0

Of the 60 series in my dataset, 26 don't exhibit an ARCH effect. I have first fitted an ARIMA model (auto.arima() in R) and tested it's squared residuals for autocorrelation using a Ljung-Box test (Box.test() in R).

26 series do not have an ARCH effect. Can I proceed with multivariate GARCH analysis using all series or should I exclude these series from further analysis?

Ferdi
  • 4,882
  • 7
  • 42
  • 62
user134924
  • 53
  • 6
  • 1
    The main idea of using a multivariate GARCH is to both capture time-varying volatility and correlation between assets. In theory you can have a multivariate GARCH model, where some of the volatility processes are constant. If your application portfolio optimization I do not think you should remove the assets. – Johan Stax Jakobsen Jul 17 '18 at 10:37
  • Thanks, Johan. Yes, portfolio optimization is what I will be heading to after measuring time-varying correlations between series in my dataset. – user134924 Jul 18 '18 at 00:04
  • Yes you may, then perform backward selection. – akkp Jul 22 '18 at 11:53
  • @akkp, backwards selection is almost always a bad idea. To understand further, it may help to see my answer here: [Algorithms for automatic model selection](https://stats.stackexchange.com/a/20856/7290). – gung - Reinstate Monica Jul 22 '18 at 16:56

0 Answers0