ARCH GARCH Models

Let be the log return of an asset at time . We are going look at the study of
volatility, the prime idea being that the series is either serially uncorrelated or with
minor lower order serial correlations, but it is a dependent series. Examine the ACF
for the residuals and squared residuals for the calamari catch data (Figures 1,2). The
catch data had a definitive seasonality, which was removed. Then the remaining
series was modelled with an AR(5) model, so it is the residuals of this modelling that
I refer to. There are various definitions of what constitutes weak dependence of a
time series. However, the operational definition of independence here will be that
both the autocorrelation functions of the series and the squared series show no
autocorrelation. If there is no serial correlation of the series but there is of the squared
series, then we will say there is weak dependence. This will lead us to examine the
volatility of the series, since that is exemplified by the squared terms. Figures 3,4,5
show the situation for the Intel series mentioned on Page 99 in the text.

Autocorrelation Function for RESI1

(with 5% significance limits for the autocorrelations)

1.0
0.8
0.6
0.4
Autocorrelation

0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Lag

Figure 1: Residuals after AR(5) fitted to the deseasoned calamari data

 Autocorrelation Function for C3
(with 5% significance limits for the autocorrelations)

1.0
0.8
0.6
0.4
Autocorrelation

0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Lag

Figure 2: Autocorrelation of the squared residuals

Autocorrelation Function for C1

(with 5% significance limits for the autocorrelations)

1.0
0.8
0.6
0.4
Autocorrelation

0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 5 10 15 20 25 30 35 40 45 50 55 60
Lag

Figure 3: Autocorrelation for the log returns for the Intel series
 Autocorrelation Function for C2
(with 5% significance limits for the autocorrelations)

1.0
0.8
0.6
0.4
Autocorrelation

0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 5 10 15 20 25 30 35 40 45 50 55 60
Lag

Figure 4: ACF of the squared returns

Partial Autocorrelation Function for C2

(with 5% significance limits for the partial autocorrelations)

1.0
0.8
0.6
Partial Autocorrelation

0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 5 10 15 20 25 30 35 40 45 50 55 60
Lag

Figure 5: PACF for squared returns

Combining these three plots, it appears that this series is serially uncorrelated but
dependent. Volatility models attempt to capture such dependence in the return series.
Model Building

Building a volatility model for an asset return involves four steps:

1. Specify a mean equation by testing for serial dependence in the data, and if
necessary, building an ARMA model for the series;

(1)

2. Use the residuals of the mean equation to test for ARCH effects;

The sample ACF and PACF of the squared returns in Figures 4,5 show the
existence of conditional heteroscedasticity. This phenomenon often appears in
terms of volatility clustering, high for some periods and low for others. A
Ljung-Box test (see page 27) can confirm this. The sample ACF and PACF
infer that we may have an AR(3) model for the volatility.

3. So, we fit the model

(2)

After we fit it using the standard procedures in Minitab, we obtain

(3)
4. Check the fit of the model.
 Autocorrelation Function for RESI1
(with 5% significance limits for the autocorrelations)

1.0
0.8
0.6
0.4
Autocorrelation

0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 5 10 15 20 25 30 35 40 45 50 55 60
Lag

Figure 6: ACF of the residuals for the volatility model

As we can see from Figure 6, and the Ljung-Box statistics,

Modified Box-Pierce (Ljung-Box) Chi-Square statistic

Lag 12 24 36 48
Chi-Square 11.5 37.3 39.0 41.8
DF 8 20 32 44
P-Value 0.174 0.011 0.185 0.565

there are some significant residual lags.

Properties of ARCH Models

See Section 3.4.1.

Weaknesses of ARCH Models

1. The model assumes that positive and negative shocks have the same effects on
volatility because it depends on the square of the previous shocks. This is not
reasonable.
2. For an ARCH(1) model, must be in the interval This will restrict the
ability to deal with leptokurtic series.
3. It often requires many parameters to describe the volatility process of an asset
return.

Example:
Monthly excess returns of the S%P 500 Index (p. 116).

This series has 792 observations from 1926. See the text for the reasons why we fit
an AR(3) model to the series to get

If we look at the ACF and PACF of the squared residuals , we get Figures 7,8.

Autocorrelation Function for C5

(with 5% significance limits for the autocorrelations)

1.0
0.8
0.6
0.4
Autocorrelation

0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 5 10 15 20 25 30 35 40 45 50 55 60 65 70
Lag

Figure 7: ACF for squared residuals for S&P 500 excess returns
 Partial Autocorrelation Function for C5
(with 5% significance limits for the partial autocorrelations)

1.0
0.8
0.6
Partial Autocorrelation

0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0

1 5 10 15 20 25 30 35 40 45 50 55 60 65 70
Lag

Figure 8: PACF for squared residuals for S&P 500 excess returns

The “best” model for this series is an AR(9) – this has to be fitted on SPSS. To fit a
more parsimonious model, we turn to the generalised ARCG model (GARCH).

The GARCH Model

Instead of a high order AR(p) model, we fit a so-called GARCH model to the residual
series.

, where , iid.

This is called a model.

Estimating a GARCH Model

We describe the joint estimation of the AR(3)-GARCH(1,1) model for the S&P 500.
We have already estimated the AR(3) model for the return series . We now take
as an observed series and model it as an ARMA(p,q) series, with parameter
estimates . The GARCH estimates are then

In this case we fit an ARMA(1,1). We get

.
From this, we get

and