DataCamp GARCH Models in R

GARCH MODELS IN R

There are old traders and

there are bold traders, but...

Kris Boudt
Professor of finance and econometrics
 DataCamp GARCH Models in R

About the instructor

Kris Boudt
PhD in financial risk forecasting
Use GARCH models to win by not
losing (much)
R package rugarch of Alexios
Ghalanos.
 DataCamp GARCH Models in R

Calculating returns
Relative financial gains and losses, expressed in terms of returns

Function CalculateReturns in PerformanceAnalytics

# Example in R for daily S&P 500 prices (xts object)
library(PerformanceAnalytics)
SP500returns <- CalculateReturns(SP500prices)
 DataCamp GARCH Models in R

Daily S&P 500 returns

Properties of daily returns:

The average return is zero

Return variability changes through
time

Standard deviation = measure of return

variability.

Synonym: Return volatility.

Greek letter σt .
DataCamp GARCH Models in R
 DataCamp GARCH Models in R

How to estimate return volatility

Function sd() computes the standard deviation:
# Compute daily standard deviation
> sd(sp500ret)
[1] 0.01099357

Corresponding formula for T daily returns:


 1
^=
T
∑(Rt − μ
⎷T − 1
σ ^ )2 ,
t=1

where μ
^ is the mean return.
 DataCamp GARCH Models in R

Annualized volatility
sd(sp500ret) is daily volatility

Annualized volatility = √252 × daily volatility

# Compute annualized standard deviation
> sqrt(252)*sd(sp500ret)
[1] 0.1745175
DataCamp GARCH Models in R
 DataCamp GARCH Models in R

Rolling volatility estimation

Rolling estimation windows :

Window width? Multiple of 22 (trading days).

 DataCamp GARCH Models in R

Function chart.RollingPerformance()
library(PerformanceAnalytics)

chart.RollingPerformance(R = sp500ret ,
width = 22,
FUN = "sd.annualized",
scale = 252,
main = "Rolling 1 month volatility")
DataCamp GARCH Models in R
 DataCamp GARCH Models in R

About GARCH models in R

Estimation of σt requires time series models, like GARCH.
DataCamp GARCH Models in R

GARCH MODELS IN R

Let's refresh the basics of

computing rolling standard
deviations in R
 DataCamp GARCH Models in R

GARCH MODELS IN R

GARCH models: The way

forward

Kris Boudt
Professor of finance and econometrics
 DataCamp GARCH Models in R

Inventors of GARCH models

Robert Engle Tim Bollerslev

 DataCamp GARCH Models in R

Notation (i)
Input: Time series of returns
 DataCamp GARCH Models in R

Notation (ii)
At time t-1, you make the prediction about the the future return Rt , using the
information set available at time t − 1:
 DataCamp GARCH Models in R

Notation (iii)
Predicting the mean return: what is the best possible prediction of the actual
return?
 DataCamp GARCH Models in R

Notation (iv)
We then predict the variance: how far off the return can be from its mean?
 DataCamp GARCH Models in R

From theory to practice: Models for the mean

We need an equation that maps the past returns into a prediction of the mean

For AR(MA) models for the mean, see Datacamp course on time series analysis.
 DataCamp GARCH Models in R

From theory to practice: Models for the variance

We need an equation that maps the past returns into predictions of the variance
 DataCamp GARCH Models in R

ARCH(p) model: Autoregressive Conditional Heteroscedasticity

We need an equation that maps the past returns into predictions of the variance
 DataCamp GARCH Models in R

GARCH(1,1) model: Generalized ARCH

We need an equation that maps the past returns into predictions of the variance
 DataCamp GARCH Models in R

Parameter restrictions

To make the GARCH process realistic, we need that:

1. ω, α and β are > 0: this ensures that σt2 > 0 at all times.

2. α + β < 1: this ensures that the predicted variance σt2 always returns to the
long run variance:

The variance is therefore "mean-reverting"

The long run variance equals ω
1−α−β
 DataCamp GARCH Models in R

R implementation - Specify the inputs

Let's familiarize ourselves with the GARCH equations using R code:

# Set parameter values

alpha <- 0.1
beta <- 0.8
omega <- var(sp500ret)*(1-alpha-beta)
# Then: var(sp500ret) = omega/(1-alpha-beta)

# Set series of prediction error

e <- sp500ret - mean(sp500ret) # Constant mean
e2 <- e^2
 DataCamp GARCH Models in R

R implementation - compute predicted variances

# We predict for each observation its variance.
nobs <- length(sp500ret)
predvar <- rep(NA, nobs)

# Initialize the process at the sample variance

predvar[1] <- var(sp500ret)

