I became interested in the GARCH lecture on Quantopian so I tried to implement my own solution. It did not generate any crazy returns but the algorithm shows the concept of what you could use the GARCH for.
Here is what I'm doing:
- Every week, forecast new stock prices for the end of the week by using ML-estimated GARCH parameters for forecasting daily sigma and log returns. The forecasted prices are calculated using a Monte Carlo simulated brownian motion
- Buy the stock with the highest forecasted returns and repeat
I use a different minimizer function than the one used in the Quantopian lecture series, this does speed up the process a bit. I also used another version of the probability function for the logaritmic returns when doing the MLE of the GARCH parameters.