Page 1 of 1
Optimizing Auto Systems
Posted: Wed Mar 29, 2017 6:39 am
Hi, is anyone using TB for optimising auto trading strategies? I am totally new to TB and have recently started coding my own strategies that have been working fine - but I want to optimise them for sustainability - specifically sing the walk forward systems - I have used MT4 strategy tester which is really curve fitting and not reliable - so my question is - if you do use it for auto trading strategies - do you have any tips??
Many thanks and greatly appreciate your response!!
Re: Optimizing Auto Systems
Posted: Wed Mar 29, 2017 9:40 am
By "Auto Trading Strategies" I assume you mean mechanical or systematic trading? Yes that is all we do here. Not many discretionary traders on this forum.
Yes people use TB for optimization all the time. The trick is looking for what we call Robust parameters. Robust parameters are those that have a high probability of generating futures results consistent with the past. It's super easy as you know to curve fit strategies to the past, and the more you curve fit the worse the system will perform in the future. So the Robust parameters are not necessarily the best parameters from the past but the safest or most solid.
There are lots of ways to do this, starting with using a large diverse set of markets and a large set of data. It's important to not optimize parameters for particular markets or for particular economic timeframes. The more diverse and uncorrelated the markets are, and the more economic influences on the data time frame, the better.
Another strategy is to use the multi parameter stepping surface charts. They show the 'safe' spots on the chart that are not peaks. When you simply take the best parameter set you will often get the peak, and the areas around it could be quite bad. So it's important to find 'flat tops' or safe spaces where if the futures is different than the past (which it will be) the results are consistent. Think of the future best parameters as being some jiggle from your current parameters and make sure that you are still good to go regardless of a x% move in best parameters one way or the other.
Another option as you mention is the Walk Forward. This is a nice way to use Out of Sample data (OOS) to test if the optimization was successful. However, this approach uses the Peak method to find the best parameter set, and is also often used again and again to find the best parameters, which in itself is curve fitting and the OOS becomes In Sample quickly. So I would be careful with this, or research carefully what you are doing and whether in the end you have Robust parameters.
Hope that helps.