Advances in Economics, Business and Management Research, volume 189

Proceedings of the 2nd International Conference on Management, Economy and Law (ICMEL 2021)

Application of Bayesian Regression Model in

Financial Stock Market Forecasting
Xuejun Zhao1,*
1
College of Letter and Science, University of California, Santa Barbara, CA 93106, USA
*
Corresponding author. Email: [email protected]

ABSTRACT
The Bayesian method is a statistics field targeting the Bayes theorem in interpreting probabilities. The Bayesian
formula provides an insight into conditional probability based on present data and prior information. Due to the
efficiency of the Bayesian model in predicting future outcomes, the model is integrated with regression analysis
which is a set of statistical methods utilized for estimating relationships between dependent and independent
variables. Bayesian regression analysis is a reliable model for investigating variables having a significant impact
on the output of a particular process, such as financial stock market forecasting considered in this research. To
fulfill the study's aim, the research adopts secondary research on published journals, case studies, and reports
documented by scholars in the field. Due to the stochastic nature of stock market variables, inadequate data or
poorly dispersed data can be addressed using Bayesian linear regression allowing investors to make better
decisions and cut larger profit margins. The vector autoregression and the classical frequentist approach achieve
a higher probability accuracy than non-Bayesian methods such as the Auto-Regressive and Moving Average
Model time series models. The author found that by studying the vector Bayesian autoregressive prediction
model, it is possible to analyze how investors use the Bayesian model to predict stock market volatility.

Keywords: Bayes’ theorem, ARMA model, Regression analysis, Frequentist approach, Financial stock
market.

1. INTRODUCTION dependent variable based on the independent

factors. Regression analysis might be superficial,
A country's economic and social structure is multiple linear, nonlinear, or Bayesian[1]. If a
heavily reliant on the stock market. The huge regression modeling approach utilizes the Bayes’
demand and complexity of stock market forecasts theorem, the process yields the Bayesian linear
are determined by factors such as the input of stock regression, which includes a prior likelihood
price time series, stocks, raw materials, non- distribution and a posterior distribution[2]. The
parameters, and dynamics. A stock market is an Bayesian regression model calculates parameter
open market where company stock is bought and values using the posterior distribution multiplied by
sold at an agreed-upon price. A stock exchange is the prior distribution. Under the standard
one of the major stock market components. A stock distribution error assumption, a normally
exchange is an organization that gives trading distributed error in a linear regression model
facilities for stock traders and brokers to trade employs the Ordinary Least Squares estimation.
equities. The author will analyze how investors The Bayes theorem was first announced nearly
predict stock market volatility by the Statistics a year after the demise of its inventor Thomas
method set. Bayes in 1764[3]. Before 1764, Bayesian analysis
Regression analysis is a statistical approach had a tumultuous history due to the limited
used to show how a variable of interest relates to a applicability of the method and the lack of
collection of already known variables. The computing devices. Due to limited computational
objective is to construct a model that helps ability, Bayes' models failed due to the technique’s
statisticians explain, control, and forecast the incapacity in dealing with prior probabilities

Copyright © 2021 The Authors. Published by Atlantis Press International B.V.

This is an open access article distributed under the CC BY-NC 4.0 license -http://creativecommons.org/licenses/by-nc/4.0/. 140
 Advances in Economics, Business and Management Research, volume 189

