0

I'm doing an analysis of integrated price series for two different types of crude oil. These crudes are priced off a differential to a main type ("marker"). I built an exogenous variable matrix that includes a main marker for Brent crude and dummy variables for day of the week. When I run the VAR, the t-stats for certain day of the week dummy variables are not significant (Thu/Fri). Is there any precedent for using stepwise regression with a VAR model to remove insignificant features?

Richard Hardy
  • 54,375
  • 10
  • 95
  • 219
Michael Westerman
  • 33
  • 4
  • 2
    It's generally best not to use stepwise regression at all (see: [Algorithms for automatic model selection](https://stats.stackexchange.com/a/20856/7290)), for anything... Moreover, the tests of the individual levels of a categorical variable don't mean what people typically think they mean. If you really need to combine levels of a categorical variable for some reason, you could try: [Principled way of collapsing categorical variables with many levels?](https://stats.stackexchange.com/q/146907/) – gung - Reinstate Monica Oct 26 '20 at 19:34
  • If you care about prediction, consider e.g. what the `bigtime` package in R does with VAR models. – Richard Hardy Oct 26 '20 at 20:12

0 Answers0