Hi all,

I have been playing with an opening gap strategy that is marginally profitable in live trading this year so far.

However, against my expectations, it is not profitable at all on optimisation in zorro.

I think this may be due to the fact that there is no real edge to it, but I'm suspicious that the optimisation is a little too simplistic.

My understanding is that the first parameter is optimised first, before moving on to optimising the second, and subsequent parameters.

This presupposes that you know in advance which parameter will have the greatest effect on the strategy. I don't know how other than gut feel you would know that.

It also assumes that there is one cluster in the first parameter leading to superior profitability. There may however be 2 or more clusters that are profitable depending on the correct parameters in the other variables. Yet in my current understading these clusters will not be tested.

The gap trading strategy for example may (probably not!) have profitabilty clusters with a close target far stop and small gap threshold giving a high win rate but low risk reward or it may be profitable with a tight stop, high risk reward and a high threshold.

I'm currently manually live trading it somewhat successfully with a 2:1 risk reward, medium threshold and full gap close target. I was hoping that zorro's optimisation would identify a cluster of more profitable parameters, but I now think it cannot optimise across the full spectrum of the parameters, due to its sequential optimisation, or maybe I simply don't understand how the optimisaiton works.

Any help anyone?

You understand it correct. Zorro identifies only one cluster in every parameter range and disregards the rest, even though finding the highest peak would certainly lead to a higher profit in the optimization.

The optimization has not the purpose to find the most profitable parameter combination. For this you would use a different method, such as a genetic or brute force algorithm. The purpose of the optimization is to find the most robust combination where small parameter changes have the least effect on the profit.

Your described gap trading system with the two peaks is in fact two different systems, each one probably based on a different effect. So I would treat them as different systems and optimize them separately with different parameter ranges.

Ah yes, I've been lazy, I was hoping zorro would run them all and tell me the answer

I can of course look at the options as different strategies and optimise them individually before comparing them.

With gap trading, it's more likely I've been lucky and in the long term is is not profitable.

One further thought occurs ...

As a non statistician I know little of these things, but can you explain whether there is a way mathematically to measure and rank the most important parameters on the basis of how much changes in that parameter impact the final result?

That may fall out of the optimisation process and could be reported in the test report? It might signal whether it was worthwhile to re-run the optimisation with a different order of parameters?

BTW please recommend more strongly the importance of reading the manual fully. My gap trading script would have been considerably easier if I had known of the existence of the day functions. I'm sure there are many idiots like me who read the tutorial and dive straight in only reading the manual on a just in time basis.

Yes, it's mathematically possible to determine the most important parameters, but this would require shuffling the parameter order and extremely increase the optimization time. Therefore Zorro does not do that automatically.

When you run an in-sample optimization - no WFO or DataSplit - you'll get the parameter range charts as described in workhop 5. You can in most cases see in those charts which parameters are important. Eliminate the flat parameters that have no noticeable maximum in the chart.

And yes, I'll certainly follow your advice and will even more strongly recommend reading the whole manual .

I was hoping that the Robot had read the manual....