Hi all, New here so go easy on me I'm currently developing a strategy which shows very promising backtest results. In my analysis I've included various indicator parameters and parameters such as key support / resistance levels, day of the week, position direction etc... My question is, when optimising the strategy do you set the parameters to the maximum individual return or the total return across all parameters. To explain what I mean - two variables I am struggling with are day of the week (Fridays negative return) and proximity to identified support or resistance. As a standalone the optimum return as a rule for the system is not to enter a trade, for example, when price is within 20 points of identified support or resistance. However, when I combine with, say, not trading Fridays the optimal return point changes to a lower figure... If I plan not to trade on Fridays then do I assume the lower rule for support and resistance is the correct one? I'm sure this is simple and I am leaning towards optimising each individual parameter not the total as this is probably just data bias (unless the variables are correlated of course).... Thanks all
The goal of course should be to optimize the system performance (note I'm not saying return). The simple approach to that is to identify individual parameters and optimize them, expecting that it will result in a better total system, however this is not always the case. It could be that certain parameters are actually negatively correlated within your own system such that optimizing for one actually hurts the total. If I understand you correctly, you determined the 20 points from S/R is optimal when trading on Friday, but the system does better when not trading on Friday. However, when not trading on Friday, the optimal point is less than 20. Your question is, does that mean the lower value is correct or is 20 correct? The answer is it depends on whether you're trading on Friday or not. You may find that using a value greater than 20, but not trading on Thursday is better. Or that using 20 and not trading on Wednesday is better. Or trading Mon-Thur mornings and not trading Friday afternoon is better, but only when the value is 10. The point here is that what is optimal in one case (trading on Friday) is not necessarily optimal in another case (not trading on Friday). What you need to do is test all 4 cases: 20 + trade on Friday 20 + not trade on Friday lower value + trade on Friday lower value + not trade on Friday Whatever gives you the best results overall is the optimal strategy. Optimizing individual pieces is a way to get started, but when you get in to more complex systems, optimizing a single parameter can have a net negative impact on the system overall. This is not unique to trading systems mind you, but is applicable in many areas of modeling. Also note that "return" is not the only criteria by which a system should be judged. To each their own. Good luck in your quest to optimize!
Thank you Lee for the quick reply.. I guess you are right, in this simplistic example I am basically cutting out more trades with the higher figure but when combined with non trade of Fridays its actually cutting more winners than losers. In reality I am also combining with a myriad of other metrics so I very quickly end up with a huge amount of combinations to test. I suppose I want to avoid overly 'curve fitting' to the data set but I don't really know how to do this (out of sample testing also yields good results). I am using more than one metric for measurement of performance and the lower S/R figure improves all of these. Thank you for the watch out. The main metrics I am looking at are - Total return, win rate, risk:reward, T-test, average win/loss, average return, standard deviation. I will get around to sharpe ratio but if I am honest I dont really know what a good figure is here! Are there any more measurements you'd recommend?