# Loop starting at 2 because of the lagged predictor

for (t in 2:nobs){
# GARCH(1,1) equation
predvar[t] <- omega + alpha * e2[t - 1] + beta * predvar[t-1]
}
 DataCamp GARCH Models in R

R implementation - Plot of GARCH volatilities

# Volatility is sqrt of predicted variance
predvol <- sqrt(predvar)
predvol <- xts(predvol, order.by = time(sp500ret))

# We compare with the unconditional volatility

uncvol <- sqrt(omega / (1 - alpha-beta))
uncvol <- xts(rep(uncvol, nobs), order.by = time(sp500ret))

# Plot
plot(predvol)
lines(uncvol, col = "red", lwd = 2)
DataCamp GARCH Models in R
DataCamp GARCH Models in R

GARCH MODELS IN R

Let's practice!
 DataCamp GARCH Models in R

GARCH MODELS IN R

Alpha - Beta - Sigma:

The rugarch package

Kris Boudt
Professor of finance and econometrics
 DataCamp GARCH Models in R

The normal GARCH(1,1) model with constant mean

The normal GARCH model

Four parameters: μ, ω, α, β.
Estimation by maximum likelihood: find the parameter values for which the
GARCH model is most likely to have generated the observed return series.
 DataCamp GARCH Models in R

Alexios Ghalanos
library(rugarch)
citation("rugarch")

When using rugarch in publications, please cite:

To cite the rugarch package, please use:

Alexios Ghalanos (2018). rugarch: Univariate GARCH models. R package version 1.4
 DataCamp GARCH Models in R

Workflow

Three steps:

ugarchspec(): Specify which GARCH model you want to use (mean μ t ,

variance σt2 , distribution of et )

ugarchfit(): Estimate the GARCH model on your time series with returns
R1 , ..., RT .

ugarchforecast(): Use the estimated GARCH model to make volatility

predictions for RT+1 ,...
 DataCamp GARCH Models in R

Workflow in R
ugarchspec(): Specify which GARCH model you want to use.
# Constant mean, standard garch(1,1) model
garchspec <- ugarchspec(
mean.model = list(armaOrder = c(0,0)),
variance.model = list(model = "sGARCH"),
distribution.model = "norm")

ugarchfit(): Estimate the GARCH model

garchfit <- ugarchfit(data = sp500ret ,
spec = garchspec)

ugarchforecast(): Forecast the volatility of the future returns

garchforecast <- ugarchforecast(fitORspec = garchfit,
n.ahead = 5)
 DataCamp GARCH Models in R

ugarchfit object
The ugarchfit yields an object that contains all the results related to the

estimation of the garch model.

Methods coef, uncvar, fitted and sigma:
# Coefficients
garchcoef <- coef(garchfit)

# Unconditional variance
garchuncvar <- uncvariance(garchfit)

# Predicted mean
garchmean <- fitted(garchfit)

# Predicted volatilities
garchvol <- sigma(garchfit)
 DataCamp GARCH Models in R

Estimated GARCH coefficients for daily S&P 500 returns

print(garchcoef)

mu omega alpha1 beta1

5.728020e-04 1.220515e-06 7.792031e-02 9.111455e-01

Estimated model:

sqrt(garchuncvar)

0.01056519
 DataCamp GARCH Models in R

Estimated volatilities
garchvol <- sigma(garchfit)

plot(garchvol)
 DataCamp GARCH Models in R

What about future volatility?

tail(garchvol, 1)

2017-12-29 0.004862908

What about the volatility for the days following the end of the time series?
 DataCamp GARCH Models in R

Forecasting h-day ahead volatilities

Applying the sigma() method to the ugarchforecast object gives the volatility

forecasts:
sigma(garchforecast)

2017-12-29
T+1 0.005034754
T+2 0.005127582
T+3 0.005217770
T+4 0.005305465
T+5 0.005390797
 DataCamp GARCH Models in R

Forecasting h-day ahead volatilities

Applying the fitted() method to the ugarchforecast object gives the mean
forecasts:

fitted(garchforecast)

2017-12-29
T+1 0.000572802
T+2 0.000572802
T+3 0.000572802
T+4 0.000572802
T+5 0.000572802
 DataCamp GARCH Models in R

Application to tactical asset allocation

A portfolio that invests a percentage w in a risky asset (with volatility σt ) and
keeps 1 − w on a risk-free bank deposit account has volatility equal to

σp = wσt .

How to set w? One approach is volatility targeting: w is such that the predicted
annualized portfolio volatility equals a target level, say 5%. Then:

w∗ = 0.05/σt

Since GARCH volatilities change, the optimal weight changes as well.

DataCamp GARCH Models in R

GARCH MODELS IN R

Let's play with rugarch!