Robustness results

Hi,

I am during a trial, and trying to understand what would be a “good” result for a robustness test. I do the Monte Carlo analysis with 1000 simulations and all methods included.

I found this type of testing example (attached) in the forum of one of the top members, which seems insane to me (What does 52666% AR even mean?)

Do we have any guideline on a “good” robustness results for a strategy? How do you judge your robustness results and how you choose which strategies to dismiss ?

Thanks

As contrast, my best Monte Carlo results look something like the file attached..

I am not sure if I should consider a strategy like that further or discard? Not sure what to aim at

Sorry, I didn’t expose the results till 100%, here they are

Hello, I look at this for max DD % value. I want to that in 95% of time max DD% is not bigger than 15%. I want to see no huge disturbances in curves. This looks ok to me

Hi Tomas, do you mean 15 or 25? (I saw you’ve written 25% in the past).

Thanks

strategy with RDD 68 is nonsense…you could build it, but you will have most probably wrong settings…this is not gone work on real trading

about the MC tests – its not neccessary to do the test alltogether – you could live with 200 iterations and its better to do the tests one by one

and the second picture RDD for GOLD 6 is not enough by my opinion – it should be for the whole history 2006-2018 something like 20-30

to notch: could you show us your best strategy made by SQ and traded live with some history?

Hi Notch, yep it’s your photo. Thanks for your input. Would you mind answering my main question in this topic? Specifying general factors and statistics which will define a strategy “robust” for you, when using Monte Carlo, WF etc?

Thanks.

For example: today I am trying to figure out a new strategy so I have to manually trade it to ensure the rules can be automised. The circles are where I bought and sold (limits). Only one losing trade and currently short and in the money. Problem ,I hate to manually trade although I’m very good at it and I also hate to program but hate manually trading more. (I guess I am the quienessential lazy bugger). So you use SQ. Find a few good entry filters, generate the MQL4 and get programming.

Hi Again Notch, thank you for the extensive reply, helps a lot. I’ll keep training.

I’ve been a technical manual trader for 3 years now as a hobby / capital growing method, going quite nicely in the last 2 years (First year I lost around 4K$ 🙂 ) but thing is I am actually a full time doctor, so I’m short of time, and I don’t enjoy seating in front of the screen for hours per day to trade.. I’m a former programmer (Java, Python, TypeScript) so I decided to combine those two side-interests (And strategyquant makes it damn easy). I am just still confused on the most effective and acceptable settings / parameters / robustness outcomes.. but the sky is the limit with SQ options, so I’ll soon get the hang of it I hope.

Thanks again and good luck.

Notch, i like your walkforward aproach and do something simular. I am not sure thought that it helps alot to use walkforward in Forex last couple of years. Well maybe if a larger change of stops and targets are allowed but then most strategies would have failed in the past made by SQ3 allthought looking at your manual entries it looks like you are over trading and can not be profitable in the long run. I started trading 2002 on futures and overtraded   as well  for 10 years. SQ strategies works fine by them selves it is no use trying to implements your own changes . Trading automatic generated strategies has only one enemy and that is performance per time . The only thing needed to be figured out is when to trade which strategy for optimum performance.

Without wanting to provoke anyone here right now:

I think that the whole thing here has become a self-portrayal of individual persons and no one actually answer anymore to the question of Ilya. For this reason I would now like to come back to the question of Ilya and answer correspondingly helpful:

Basically I evaluate the robustness of the strategies according to their RET/DD value. This means that the RET/DD value must still have at least 50% of the RET/DD value it had from the robustness test after the test. I hope this helps you

i think use 200 simulations on monte carlo analysis enough…if some of the line go below of 0 which is negative that mean the strategy might be not quite good