_size = 5

, epochs = 100 )

# Creating trained model

self . trainedSet = self . trainingData . values
self . input = np . reshape ( self . trainedSet , len ( self .
trainedSet ) , 1 )
self . input = MinMaxScaler () . transform ( self . input )
self . input = np . reshape ( self . input , ( len ( self . input ) , 1 , 1 )
)
self . trainedModel = self . RNN . predict ( self . input )
self . trainedModel = MinMaxScaler () . i nvers e_tran sform ( self .
trainedModel )

A.8 backtestingCore
Code Listing A.8: backtestingCore file code
# BACKTESTING BOT CLASS FILE : functions defined to implement the
backtesting .

# Python libraries .
import pandas as pd
import math
import backtrader as bt
import backtrader . analyzers as btanalyzers
import matplotlib . pyplot as plt
from matplotlib . font_manager import FontProperties
import datetime as dt
import csv

# Imported files .
from settings import FREQUENCIES
import downloadData
import strategy

class BacktestingClass :

def __init__ ( self , assetType : int , pair : dict , stock : str , date
: dict , dataFrequency : list ) :

’’’
Constructor of the class : setting all the information for
the backtesting and
downloading the
subsequent candlestick
data ,
which is entered into the strategy class .

@param assetType : type of asset selected ( 0 for

cryptocurrency , 1 for
stocks ) .
@param pair : cryptocurrency pair selected ( must be in
Yahoo Finance list ) .

José Manuel Rodrı́guez Lomeña 142

 A.8. BACKTESTINGCORE

@param stock : stock selected ( must be in Yahoo Finance

list ) .
@param date : start and end dates selected ( year - month -
day ) .
@param dataFrequency : data frequencies selected .

’’’

self . pair = pair

self . stock = stock

self . dtStartDate = dt . datetime ( int ( date [ " StartDate " ] [ 0 : 4 ] ) ,

int ( date [ " StartDate " ] [ 5 :
7 ] ) , int ( date [ " StartDate "
] [ 8 : 10 ] ) )
self . dtEndDate = dt . datetime ( int ( date [ " EndDate " ] [ 0 : 4 ] ) , int
( date [ " EndDate " ] [ 5 : 7 ] ) ,
int ( date [ " EndDate " ] [ 8 : 10 ]
))
self . cerebro = bt . Cerebro ()

for freq in dataFrequency :

if assetType = = 0 :
# Downloading data for cryptocurrency
downloadData . dataCrypto ( FREQUENCIES , freq , self .
pair , date [ "
StartDate " ] , date
[ " EndDate " ] )

# Cryptocurrency data feeded into the backtesting

data = bt . feeds . Y ah o o Fi n a nc e C SV D a t a (
dataname = ( " MarketData / Crypto / { } { } / Freq_ { } . csv
" . format ( self
. pair [ " Base " ]
, self . pair [ "
Quote " ] , freq
)),
fromdate = self . dtStartDate ,
todate = self . dtEndDate ,
reverse = False )

elif assetType = = 1 :
# Downloading data for stocks
downloadData . dataStocks ( FREQUENCIES , freq , self .
stock , date [ "
StartDate " ] , date [
" EndDate " ] )

# Stocks data feeded into the backtesting

data = bt . feeds . Y ah o o Fi n a nc e C SV D a t a (
dataname = ( " MarketData / Stocks / { } / Freq_ { } . csv " .
format ( self .
stock , freq ) )
,

143 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

fromdate = self . dtStartDate ,

todate = self . dtEndDate ,
reverse = False )

self . cerebro . adddata ( data )

def setInitialMoney ( self , money ) :

’’’
Setting initial cash of the portfolio .

@param money : cash of the portfolio .

@return none
’’’

self . cerebro . broker . setcash ( money )

self . InitialMoney = money

def pri ntCurr entMon ey ( self ) :

’’’
Printing the current portfolio value .

@return none
’’’

print ( " Current Portfolio Value : % . 2f " % self . cerebro . broker

. getvalue () )

def setComissions ( self , comissions ) :

’’’
Setting Binance commissions .

@param comissions : broker comission for each trade .

@return none
’’’

self . cerebro . broker . setcommission ( commission = comissions )

def setSizers ( self , sizer ) :

’’’
Setting portfolio percentage for each trade .

