Thread: Trading Station Back Testing and Optimization FAQ

When I optimize including a particular set of conditions/variables as part of an optimizer "matrix" I get a particular answer for that set in terms of profit made, number of profit trades compared to loss trades.

When I backtest those particular set of conditions/variables over the same time I get a different answer in terms of profit made, number of profit trades compared to loss trades.

Why is it so?

truly blinded

Hi blinded,

Could you provide us more specific information:

1) The name of the strstegy you optimize and backtest;
2) Screenshots with chosen Strategy Parametres (for both: Backtester and Optimizer);
3) Screenshots with results (for both: Backtester and Optimizer);

Please, also check that you have choosen the same Account parametres for both Backtester and Optimizer. Currency, Initial balance, Account lot size and Maintenance type should be equal. Margin requiments should also be equal.

Anyway, when you choose parametres in Backtester, you enter a specific date. In Optimizer you enter some interval. And in this case, Optimizer gives you more extensive results depending on the interval you have chosen. For example, in the DMACD strategy you should choose shot and long EMA period. In Backtester there will be 2 figures, one for short and one for long (e.g. 5 and 34). In Optimizer you should enter the range for at least one parameter (min;max;step), that is 3 figures. And Optimizer will give you all possible result. Then you can choose those parametres, that are the most profitable for the exact startegy.
Please, read more information about Backtester and Optimizer.

I'm looking forward for the information and I'll do my best to help you.

Best regards,
Ekaterina

I have decided not to send you the information you request.

I will work on it some more before I bother you further. It may be that I have the same issues as the previous poster in the code for the strategy.

(well and truly) blinded.....

Ok blinded,

but if you need help, don't hesitate to bother me=)

What does

Cannot load file"/pricearchive/AUDUSD/2012/30.7z". Error: 12150

mean and how is it fixed?

Hi blinded!

I used to face this problem during backtesting process yesterday. Do you still see this error message today?

Yes; I have only ever seen it today, last time was 90 seconds ago

As far as I know, FXCM is working on this problem. Perhaps not all databases are fixed yet.

If you still experiencing the issue, please contact FXCM on-line support. Use Chat command in the TS tool bar for this purpose.

Best regards,
Ekaterina

I will send you a private message in a couple of minutes, please read it.

Hi,

I have reported this in FXCodeBase already, but I believe that this issue is so essential I want to make sure that FXCM gets aware of this directly.

The backtest price data is not consistent with LIVE data scrolling back in time in MarketScope, which produces wrong indicator data outputs in backtesting my strategies. Here is what I see:

I did a small backtest using EUR/USD instrument from 01.01.2012 up to including 30.07.2012. On this last weekend, there is price data on 27.07. up to 21:00, looking at it in m1 resolution, the last candle is 21:00. This means that in H1 resolution I get 4 candles too many (18:00 to 21:00).

I think it should end at 17:00 or 18:00 depending on what regulation you use for the Friday close.

Price re-starts on 30.07. at 5:00 (for both time frame resolutions, of course). This, IMHO, is also too late, unless I am missing something. It should be on 29.07. at 18:00 for the market re-open.

I do not see any way how to get a consistent price picture for both backtest and LIVE, which I believe, is essential to be able to backtest and optimise my strategies.

Going back, the second last weekend, backtest data stops on 20.07. at 21:00 and continues 22.07. at 22:00. This pattern seems to repeat then going further back.

This looks to me as if something is wrong with time zone interpretation of backtest data, the close and open times are wrong.

This has to be corrected in the historical data usage.

One more thing is that it would be nice to have the weekend gap data correctly displayed. I know this is an issue of FXCM servers providing this information to you (looking at other posts).

I would hope to get an indication of when roughly a future release can be expected that fixes this important issue.

Cheers,
Martin

Hello blinded, and thank you for your post!


What is this error message and how do I fix it?

There was a known issue with downloading some of the price history data in the back tester. The issue occurred from Market Open on Sunday 07.29.2012 until Tuesday 07.31.2012 at around 5PM EST. Due to the nature of the fix it could have taken up to 24 hours from the time it was fixed for some clients to see the issue resolved on their platform. By now the issue should be resolved for all clients. Please let me know if the issue persists for you and we can continue to investigate your specific situation.