delaying its appreciation towards the later end of family of distributions are the only ones that have a
the 19th century[4]. As the computation devices conjugate prior. A family with many joint
continued to advance within the century, Bayesian distributions is known as the exponential family,
analysis garnered more physics, statistics, and other especially if the probability distribution function
fields. However, Bayesian achieved its full glory in has an exponential term, but Conjugate prior does
the late 20th century when digital computers with not exist for Poisson’s distribution[2]. Conjugate
decent processing specifications became widely priors for Negative Binomial distributions exist if
available for scholars[2]. Many advances in the the parameter of a Poisson distribution follows a
Bayesian method have followed from the Gamma distribution[6]. The prior and the
development of global interest in Bayesian statistics. probability product is the exponential of the sum of
The application of probability to statistical issues is the two parameters and is calculated after the
made easier using Bayesian statistics. It allows posterior distribution parameter is found.
scholars to investigate the impact of several factors
Probability distributions are inclusive for linear
on the dependent variable. Since Bayes' theorem
regression in the Bayesian model. The response
defines the conditional probability of an event,
(dependent variable) is not expected to be chosen
depending on facts and prior information, while
from a single probability distribution, but instead,
regression analysis is a statistical method for
the distribution itself is used to approximate the
estimating relationships between variables, the
response[2]. The Bayesian Linear Regression
Bayesian regression analysis model can be an
model, in which the response variable is drawn
effective tool in forecasting the stock market. The
from a normal distribution, is represented by,
model is essential for investors due to the
stochasticity of stock market variables. y~N(    ,  2l )
Additionally, the best-known and most extensively Y (response) is generated from a normal
utilized Bayesian economic forecasting model is distribution, with a mean of zero and a variance
the vector autoregression model (VAR). Vector equal to The inverse of the weight matrix
autoregression models are built on the concept of multiplied by the predictor matrix transposes the
replication which is a recurring theme in these matrix of inverse weights hence the linear
models. This article will study the significance of regression[6]. Since the Bayesian regression model
the Bayesian regression model for a financial parameters are distributed, posterior probabilities
forecasting model in the stock market. The author for model parameters are computed conditionally
will analyze how investors use the Bayesian model from the training inputs and outputs.
to predict stock market volatility by studying the
vector Bayesian autoregressive prediction model. [ P(  | y, X ) * P（ | X）
]
P(y|β, X)=
[ P( y | X )]
2. LITERATURE REVIEW
The posterior probability distribution of model
Using a prior distribution is the most parameters is expressed by the above formula. The
contentious feature in Bayesian regression analysis. parameter (P(y|β, X)) is multiplied by the prior
Scholars utilize these prior distributions differently, probability of the model (P(y|β, X)) and then
either via the Conjugate or the Noninformative divided by a normalization constant (P(y|β, X)).
prior. The Conjugate prior is when the posterior
Liliihood * Prior
distribution and the initial distribution have the Posterior =
same form, while the Noninformative is when they Normalization
differ[5]. The Noninformative Prior distribution is In frequentist models, the domain knowledge is
utilized when scholars have limited previous assumed, while in the Bayesian model’s previous
knowledge or information. If an analyst can acquire data is included. Before conducting quantitative
adequate posterior distribution information or estimation, Scholars apply noninformative priors,
deduce accurate predictions, they use the Conjugate such as a normal distribution on the parameters.
Priors, an example of conjugate predicates[6]. However, the posterior distribution of model
Hence, if the posterior is conjugate to the parameters based on prior data is generated via
probability, the prior is referred to as a conjugate Bayesian Linear Regression. The quantification of
prior. When a conjugate prior already exists, the uncertainties in the Bayesian regression model
posterior distribution function can be found, and the facilitates accurate estimation of the model's
parameters can be calculated using the posterior distribution utilizing fewer data points.
distribution. Loss functions in the exponential