@param sizer : sizers class .

@return none
’’’

self . cerebro . addsizer ( sizer )

def setStrategy ( self , strategy ) :

José Manuel Rodrı́guez Lomeña 144

 A.8. BACKTESTINGCORE

’’’
Setting backtesting strategy .

@param strategy : strategy class .

@return none
’’’

self . cerebro . addstrategy ( strategy )

# Run the implemented strategy

def runStrategy ( self ) :

’’’
Running the implemented class .

@return none
’’’

self . strats = self . cerebro . run ()

self . strat = self . strats [ 0 ]
self . results = self . strat . analyzers . ta . get_analysis ()

def p l o t B a c k t e s t i n g R e s u l t s ( self ) :

’’’
Plotting backtesting results : market analyzers and
backtesting plots .

@return none
’’’

self . p r i n t M a r k e t A n a l y z e r s ()
self . cerebro . plot ()

def setBotAnalyzers ( self ) :

’’’
Setting bot analyzers ( see documentation :
https :// www . backtrader . com / docu / analyzers / analyzers /)

@return none
’’’

self . cerebro . addanalyzer ( bt . analyzers . TradeAnalyzer , _name =

" ta " )

def p r i n t M a r k e t A n a l y z e r s ( self ) :

’’’
Printing market analyzers plot . If there are no trades ,
this function
warns the user .

145 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

@return none
’’’

results = self . strat . analyzers . ta . get_analysis ()

if results . total . total = = 0 :

print ( " There have been no trades for the current
backtesting . " )
return

else :
closedTrades = results . total . closed
wonTrades = results . won . total
winRatio = wonTrades / closedTrades
netProfit = results . pnl . net . total
totalReturn = ( netProfit / self . InitialMoney ) * 100

totalDays = ( self . dtEndDate - self . dtStartDate ) . days

monthlyReturn = totalReturn * 30 . 0 / totalDays

fig , ax = plt . subplots (1 , 1 )

column = [ " Results " ]

rows = [ " Closed trades " ,

" Won trades " ,
" Win ratio " ,
" Net profit " ,
" Total return " ,
" Start period " ,
" End period " ,
" Monthly return " ]

results = [ [ " { } " . format ( closedTrades ) ] ,

[ " { } " . format ( wonTrades ) ] ,
[ " { :. 2f } % " . format ( 100 * winRatio ) ] ,
[ " { :. 2f } { } " . format ( netProfit , self . pair [
" Quote "
])],
[ " { :. 2f } % " . format ( totalReturn ) ] ,
[ " { } " . format ( self . dtStartDate ) ] ,
[ " { } " . format ( self . dtEndDate ) ] ,
[ " { :. 2f } % " . format ( monthlyReturn ) ] ]

df = pd . DataFrame ( results , columns = column )

ax . axis ( ’ tight ’)
ax . axis ( ’ off ’)
ax . table = plt . table ( cellText = df . values ,
rowLabels = rows ,
colLabels = column ,
colWidths = None ,
loc = " center " )
ax . table . set_fontsize ( 20 )
ax . table . scale ( 1 .8 , 1 . 8 )

José Manuel Rodrı́guez Lomeña 146

 A.9. BACKTESTINGMAIN

for ( row , col ) , cell in ax . table . get_celld () . items () :

if ( row = = 0 ) or ( col = = - 1 ) :
cell . set_text_props ( fontproperties =
FontProperties
( weight = ’ bold
’) )

ax . table . a u t o _ s e t _ c o l u m n _ w i d t h ( col = list ( range ( len ( df .

columns ) ) ) )
plt . title ( " Backtesting Results " , fontsize = 30 ,
fontweight = " bold " )

A.9 backtestingMain
Code Listing A.9: backtestingMain file code
# BACKTESTING MAIN CODE : main file of the backtesting tool .

