When building a strategy, how do you evaluate if it is 'good enough'? Some may say 'Well if it has positive return and low drawdown it should be a good indication', and others will add more 'things it should do' to that metrics list. So what, should I come up with a number of requirements for my strategy and see how close the strategy gets? For instance 'return > 10%, drawdown > -5% , trades < 30', etc.? That's super arbitrary and I haven't much experience with trading so will set these thresholds wrong - just like many others probably would.

Here's the method I'm thinking to implement to improve this.

A) - Stop thinking in binary terms. The higher the return the better, the lower the drawdown, less trades, etc. - no matter if it's past some arbitrary threshold.

B) - Work against a baseline. Hypothesise a 'perfect strategy' for your test data, one that always makes good trades and always makes you money. Another story is how to actually create such strategy (obviously with a look-ahead bias) and how perfect it is, but for now I'll just stick with finding the peaks and valleys of the test data using a modified scipy find_peaks function enter image description here

Each peak is a Sell, each valley is a Buy order. Now, calculate the usual metrics: returns, drawdowns, number of trades, etc. This becomes the baseline.

Having the baseline values, I can now calculate how much my actual strategy deviates from the 'perfect strategy'. For each metric, I can calculate a percentage of baseline my current strategy achieves. The closer to 100% the better. This way I can see which of my metrics are performing worse than others and try to improve these. (In the example below the percentage is replaced by 0-1 fractional values in the fraction column, but it's the same idea)

Red - my strategy, Blue - 'perfect strategy', fraction column - how closely I get, '1' is perfect match. (Red - my strategy, Blue - 'perfect strategy', fraction column - how closely I get, '1' is perfect match.)

The example above is hypothetical and based on a horrible strategy, but for instance I could look at it and conclude 'ah, I should improve everything, but most of all the Return [%], given it's low fraction value of -0.031'.

Does this make sense? Would this be useful and conclusive? Is this already a thing, and if so what is it called? Any thoughts appreciated


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.