This is a simply strategy that use ARIMA-GARCH model to forecast the return of the stock, if the return > 0 we long, otherwise, we short.

First, we find the best p,d,q value of ARIMA according to AIC. Then we combine it with GARCH(1,1) to build our final model.

(*not a good trading strategy, made me lose some money)

Conclusion:

For The arima-Garch model, it actually cannot forecast the stock return pretty well. There are several reasons.

(1)The garch is not able to perform well with asymmetry. The asymmetry occurs when values in time series have wider swings from negative territory to positive territory.

(2)It cannot capture long-term dependency or long memory(Shi & yang, 2018). The long memory occurs when the autocorrelation function decays very slowly and remains significant persistently for a prolonged time.