# Python libraries
import pandas as pd
import numpy as np
import csv
import datetime as dt
import math
import os
import random

# Python libraries options

pd . options . mode . ch ai ned _a ss ig nm en t = None # default = ’ warn ’

# Imported files
from settings import PAIR , STOCK , ASSET , DATE
import downloadData
import backtestingCore

# Import of strategy and sizers

import strategy as strat
import sizers

# Strategy is backtested .
backStrategy = backtestingCore . BacktestingClass ( ASSET [ " AssetType " ] ,
PAIR , STOCK [ " Ticker " ] , DATE ,
strat . dataFrequency )
# Bot_BackTest . setComissions ( Bot_1 . makerCommission )
backStrategy . setInitialMoney ( 1000 . 0 )
backStrategy . setSizers ( sizers . FullMoney )
backStrategy . setBotAnalyzers ()
backStrategy . setStrategy ( strat . Strategy )
backStrategy . runStrategy ()
backStrategy . p rintCu rrentM oney ()
backStrategy . p l o t B a c k t e s t i n g R e s u l t s ()

147 José Manuel Rodrı́guez Lomeña

APPENDIX A. BOT CODE

José Manuel Rodrı́guez Lomeña 148

Appendix B

Studies Codes

This appendix aims to state constancy of all the code developed in order to perform
the studies presented in Chap. 12. Please note that not all codes are written using
the same programming language. Some have been implemented in Python, while
others in Matlab, for convenience when coding.

B.1 Simulation on Random Trade Closures

Code Listing B.1: File code for the simulation on random trade closures
import pandas as pd
import random
import matplotlib . pyplot as plt
import numpy as np
import seaborn as sns
import combinatory

minWon = 48 # %
maxWon = 52 # %
if minWon > maxWon :
minWon , maxWon = maxWon , minWon

div = 100
quan = ( maxWon - minWon ) * div + 1
percWonTrades = [ ]
for i in range ( quan ) :
perc = minWon + i / div # %
percWonTrades . append ( perc )

init ialIn vestme nt = 15000 # u . m .

isComission = False
makerComission = 0 . 018 # %
maxTakeProfitPer = 10 # %

# LIMITS DEFINITION
maxPer = 6 # Maximum percentage
minPer = 5 # Minimum percentage
numDiv = 2 # Number of divisions in each % unit

149
APPENDIX B. STUDIES CODES

perBetTrade = [ ]
quantity = ( maxPer - minPer ) * numDiv

# In case of minimum percentage being null , the first trade

percentage must be 0 . 5 %
if minPer = = 0 :
for i in range ( quantity ) :
percentage = minPer + ( maxPer - minPer ) * ( i + 1 ) /
quantity
perBetTrade . append ( percentage )
else :
quantity2 = int ( quantity + 1 )
for i in range ( quantity2 ) :
percentage = minPer + ( maxPer - minPer ) * i / ( quantity2 -
1)
perBetTrade . append ( percentage )

numTrades = 40
numIt = 10
countPos = - 1

for botPerc in percWonTrades :

# Combination without repetition

r = int ( numTrades * botPerc / 100 )
numIt = math . factorial ( numTrades ) / ( math . factorial ( r ) * math .
factorial ( numTrades - r ) )
print ( len ( combinaciones ( comb , r ) )

for per in perBetTrade :

per = per / 100 # unitary percentage

for i in range ( numIt ) :

currentCapital = in itialI nvest ment
winCount = 0

for j in range ( numTrades ) :

if isComission :
betTrade = currentCapital * per * ( 1 -
makerComission
/ 100 )
else :
betTrade = currentCapital * per ;
winLossNum = random . uniform (0 , 100 )
if winLossNum > botPerc :
profitPer = - random . uniform (0 ,
maxTakeProfitPer
) / 100
else :
profitPer = random . uniform (0 , maxTakeProfitPer )
/ 100
winCount + = 1
if isComission :
finalTrade = betTrade * profitPer * ( 1 -
makerComission

José Manuel Rodrı́guez Lomeña 150

 B.1. SIMULATION ON RANDOM TRADE CLOSURES

/ 100 )
else :
finalTrade = betTrade * profitPer
currentCapital = currentCapital + finalTrade

ROI = ( currentCapital - initia lInves tment ) /

ini tialIn vestme nt *
100
winPer = winCount / numTrades * 100
countPos + = 1
results [ countPos ] = [ botPerc , per * 100 , round ( ROI , 2 ) ,
round ( winPer , 2 ) ]