141
 Advances in Economics, Business and Management Research, volume 189

3. METHODOLOGY The classical (frequentist) approach looks at the

likelihood of an event given a long-term frequency.
The data collection phase involves secondary Stock market factors have random sampling
research on published research journals from 2010 attributes from the underlying variable population,
to 2021 on the Bayesian regression model in with a real magnitude that is randomly
forecasting problems. The author searches and finds distributed[9]. Inference rules incorporate universal
Bayesian method models, sorts out the results constraints based on unbiased estimators,
obtained, and leaves ten research papers that are hypothesis testing, and confidence intervals, relying
suitable for financial forecasting using Bayesian on large-sample approximations[5]. However, it is
regression models. The author also excluded infrequent for financial data to adhere to these
duplicate papers, papers involving imprecise topics, assumptions. For example, periods of economic
papers not related to financial forecasts, and papers crisis such as the Great Depression do not reoccur
covering imprecise topics. often enough to provide adequate samples of time-
series data. In addition, volatility and correlation
4. FINDINGS between various stock markets could either
appreciate or depreciate in a crisis adding an
Inadequate data or poorly dispersed data can be element of stochasticity. A common assumption in
addressed using Bayesian linear regression. The these cases is that the underlying parameters are
statistical model allows a stock market analyst to random variables[8]. It is crucial to bear in mind
attach prior information to the coefficients and that many of these strong assumptions are
noise in the absence of data, allowing the deduced completely irrelevant in a Bayesian context. In
priors to take over when the data is Bayesian statistics, the dataset is considered fixed,
missing[7][8][9]. With a Bayesian linear regression, and the parameters are considered uncertain[5].
investors may use this tool to find variables with Hence, probability distributions can be altered
high confidence and others with low when new information becomes available under the
confidence[10]. The linear relation and the Bayesian framework.
conviction on the relation are calculated utilizing
Bayes’ theorem[7][6]. The entire posterior The vector autoregression model use has
distribution on the Bayesian regression model expanded since it improves predictions and can be
comprises the variables stock market analysts are utilized to convey uncertainty[8]. The model is
uncertain about. The posterior distribution involves used for various purposes, including economic
the current noise estimate and its probability forecasting whereby Bayesian inference is
distribution. commonly employed. Econometricians have
discovered that non-Bayesian optimization is, in
5. DISCUSSION fact, more accessible to solve than the Bayesian
integration issue[6]. There is an essential link
Sequential ordering and correlations frequently between inference, computing, and models with
manifest in time series data analysis in stock market practical applicability. By comparison, the
securities. The pattern in which sequential data significant advancements in posterior simulators in
evolves as the buying and selling processes the past two decades have a notable impact on
progress involves time series analysis[3]. While developing new models[5]. Since these models'
classical statistics do not apply in many non- applicability depends on available computational
standard cases, Bayesian inference offers tools, powerful processors facilitate the making of
substantial advantages over such situations[7]. The influential previous assumptions, which are
ARMA time series models used in financial successful in compensating for a wasteful
markets are highly unconventional even if analysts parameterization.
preserve normalcy assumptions. For instance,
The market structure shifts over time, and
financial analysts rely on an array of asymptotic
occasionally, structural change happens suddenly.
behaviors, consistency, asymptotic normalcy, and
To successfully manage and reconcile the concept
efficiency in their ARMA models to maintain
of probability, it is imperative to use dynamic
normalcy[8]. Bayesian statistics are not
Bayesian modeling processes[5]. These methods
significantly affected by small sample size because
control the process of updating expectations while
Bayesian regression models achieve a higher
incorporating new information with careful respect
efficacy than standard time-series techniques.
to fundamental laws of probability[1]. Bayesian
forecasting requires cautious attention to

