Topic: Too good to be true?

I've been playing around with Monte Carlo simulation trying to evaluate the tool, and then the strategy. Having read some relative sites and papers regarding the simulation with Monte Carlo I still haven't come to a decision as to what a good graph looks like. Looking at the Confidence Table below one can assume this look pretty good. The settings are as follows (25 Counts):

Market variations
Randomize history data: Yes
Randomize payout: Yes

Execution problems
Randomize slippage: Yes
Randomly skip position entry: Yes

Strategy variations
Randomize indicator parameter: Yes
Randomize backtest starting bar: No

Data range change ATR %: 20
Minimum payout %: 70
Maximum slippage (points): 5
Skip entry probability %: 2
Indicator change probability %: 20
Indicator max change %: 20

I tried few valuations and realised that the default settings are a good set. However, I don't think the indicator change probability should be as high. Especially when we consider one modifies/updates their strategy often?  What is your take? The confidence table looks took good to be true..?

Post's attachments

MCS.jpg 213.95 kb, file has never been downloaded. 

You don't have the permssions to download the attachments of this post.

Re: Too good to be true?

It does look great. How big is your data set for this chart? I haven't actually had great luck with Monte as it's all a matter of interpretation, been reserving a small sample of data for last check-up, kind of like another OOS, and things have gone every way - quite nice Monte with disastrous behaviour in small sample, nice Monte with nice OOS and not so good Monte with acceptable OOS. So yeah, wish we had one correct answer.

3 (edited by Stav 2016-10-03 22:33:19)

Re: Too good to be true?

I generated a strategy with 40% OOS. The counts of test are 25, tried 50 and 100 as well, not much difference in w/l ratio, only a small 1-2%. I imported 250000 bars for the backtest and this generated 181 trades. The 250000 bars take us back to May/June 2013.

I guess the best way is to test the EA on a  demo, see how it behaves, but what good is that for a month, or two..

Re: Too good to be true?

It will be hard to have a good reading out of a month's test - it will make ~5 trades only if we consider its average performance, can you read something useful out of it?

Apart from that, having such a curve on 250k bars is quite an achievement.

Re: Too good to be true?

can you read something useful out of it?

4 trades in two days actually as I had it on on a demo since Sunday night. Score is 4-0

having such a curve on 250k bars is quite an achievement.

I believe the EA trades the gaps, a usual phenomenon between closing of a bar and opening of a new. And at the same time take into consideration the BB level. All 4 trades have a different close to open price. See the picture of one example:

Post's attachments

EurNzd 30M Trade.png
EurNzd 30M Trade.png 4.27 kb, 1 downloads since 2016-10-04 

You don't have the permssions to download the attachments of this post.