Results = pd . DataFrame ( results , columns = [ " Set Win Perc " , " Bet
Trade Perc " , " ROI " , " Real Win
Perc " ] )
pd . set_option ( ’ display . max_rows ’ , None , ’ display . max_columns ’ , None
)

sns . set_theme ( style = " dark " )

sns . color_palette ( " viridis " , as_cmap = True )

figureResults , ax = plt . subplots ()

for botPerc in percWonTrades :

p er B e t Tr a d eA n a ly s i s = [ ]
ROIAnalysis = [ ]
ResultsFiltered = Results [ Results [ ’ Set Win Perc ’] = = botPerc ]

for per in perBetTrade :

ResultsFiltered2 = ResultsFiltered [ ResultsFiltered [ ’ Bet
Trade Perc ’] = = per ]
ROIMean = ResultsFiltered2 [ " ROI " ] . mean ()
p er B e t Tr a d eA n a ly s i s . append ( per )
ROIAnalysis . append ( ROIMean )

sns . lineplot ( perBetTradeAnalysis , ROIAnalysis , linewidth = 2 .0 ,

label = " { } " . format ( botPerc ) )

ax . set_title ( " ROI vs Bet Trade % " , size = 14 , fontweight = " bold " )
ax . set_ylabel ( " ROI ( % ) " , size = 12 , fontweight = ’ bold ’)
ax . set_xlabel ( " Bet Trade % " , size = 12 , fontweight = ’ bold ’)
ax . set_xlim ( minPer , maxPer )
ax . grid ()
ax . legend ( shadow = True , fontsize = 16 )

plt . show ()

151 José Manuel Rodrı́guez Lomeña

APPENDIX B. STUDIES CODES

B.2 Combination Without Repetition

Code Listing B.2: File code for computing the combination without repetition
# Computes a combination without repetition
def combWithoutRep (n , r ) :

c = [ i for i in range ( r ) ]
cIni = c . copy ()

comb = [ c . copy () ]
i = 0

while i < len ( c ) - 1 :

if c [ i ] = = c [ i + 1 ] - 1 :

i += 1

if i = = len ( c ) - 1 :

if c [ i ] < n - 1 :

c[i] += 1
i -= 1

while i > - 1 :

c [ i ] = cIni [ i ]
i -= 1

comb . append ( c . copy () )

i = 0
else :
c[i] += 1
i -= 1

while i > - 1 :

c [ i ] = cIni [ i ]
i -= 1

comb . append ( c . copy () )

i = 0

return comb

José Manuel Rodrı́guez Lomeña 152

 B.3. ROI VS NUMBER OF TRADES FOR ωp RANGE

B.3 ROI vs Number of Trades for ωp Range

Code Listing B.3: ROI vs Number of Trades for ωp Range
p = 0 . 0005 ;

A = 1;

wp = linspace ( 0 . 48 , 0 . 52 , 5 ) ;
wpLen = length ( wp ) ;

n = [ 1 : 40000 ] ;
nLen = length ( n ) ;

negLog = log ( 1 - A * p ) ;
posLog = log ( 1 + p ) ;

ROI = zeros ( wpLen , nLen ) ;

for j = 1 : wpLen

for i = 1 : nLen
res = n ( i ) * ( wp ( j ) * ( posLog - negLog ) + negLog ) ;

ROI (j , i ) = 10 ˆ res - 1 ;
end
i = 1;
end

plot (n , ROI (1 , : ) )
title ( ’ ROI vs number of trades [ A = 1 , p = 0 . 0005 ] ’)
xlabel ( ’ Number of trades ( n ) ’)
ylabel ( ’ ROI ’)
grid on
grid minor
hold on
plot (n , ROI (2 , : ) )
plot (n , ROI (3 , : ) )
plot (n , ROI (4 , : ) )
plot (n , ROI (5 , : ) )
legend ( ’ wp = 0 . 48 ’ , ’ wp = 0 . 49 ’ , ’ wp = 0 . 50 ’ , ’ wp = 0 . 51 ’ , ’ wp = 0 . 52 ’)
hold off

153 José Manuel Rodrı́guez Lomeña

APPENDIX B. STUDIES CODES

B.4 ROI vs Number of Trades for p Range

Code Listing B.4: ROI vs Number of Trades for p Range
p = [ 0 . 0005 , 0 . 0015 , 0 . 0025 , 0 . 005 ] ;
pLen = length ( p ) ;

A = 1;

wp = 0 . 5008 ;

n = [ 1 : 40000 ] ;
nLen = length ( n ) ;

