1

I have financial time series with non constant variance. I suppose that using ARMA- GARCH i will create more accurate confident intervals for predictions than using ARMA model. This is how i fit model 1. Turn to log returns from original series. (It becomes stationary)

  1. Select appropriate ARMA model by AIC.

  2. Testing residual for ARCH effect using LM test. (Rejects null hypothesis - there is arch effect)

  3. Select appropriate GARCH(p,q) using AIC.

a) So my question is how should i combine this models to create mean_forecast and confidence interval?

b) Here https://stats.stackexchange.com/a/143521/180509 @Richard Hardy suggests to determine both the ARMA and the GARCH parts simultaneously. How i can do it simultaneously and how it implement it in python? Thanks

John Doe
  • 79
  • 5
  • For ARMA/GARCH models, you may be interested in my other post https://quant.stackexchange.com/questions/9351/algorithm-to-fit-ar1-garch1-1-model-of-log-returns/28427#28427 –  Apr 19 '18 at 02:42
  • Ty @wrtsvkrfm, do you have any suggestions how implement you algo in python without hand coding? – John Doe Apr 19 '18 at 09:47
  • Probably use R, or STATA. STATA has hundreds of pages in its user guide for forecasting models. –  Apr 19 '18 at 15:59

0 Answers0