142
 Advances in Economics, Business and Management Research, volume 189

parameters, variables, and modeling structure. The expanded since it improves predictions and can be
Fundamental forecasting model incorporates three utilized to convey uncertainty.
market-related elements, including the shape of the
yield curve, the performance of stocks against AUTHORS’ CONTRIBUTIONS
bonds, and the central bank's current monetary
policies[8]. By utilizing forecasting models, excess This paper is independently completed by
return estimates, forecasting error volatility, and the Xuejun Zhao.
correlation structure, the variables are fed into the
modified mean-variance optimization system to ACKNOWLEDGMENTS
generate Fundamental Currency and Fundamental
Global Macro portfolios[6]. The technical Upon the completion of the paper, I would like
forecasting model employs three market-related to express my special thanks to my Thesis Advisor.
parameters and two technical indicators derived During the process of my work, I have received her
from two different daily moving averages to careful and careful help and instruction in the topic
provide trading indications from historical price selection and conception, as well as in the research
data. Excess return forecasting errors, forecasting methods and finalization of the paper.
error volatility, and the correlation structure of the
errors are utilized as inputs to the modified mean- REFERENCES
variance optimization system to create stock market
[1] Permai, S. D., & Tanty, H. (2018). Linear
securities portfolios[2]. These portfolios act as a
regression model using the Bayesian approach
guidebook for investors trading in various
for energy performance of the residential
investment instruments such as equities,
government bonds, commodities, and currencies to building. Procedia Computer Science, 135(1),
realize higher returns. pp. 671-677. Retrieved from
https://doi.org/10.1016/j.procs.2018.08.219
6. CONCLUSION [2] DelSole, T. (2007). A Bayesian Framework
for Multimodel Regression. Journal of Climate,
Chaotic Researching stock market prediction is
2810-2826. Retrieved from
an essential task for financial researchers since
https://doi.org/10.1175/JCLI4179.1
investing in the stock market has a higher risk.
While this is feasible because of recent advances in [3] Beck, K., Niendorf, B., & Peterson, P. (2012).
artificial intelligence, the majority of the danger has The use of Bayesian methods in economic
been reduced. Forecasts are created recursively by research. Investment Management and
way of a VAR. Forecasts are generated for each Financial Innovations, 68-75. Retrieved from
variable used in the system. While there are various https://www.businessperspectives.org/images/
situations in which VARs can be used, forecasting pdf/applications/publishing/templates/article/a
related variables where a straightforward ssets/4750/imfi_en_2012_03_Beck.pdf
interpretation is not required is just one of them.
Additionally, impulse response analysis is used to [4] Andernach, I. E., Hunewald, O. E., & Muller,
determine whether one variable can be used to C. P. (2013). Bayesian Inference of the
forecast another, and forecast error variance Evolution of HBV/E. PloS ONE, 8(11), 1-12.
decomposition is used to uncover the variance of a Retrieved from
forecast's variables due to the variability of the https://doi.org/10.1371/journal.pone.0081690
other variables. Fluctuation in the financial markets [5] Chan, J. C. (2019). Asymmetric Conjugate
affects asset pricing, risk management, portfolio
Priors for LargeBayesian VARs. Purdue
management, and the overall economy. It is
University. Retrieved from
consequently critical to better understand the
https://economics.indiana.edu/documents/asy
driving forces behind financial market volatility.
mmetric-conjugate-priors-for-large-bayesian-
Studying the vector Bayesian autoregressive
vars.pdf
prediction model, it is possible to analyze how
investors use the Bayesian model to predict stock [6] Dey, S., & Dey, T. (2012). Bayesian
market volatility. The vector autoregression models estimation and prediction intervals for a
are built on the concept of replication which is a Rayleigh distribution under a conjugate prior.
recurring theme in these models, and its use has Journal of Statistical Computation and
Simulation, 82(11), 1651-1660. Retrieved

143
 Advances in Economics, Business and Management Research, volume 189

from
https://doi.org/10.1080/00949655.2011.59080
8
[7] Huang, W., & Chang, M.-S. (2021, February
03). Gold and Government Bonds as Safe-
Haven Assets Against Stock Market
Turbulence in China. Macroeconomics and
Monetary Economics, 11(1), pp. 1-12.
Retrieved from
https://doi.org/10.1177/2158244021990655
[8] Ulf, H., & Raimond, M. (2006). Portfolio
Choice and Estimation Risk. A Comparison of
Bayesian to Heuristic Approaches. ASTIN
Bulletin: The Journal of the IAA, 36(1), 135-
160. Retrieved from
https://doi.org/10.1017/S0515036100014434
[9] Nonejad, N. (2017). Modeling and forecasting
aggregate stock market volatility in unstable
environments using mixture innovation
regressions. Journal of Forecasting, 36(6),
718-740. Retrieved from
https://doi.org/10.1002/for.2466
[10] Ouysse, R., & Kohn, R. (2010, December 01).
Bayesian variable selection and model
averaging in the arbitrage pricing theory
model. Computational Statistics and Data
Analysis, 54(12), pp. 3249-3268. Retrieved
from
https://doi.org/10.1016/j.csda.2009.09.034

144