ROI = zeros ( pLen , nLen ) ;

for j = 1 : pLen

negLog = log ( 1 - A * p ( j ) ) ;
posLog = log ( 1 + p ( j ) ) ;

for i = 1 : nLen
res = n ( i ) * ( wp * ( posLog - negLog ) + negLog ) ;

ROI (j , i ) = 10 ˆ res - 1 ;
end
i = 1;
end

plot (n , ROI (1 , : ) )
title ( ’ ROI vs number of trades [ A = 1 , w_p = 0 . 5008 ] ’)
xlabel ( ’ Number of trades ( n ) ’)
ylabel ( ’ ROI ’)
grid on
grid minor
hold on
plot (n , ROI (2 , : ) )
plot (n , ROI (3 , : ) )
plot (n , ROI (4 , : ) )
legend ( ’p = 0 . 0005 ’ , ’p = 0 . 0015 ’ , ’p = 0 . 0025 ’ , ’p = 0 . 0050 ’)
hold off

José Manuel Rodrı́guez Lomeña 154

 B.5. ROI VS NUMBER OF TRADES FOR A RANGE

B.5 ROI vs Number of Trades for A Range

Code Listing B.5: ROI vs Number of Trades for A Range
A = [ 0 . 98 , 0 . 99 , 1 , 1 . 01 , 1 . 02 ] ;
ALen = length ( A ) ;

p = 0 . 0005 ;

wp = 0 . 50 ;

n = [ 1 : 40000 ] ;
nLen = length ( n ) ;

ROI = zeros ( ALen , nLen ) ;

posLog = log ( 1 + p ) ;

for j = 1 : ALen

negLog = log ( 1 - A ( j ) * p ) ;

for i = 1 : nLen
res = n ( i ) * ( wp * ( posLog - negLog ) + negLog ) ;

ROI (j , i ) = 10 ˆ res - 1 ;
end
i = 1;
end

plot (n , ROI (1 , : ) )
title ( ’ ROI vs number of trades [ p = 0 . 0005 , w_p = 0 . 50 ] ’)
xlabel ( ’ Number of trades ( n ) ’)
ylabel ( ’ ROI ’)
grid on
grid minor
hold on
plot (n , ROI (2 , : ) )
plot (n , ROI (3 , : ) )
plot (n , ROI (4 , : ) )
plot (n , ROI (5 , : ) )
legend ( ’A = 0 . 98 ’ , ’A = 0 . 99 ’ , ’A = 1 ’ , ’A = 1 . 01 ’ , ’A = 1 . 02 ’)
hold off

155 José Manuel Rodrı́guez Lomeña

APPENDIX B. STUDIES CODES

José Manuel Rodrı́guez Lomeña 156

Bibliography

[1] 3 Commas - Crypto Trading Bot. Visited on December 18, 2022. url: https:
//3commas.io/.
[2] 7 Best Bitcoin Hardware Wallet 2022 — Reviews And Comparisons. Visited
on December 6, 2022. url: https://medium.com/coinmonks/the- best-
cryptocurrency-hardware-wallets-of-2020-e28b1c124069.
[3] Backtrader - Analyzers. Visited on October 16, 2022. url: https : / / www .
backtrader.com/docu/analyzers/analyzers/.
[4] Fernando Blanco, Helena Matute, and Miguel A Vadillo. “Making the uncon-
trollable seem controllable: The role of action in the illusion of control”. In:
Quarterly Journal of Experimental Psychology 64.7 (2011), pp. 1290–1304.
[5] Blockchains: The Technology of Transactions. Visited on December 4, 2022.
url: https://towardsdatascience.com/blockchains- the- technology-
of-transactions-9d40e8e41216.
[6] Jesús Bobadilla. Machine Learning y Deep Learning: Usando Python, Scikit y
Keras. Ediciones de la U, 2021.
[7] Matthew Browne and Matthew J Rockloff. “Measuring behavioural dependence
in gambling: A case for removing harmful consequences from the assessment of
problem gambling pathology”. In: Journal of Gambling Studies 36.4 (2020),
pp. 1027–1044.
[8] Capfolio. Visited on December 18, 2022. url: https://www.capfol.io/.
[9] CFI - Cryptocurrency Exchanges. Visited on December 9, 2022. url: https://
corporatefinanceinstitute.com/resources/cryptocurrency/cryptocurrency-
exchanges/.
[10] CNMC. “Informe de Garantı́as y Etiquetado de la Electricidad”. In: Cnmc
(2018), p. 29.
[11] Coinbase - What is ”proof of work” or ”proof of stake”? Visited on December
7, 2022. url: https://www.coinbase.com/es/learn/crypto-basics/what-
is-proof-of-work-or-proof-of-stake.
[12] Coinbase - What is a smart contract? Visited on December 8, 2022. url:
https://www.coinbase.com/es/learn/crypto-basics/what-is-a-smart-
contract.
[13] Coinbase - What is mining? Visited on December 7, 2022. url: https://www.
coinbase.com/es/learn/crypto-basics/what-is-mining.
[14] CoinmMarketCup. Visited on December 8, 2022. url: https://coinmarketcap.
com/.

