I had a few classes related to time-series econometrics, however most were theory heavy. I would like to practice this, so I will try to analyse few stock prices, however I am not fully sure about the steps, so would appreciate some advice. I will use intraday stock data (very high frequency for a 6-7 year time-period, so several 100 thousand rows of data)
The main goal would be forecasting and predicting based on other time-series.
1, First is the Box-Jenkins test, we make sure about stationarity, white noise etc. this part is clear.
2, Now I'm not sure about this one. I will have to end up using the ARCH-GARCH model family, but do I have to do the ARIMA model first? In the R code I think the ARMA parameters go into the GARCH , but I did not see that being the case in theory, so is the AR(I)MA required to do, or I can safely skip this part and jump to GARCH?
3, Is there any formula to determine which GARCH is the best, or I would have to trial and error a few that I suspect to be fitting? (mostly likely Treshold (-> GJR) for financial data).
4, Something was mentioned that there is a difference between aggregating the intraday data and if using it by itself. Related to the HAR/realized volatility models, but I'm not sure what is the exact difference between the 2 usages? When should I do these 2 methods? Similarly, if I do HAR, do I need to do the ARIMA/GARCH process before?
5, What's the main difference between multivariate GARCH and VAR/VECM models? From what I understood multivariate-GARCH models are very intense computationally and harder to interpret compared to a VAR/VECM? Any reason to use them?
EDIT: Also I would be open to any books/resources that explain these models in detail, with explanations on what each parameter means in the equations etc.
