What is the best way to optimize the lookback parameters for the long only crypto momentum strategy without overfitting? For the lookback in 'n-day returns', 'n-day standard deviation' and 'n-day RSI' what is the best way to optimize for n under each of these parameters?
The best way to optimize the strategy parameters is to divide the data into two parts namely train and test dataset. Then, use train dataset to optimize the parameters such as 'n-day returns', 'n-day standard deviation' and 'n-day RSI'. And then check the strategy performance in the test dataset. If the performance in test and train is not much different then you have successfully optimized the parameters.
Note: Repeatedly optimizing in train dataset and then verifying in test dataset until you find the strategy which works in both dataset might lead to overfitting. In such a scenario, it is advisable that you do paper trading for 3-6 months and check if the performance is not significantly different in paper trading compared to performance in the train and test dataset.
I hope this helps.
Sounds good - thank you for the insights - will be careful of overfitting. Is there a suggested method on how to implement optimization? Currently I have been looping over a range of predefined values for my parameters.