157
BIBLIOGRAPHY

[15] crypto.com - Cryptography and Cryptocurrencies – Putting the Crypto into Cur-
rency. Visited on December 4, 2022. url: https://crypto.com/university/
what-is-cryptography.
[16] CryptoVantage - A Brief History of Cryptocurrency. Visited on December 7,
2022. url: https://www.cryptovantage.com/guides/a-brief-history-
of-cryptocurrency/.
[17] EQUITY TRUST - Types of Cryptocurrency Explained. Visited on December
8, 2022. url: https://www.trustetc.com/blog/cryptocurrency-types/.
[18] EUROMONEY LEARNING - How does a transaction get into the blockchain?
Visited on November 13, 2022. url: https://www.euromoney.com/learning/
blockchain-explained/how-transactions-get-into-the-blockchain.
[19] Fan Fang et al. “Cryptocurrency trading: a comprehensive survey”. In: Financial
Innovation 8.1 (2022), pp. 1–59.
[20] GitHub - blackbird. Visited on December 18, 2022. url: https://github.com/
butor/blackbird.
[21] GitHub - catalyst. Visited on December 18, 2022. url: https://github.com/
scrtlabs/catalyst.
[22] GitHub - ccxt. Visited on December 18, 2022. url: https://github.com/
ccxt/ccxt.
[23] GitHub - CryptoSignal. Visited on December 18, 2022. url: https://github.
com/CryptoSignal/crypto-signal.
[24] GitHub - freqtrade. Visited on December 18, 2022. url: https://github.com/
freqtrade/freqtrade.
[25] GitHub - golang-crypto-trading-bot. Visited on December 18, 2022. url: https:
//github.com/saniales/golang-crypto-trading-bot.
[26] GitHub - Krypto-trading-bot. Visited on December 18, 2022. url: https :
//github.com/ctubio/Krypto-trading-bot.
[27] GitHub - StockSharp. Visited on December 18, 2022. url: https://github.
com/StockSharp/StockSharp.
[28] Google Finance. Visited on December 8, 2022. url: https://www.google.
com/finance.
[29] Great Learning - Types of Neural Networks and Definition of Neural Network.
Visited on January 5, 2023. url: https://www.mygreatlearning.com/blog/
types-of-neural-networks.
[30] IBM - What are neural networks? Visited on January 4, 2023. url: https:
//www.ibm.com/topics/neural-networks.
[31] IBM - What are recurrent neural networks? Visited on January 5, 2023. url:
https://www.ibm.com/topics/recurrent-neural-networks.
[32] INFOSEC INSIGHTS - What Is Crypto Mining? How Cryptocurrency Mining
Works. Visited on December 7, 2022. url: https://sectigostore.com/blog/
what-is-crypto-mining-how-cryptocurrency-mining-works/.
[33] Investopedia - ARIMA Model: Autoregressive Integrated Moving Average. Vis-
ited on January 5, 2023. url: https://www.investopedia.com/terms/a/
autoregressive-integrated-moving-average-arima.asp.

