Scribd
Upload
108 views4 pages

Financial Econometrics Notes

This document discusses modeling and forecasting volatility in daily exchange rate data using autoregressive conditional heteroscedasticity (ARCH) and generalized ARCH (GARCH) models in PcGive. It outlines exercises to identify an appropriate AR model for each exchange rate series, test for ARCH effects, estimate ARCH and GARCH models, and use the preferred GARCH model to produce forecasts and compare to ARMA forecasts. The goal is to select the best volatility model to analyze and predict fluctuations in daily currency exchange rates.

Uploaded by

Gil Emerson
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
108 views4 pages

Financial Econometrics Notes

This document discusses modeling and forecasting volatility in daily exchange rate data using autoregressive conditional heteroscedasticity (ARCH) and generalized ARCH (GARCH) models in PcGive. It outlines exercises to identify an appropriate AR model for each exchange rate series, test for ARCH effects, estimate ARCH and GARCH models, and use the preferred GARCH model to produce forecasts and compare to ARMA forecasts. The goal is to select the best volatility model to analyze and predict fluctuations in daily currency exchange rates.

Uploaded by

Gil Emerson
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

ECON41615 Econometrics II

Volatility Models with PCGIVE

March, 2011

Purpose of this session:

This workshop covers ARCH and GARCH models, as well as forecasting by using the
GARCH model and the ARMA model. You will see how to estimate the various models
and perform appropriate diagnostic testing to assist in choosing a preferred model.

We will work with a dataset comprising exchange rate data for the period 1 January 1980
to 21 May 19871 . The contents of this dataset are listed in Table 1.

Table 1: Data contained in GARCH.XLS


DAY Day of week (1 is Monday)
BP US$ British Pound
CD US$ Canadian Dollar
DM US$ Deutsche Mark
DY US$ Japanese Yen
SF US$ Swiss Franc

and DUM is a dummy variable taking the value 1 is the previous day was a trading day,
and 0 if the previous day was a weekend or holiday.

Exercise 1: Volatility in Daily Exchange Rates - ARCH


Models
Download the data from duo and load into GiveWin.

dlog all of the exchange rate series, multiplying by 100 so we have daily percentage
changes in each exchange rate, i.e. for the DM it would be

[ln (DMt ) ln(DMt 1 )] 100:

Plot the daily change data in GiveWin; what conclusions do you come to?
1
This data is from Verbeek (2004) and pertains to the GARCH example illustrated on p.303.

1
Identify an appropriate AR model for each of the exchange rate series.
Hint: You will want to use ACF/PACF to help you here, as well as employ ‘over-
…tting’ to check the speci…cation. You may …nd that there is no evidence of serial
correlation and thus changes in the exchange rate do not have persistence and the
exchange rate itself can be regarded as a random walk.

Test for ARCH e¤ects by going to Test –> Test... and selecting ARCH test
with order 1. At a minimum you should test an order of 1 and 6. Are ARCH
e¤ects present?

Estimate an appropriate ARCH model based on your results from the previous
step. To do this in PcGive, go to Package –> Volatility Models (Garch) and
then Model –> Formulate. . . Select the variables from your AR model and click
OK. In the Model Settings dialogue box select the appropriate speci…cation; for
example, an ARCH(6) model will be p = 0 and q = 6. Click OK and estimate using
maximum likelihood.

Analyze the results from your model. In particular, you will want to check if any
of the parameters are insigni…cant, and if so consider a more parsimonious speci…-
cation. It may be that you want to perform a Wald test if you have more than one
insigni…cant parameter. This can be done via Test –> Exclusion Restrictions. . .

Perform further tests on your model from Test –> Test. . . Recall the Portmanteau
is the Ljung Box test. Can you improve the model?

We will now see if it is advantageous to formulate a more parsimonious GARCH(1,


1) model and examine di¤erent the types of GARCH models that we can estimate with
PcGive:

Exercise 2: Volatility in Daily Exchange Rates - GARCH


Models
Using the same AR speci…cation identi…ed above, go to the Model Settings dialogue
box, but this time specify p = 1 and q = 1 so we have a GARCH(1, 1) model.

Interpret the results. Consider if you can improve the model with a change in the
AR structure. As above, check the diagnostics.

Estimate an EGARCH(1, 1) model using the AR speci…cation you identi…ed above.


To do this you will need to check the EGARCH box in Model Settings.

Estimate a GARCH(1, 1) model with Student’s t-distribution, using the AR spec-


i…cation you identi…ed above. To do this you will need to check the Non-normal
error distribution box in Model Settings.

Estimate a GJR-GARCH(1, 1) model (this type of GARCH speci…cation is from


Glosten et al. (1993). To do this, open Model Settings and expand the GARCH
variations section. Select Threshold GARCH.

2
Estimate a GARCH(1, 1)-in-Mean. To do this, open Model Settings and expand
the GARCH variations section. Select h_t in mean.

For all of the above, you should carefully evaluate the adequacy of your model with
particular reference to the diagnostic tests.

We will now use our GARCH models to produce forecasts:

Exercise 3: Forecasting
Using your preferred AR(p) – GARCH(1, 1) model, in the Estimate Model di-
alogue box hold back a number of observations with the Less Forecasts setting.
Initially, you may wish to try 14.
After estimating the model, go to Test –> Forecast. . . Select the number of
forecasts you would like. If you look under Options you are able to control the
number of pre-forecast observations that are graphed and there is also a check box
for Write results instead of graphing if you desire a print-out of the forecasts
rather than a graph.
Examine the forecasts from your model. Note that you can use Test –> Store
in Database to commit the Forecasts and Forecasts standard errors to the
database.

The procedure for forecasting with ARMA models is similar to the above:

Exercise 4: Forecasting with ARMA Models


Estimate an appropriate ARMA model, holding observations back for forecasting.
Use Test –> Forecast... to produce a forecast from your model.
If you employ Write results instead of graphing you will see that PcGive
displays the HMSE (Heteroscedasticity-ajusted mean square error). This is the loss
function used by Bollerslev & Ghysels (1996) for model comparison. You will also
see RMSE (root mean square error) and MAPE (mean absolute error).
Evaluate the forecasting ability of your model. You may wish to compare it with
another suitable speci…cation.

Points to Note
More details about PcGive’s implementation of volatility models, together with a
tutorial, may be found in Hendry and Doornik (2001).
It is often useful to use Model –> Progress to produce a summary table that can
aid analysis of competing models.

3
References
[1] Brooks, C. (2002): Introductory Econometrics for Finance, Cambridge University
Press.

[2] L. Glosten, R. Jagannathan, and D. Runkle (1993): “On the relation between the
expected value and the volatility of the nominal excess return on stocks”, Journal of
Finance, 48(5).

[3] Harris, R., and Sollis, R (2003): Applied Time Series Modelling and Forecasting, John
Wiley & Sons, Inc.

[4] D. F. Hendry and J. A. Doornik (2001): Empirical Econometric Modelling Using


Pc-Give, Volume 1. Timberlake Consultants Ltd., 3rd edition.

[5] M. Verbeek (2004): A Guide to Modern Econometrics. John Wiley & Sons, Inc.

You might also like