Data format for GARCH (1,1)

[alexnovitzky]

Hi All,


I am looking at modelling the volatility of numerous economic variables using GARCH (1,1) analysis. These variables include exchange rates, GDP, interest rates, etc

My question is, is it necessary to transform the raw data into a volatility series using a volatility metric or can I use the data as it is?

Many thanks
Alex

[trubador]

It depends on your research problem. From the technical point of view, you do not have to make any transformations as long as your data is stationary. GARCH inherently models and identifies the (conditional) variance part. In general, return series is used for exchange rates, whereas real growth rates are preferred for GDP.

[alexnovitzky]

Hi there,

I am researching the source of the volatility of the South African Rand/ US Dollar exchange rate. I'm needing some
assistance regarding the best method of constructing the model. The explanatory variables that are being tested are:

ZAR/USD exchange rate
Net purchases of bonds by non-residents
Net purchases of equities by non-residents
Commodity price movements (ToT)
Productivity shocks
Relative Equity Returns (JSE vs S&P 500)
Short term interest rate differentials (RSA vs USA)
Long term interest rate differentials (RSA vs USA)
Money supply volatility
Domestic Credit
Foreign Exchange Reserves

Currently I'm using eviews and the challenge that I'm having is in determining the best format to measure volatility using my quarterly
variables, as well as whether a GARCH model is possible. With regards to the data, I'm debating whether the variables should be used in their raw format
(differenced to be stationary if needed), transformed using a rolling standard deviation or volatility metric, use 4-quarter % changes, or a
combination of these. In addition, since I'm using quarterly data, there may not be a sufficient ARCH effects to use a GARCH model and so I may need to
use another model such as ARIMA, ARDL etc.

Please share your views?

Many thanks

[alexnovitzky]

Please see attached the data that I am using. Thanks.

[bate]

Hello There
Thanks for your answer first.
I am trying to see the impact of Volatility in GARCH (1,1) model. I have four crops and I proved that my data is non station and I changed to I(1). After doing this I also add dummies to see the impact of volatility in specific months. When I run the formula in Eviews by entering the variable and the constant in the equation estimation and the dummy in the variance regressors,I found the following results, for example one commodity.
Distribution Corn
W 0.449(0.137)
α 0.486(0.000)
β 0.617(0.000)
Dummy 15.229(0.000)

σt2  W αε(2t−1) βσ(2t−1) + Z,t

= 0.449 + 0.486ε(2t−1) + 0.617σ(2t−1) + 15.229DUMMY
( 0.137) (0.000) (0.000) (0.008)

What I am not comfortable is that how to clearly explain the results as I have some dought in my results, because of the fact that ,
1. α+β is greater than one. I thought it is suppose to be less than or equal to one.
2. I am alos looking for more explanation of the the results.
3. It is possible to calculate the specific volatility according to the results?
Highly appreciated for your help and feed back