José Manuel Rodrı́guez Lomeña 158

 BIBLIOGRAPHY

[34] Investopedia - Backtesting: Definition, How It Works, and Downsides. Visited

on December 24, 2022. url: https://www.investopedia.com/terms/b/
backtesting.asp.
[35] Investopedia - Binance Exchange. Visited on December 15, 2022. url: https:
//www.investopedia.com/terms/b/binance-exchange.asp.
[36] Investopedia - Blockchain Facts: Whay Is It, How It Works, and How It Can
Be Used. Visited on November 13, 2022. url: https://www.investopedia.
com/terms/b/blockchain.asp.
[37] Investopedia - How Does Bitcoin Mining Work? Visited on December 4, 2022.
url: https://www.investopedia.com/tech/how-does-bitcoin-mining-
work/.
[38] Investopedia - Non-Fungible Token (NFT): What It Means and How It Works.
Visited on December 8, 2022. url: https://www.investopedia.com/non-
fungible-tokens-nft-5115211.
[39] Investopedia - Stablecoins: Definition, How They Work, and Types. Visited
on December 8, 2022. url: https : / / www . investopedia . com / terms / s /
stablecoin.asp.
[40] Investopedia - Understanding Liquidity and How to Measure It. Visited on De-
cember 9, 2022. url: https://www.investopedia.com/terms/l/liquidity.
asp.
[41] Investopedia - What Does Proof-of-Stake (PoS) Mean in Crypto? Visited on
December 7, 2022. url: https://www.investopedia.com/terms/p/proof-
stake-pos.asp.
[42] Gary C Kessler. “An overview of cryptography”. In: (2003).
[43] Ahmed M Khedr et al. “Cryptocurrency price prediction using traditional
statistical and machine-learning techniques: A survey”. In: Intelligent Systems
in Accounting, Finance and Management 28.1 (2021), pp. 3–34.
[44] Ellen J Langer. “The illusion of control.” In: Journal of personality and social
psychology 32.2 (1975), p. 311.
[45] Dale T Miller and Brian R Taylor. “Counterfactual thought, regret, and super-
stition: How to avoid kicking yourself”. In: What might have been. Psychology
Press, 2014, pp. 317–344.
[46] John J Murphy. Technical analysis of the financial markets: A comprehensive
guide to trading methods and applications. Penguin, 1999.
[47] Satoshi Nakamoto. “Bitcoin: A peer-to-peer electronic cash system”. In: De-
centralized Business Review (2008), p. 21260.
[48] Python 3.11.1 documentation. Visited on December 27, 2022. url: https:
//docs.python.org/3/.
[49] Sangaku Maths - Combinations without repetition. Visited on January 2, 2023.
url: https : / / www . sangakoo . com / en / unit / combinations - without -
repetition.
[50] Sofi Learn - Guide to Crypto Staking: What It Is, How It Works, and How
to Get Started. Visited on December 7, 2022. url: https://www.sofi.com/
learn/content/crypto-staking/.

159 José Manuel Rodrı́guez Lomeña

BIBLIOGRAPHY

[51] SoFi Learn - Understanding the Different Types of Cryptocurrency. Visited

on December 8, 2022. url: https : / / www . sofi . com / learn / content /
understanding-the-different-types-of-cryptocurrency/.
[52] Saurabh Suratkar, Mahesh Shirole, and Sunil Bhirud. “Cryptocurrency wallet:
A review”. In: 2020 4th International Conference on Computer, Communication
and Signal Processing (ICCCSP). IEEE. 2020, pp. 1–7.
[53] techopedia - Constructor. Visited on December 29, 2022. url: https://www.
techopedia.com/definition/5656/constructor.
[54] The Balance - How Does Blockchain Work For Small Business. Visited on
December 4, 2022. url: https://www.thebalancemoney.com/blockchain-
explained-4773366.
[55] The Motley Fool - What Is Proof of Work (PoW) in Crypto? Visited on
December 7, 2022. url: https://www.fool.com/investing/stock-market/
market-sectors/financials/cryptocurrency-stocks/proof-of-work/.
[56] TikZ.net - Neural networks? Visited on January 4, 2023. url: https://tikz.
net/neural_networks/.
[57] Jennifer Williams. “The psychology of cryptocurrency trading: Risk and pro-
tective factors”. In: (2021).
[58] W Scott Wood and Maria M Clapham. “Development of the drake beliefs about
chance inventory”. In: Journal of gambling studies 21.4 (2005), pp. 411–430.
[59] yahoo! finance. Visited on December 24, 2022. url: https://finance.yahoo.
com/.

José Manuel Rodrı́guez Lomeña 160

 BIBLIOGRAPHY

161 José Manuel Rodrı́guez Lomeña