Optimisation ranges and nice, rounded graph of results against parameter value

[Adamus]

Thought I'd better post something in the Mechanical Trading section before the sysops close it down due to inactivity.

It occurred to me that perhaps we mechanical traders are just too anti-social. So if that's you, well don't bother replying. And don't tell me you weren't going to anyway. Anti-social ****ard :jester:

Well I can see my fingers tapping the keyboard but all I'm reading on the screen is "bla bla bla" so I better get serious.

I'm doing some optimisation on my system in the hope that God smiles on me and my walk-forward isn't the abysmal failure I'm getting used to.

I'd like to see the parameter / variables that I optimise change the result in a nice, smooth, rounded up-and-down way, so that I don't have any trouble identifying an optimal value for it and so I don't get the feeling that it's actually curve-fitting as I watch it.

I'm working on a break-out system where I enter on a break-out and exit on a trailing stop, and right now I'm optimising the trailing stop. Here's the results:

The top line is the complete system, long and short together. The bottom line is just short trades. The x axis is the factor to multiply the ATR by to get the trailing stop distance.

If anyone can correct my logic: the first line looks OK, I should choose a value of 1.6 or 1.7? The peak value is higher at 1.5, but that's the drop-off point downwards roughly speaking.

But the thing that bugs me is the second pink line. That's the profit from short trades. So what do I do?

Ignore it?

Choose 0.9 which looks to be an unstable value (peaky instead of rounded)?

Or should I give up on this system because of its apparently inherent instability?

 

[A Dashing Blade]

Wrt to your short trade, all that graph is saying to me that (net net) the market went up over your time frame.

 

[Adamus]

Yes, there's a general upwards trendd, but the short trades' optimisation values graph is all over the place. How can I choose an optimal value?

You mean I should just abandon it ?

 

[A Dashing Blade]

What are the results with a dataset (ie market) that's in a downtrend?

 

[Adamus]

Dammit, you're blowing my simplistic optimisation and walk-forward plan to pieces.

I just wanted to optimise over a 2 year period and walk it forward 3 months or 6 months to see if the system held up on unknown data.

What you ask implies that I need to go back over the market data to a period with a bear run and see what the results are like. OK, I'll try. Bear with me. pun

 

[cedricthefrog]

why do you need one factor? can't you use one for long trades and one for short trades?

 

[Adamus]

Yes I suppose I could. But where do you stop? Do you also optimise the moving avg length seperately, and also the distance to the stop?

Ultimately you could say I should treat long and short as seperate systems.

The immediate reason why I don't do that is because my software - the trade simulator and the risk analysis program - are set up to handle amalgamate long and short trade results.

My first reaction is that seperating long trading systems from short trading systems would make it easier to curve-fit the optimisation, because you have another variable in the equation, and half the number of trades. I'm happy to admit that I haven't thought that through and am willing to be told I'm wrong.