My USDJPY Daily Strategy

I thought I would share one of my strategies that I am currently testing in a portfolio on a demo account.

This is one of 12 daily strategies that I have just started testing.

It had about 3 years of IS testing and about 10 years of OOS testing.

It seemed to perform quite well.

It has a win % of about 50.0%, and has a payout ratio of 2.37.

Hi phild,

what have you done to prevent curvefitting ?

thomas

Hi phild,

what have you done to prevent curvefitting ?

 

thomas

Hey tnickel,

Basically, I used a large section of the testing data (about 10 years) was all tested on OOS data. 

I believe that this is a statistically significant timeframe for testing.

I also did some robustness testing using 500 monte carlo analysis runs (I think it was 500) and also changing a few other variables like:

1. Randomize trades order, 

2. Randomly skip trades,

3. Randomize strategy parameters,

4. Randomize starting bar,

5. Randomize history data,

I think this might be sufficient, but I am demo trading this strategy now along with 10 others together as a portfolio.

Hi phild,

I think the half of you portfolio is curvefitted.

This was is my experience in the past.

thomas

Hi phild,

I think the half of you portfolio is curvefitted.

 

This was is my experience in the past.

 

thomas

Hey tnickel, 

Okay, I would like you to explain to me:

1. How it is curve fitted (so I can rectify this problem), 

2. Also give me some details on how you would test a strategy you have developed, 

eg: do you test 10 years on IS vs 3 years on OOS.

3. If you wouldn’t mind, tell me a bit about your past experience with SQ.

I haven’t had this programme for very long, and would appreciate the advice…

Thanks.

Hey phild,

I tested the strategy using Alpari data and got very different result. What data source did you use?

Hey phild,I tested the strategy using Alpari data and got very different result. What data source did you use?

Hi,

I used data from forex tester data.

I am still demoing this strategy and a few others before I go live.

I still have about 10 trading days to go before I have 1 month of trading experience.

I will need to probably need to use different data sources to test strategies before I go live.

Hi phild,

I tried using forextester free data and got the same different result. Perhaps you should check into this also.

My guess is the broker time difference causes this. I use GMT+2 for the forextester data. Hope this help.

Hi,

the GMT can be the reason if the strategy use the time.

If the GMT is ok the strategy should the same equitycurve with different data-source.

If the equitycurve is different the strategy is curvefitted.

thomas

Yes, I think that the time the broker uses could be the problem.

Either way , i think trying a few different data sources certainly can’t hurt.

Thanks for the input all.

Also I am wondering with backtesting did you test on 1 minute interval or even tick? This will give a much honest backtest results then for example testing only on 15 min time or 1H etc…

Regards,

Dennis

Also I am wondering with backtesting did you test on 1 minute interval or even tick? This will give a much honest backtest results then for example testing only on 15 min time or 1H etc…

 

Regards,

 

Dennis

Hi Dennis,

I think I only tested it on a daily timeframe. I am not sure if the results will be substantially different if I changed the timeframe.

I am doing some walk forward optimizations on some other strategies that I have developed at the moment.

Cheers.

Hi,

You can test is on 1H interval but then use testing on 1 minute basis. This lets SQ look ‘into’ the 1 H bar. Try this and see what the results will be…

Hi Dennis,
 
I think I only tested it on a daily timeframe. I am not sure if the results will be substantially different if I changed the timeframe.
 
I am doing some walk forward optimizations on some other strategies that I have developed at the moment.
 
Cheers.

Not sure if one gets enough sample trades using daily bars, thus the % chance it is good strategy is lower for lower X # of bars in Y years. Can look darn good I’ve found in the past until it was pointed out to me the low number of trades is less chance it covers greater % of possibilities.

Hi phild,

do you have an myfxbook link for this demoaccount ?

I think this portfolio have crashed.

I have done many tests in the pasts. Most of the strategies are curvefitted.

thomas