What are the tests to perform to see if a set of returns data is suitable for a GARCH model? If returns data shows volatility clustering does that mean GARCH is suitable? Do you need to perform Ljung-Box test on the data to see autocorrelations?
This article might help:
http://economia.unipv.it/pagp/pagine_personali/erossi/note32004a.pdf
Following three points can be useful:
- Apply the tests (e.g., Ljung-Box, ARCH) to return data in an attempt to ’see if there is anything there’.
- A proposed volatility model (e.g., GARCH, or some other non-linear model) is estimated, and the tests (e.g., Ljung-Box, ARCH) applied to the standardized and squared standardized residuals in order to ’see what is left’.
- If the model under consideration is adequate, the standardized and squared standardized residuals need to be white noise.
You can always square the centered return series and look at the correlogram to see if there are significant coeffs
You can test for heteroscedasticity (via Ljung-Box for example). If you succeed in rejecting the NULL hypothesis of homoscedasticity, then you can apply an Arch Model.
Following steps to be followed to know if returns data is suitable for GARCH
1. The data should be stationary (using ADF test or correlogram)
2. There should be volatility clustering present in the dataset (run ARCH test)
if the ARCH is present in the return series then we can apply ARCH family models..
- Badges
- Science topic
- Similar topics
- Computer Communications (Networks)
- Simulators