Thank you again for your post blinded! Please feel free to post with additional questions or contact me directly at julius@fxcm.com

Originally Posted by blinded

What does

Cannot load file"/pricearchive/AUDUSD/2012/30.7z". Error: 12150

mean and how is it fixed?

Hello mstreck, and thank you for your post!


How can I fix the issue with historical data time stamps when back testing on the Trading Station platform?

This is a known issue that our productions team and developers are working on to have fixed. There is a major change coming soon to the way prices and time stamps are displayed when running back tests on the Trading Station platform. Although the expected Estimated Time of Arrival (ETA) for that is tentatively scheduled for sometime this fall, it is possible that it could take longer than that.

As for the issue with the Weekend Data option on the Trading Station that is still an issue that we are currently working on resolving. This issue however could potentially be resolved much sooner than the first issue you mentioned. As soon as I have an update for the situation I will be more than happy to follow up with you so that you remain informed.

We are working as quickly and as diligently as possible to have these issues fixed. FXCM apologizes for any inconvenience this causes you; FXCM appreciates your business and your patience while we resolve these issues.


Thank you again for your post mstreck! Please feel free to post with additional questions or send me an email directly at julius@fxcm.com

Originally Posted by mstreck

Hi,

I have reported this in FXCodeBase already, but I believe that this issue is so essential I want to make sure that FXCM gets aware of this directly.

The backtest price data is not consistent with LIVE data scrolling back in time in MarketScope, which produces wrong indicator data outputs in backtesting my strategies. Here is what I see:

I did a small backtest using EUR/USD instrument from 01.01.2012 up to including 30.07.2012. On this last weekend, there is price data on 27.07. up to 21:00, looking at it in m1 resolution, the last candle is 21:00. This means that in H1 resolution I get 4 candles too many (18:00 to 21:00).

I think it should end at 17:00 or 18:00 depending on what regulation you use for the Friday close.

Price re-starts on 30.07. at 5:00 (for both time frame resolutions, of course). This, IMHO, is also too late, unless I am missing something. It should be on 29.07. at 18:00 for the market re-open.

I do not see any way how to get a consistent price picture for both backtest and LIVE, which I believe, is essential to be able to backtest and optimise my strategies.

Going back, the second last weekend, backtest data stops on 20.07. at 21:00 and continues 22.07. at 22:00. This pattern seems to repeat then going further back.

This looks to me as if something is wrong with time zone interpretation of backtest data, the close and open times are wrong.

This has to be corrected in the historical data usage.

One more thing is that it would be nice to have the weekend gap data correctly displayed. I know this is an issue of FXCM servers providing this information to you (looking at other posts).

I would hope to get an indication of when roughly a future release can be expected that fixes this important issue.

Cheers,
Martin

Thank you Julius for your quick answer to my questions.

Unfortunately, I use quite a lot of time correlations in the entry logic of my primary strategy which uses multiple time frames, i.e. between time frames and also within a time frame.
This makes it almost impossible to verify the code in a backtest. The other issue is that I use in the same strategy daily pivot points as an additional entry filter of price levels, and there I have the issue that the indicator produces wrong values for the first day of the week as a consequence of the bad time stamps.

There may be a chance for me to produce a work-around, but for that I need to be sure of the following:
Do the candles itself represent correct data in the sense of that there is no additional or less candles used in the price streams, or in other words, are 'only' the time stamps around a weekend wrong, but everything else is correct (again, no missing candles?
I have the impression that the candle series looks OK, but I have not extensively verified this, only for a small portion of a H1 chart.
I am not sure whether I can work-around the time stamp issue, but as it looks like a repetitive pattern, there may be a way to do so, I am just not sure if it will be worth the effort.

As you say I have to wait until end of this year, it may be worth for me to do so, since I do not trade manually, only automated, and I am not willing to take the risk to go back to Strategy Trader, although this may be an option until there is a new release of TS II / MarketScope with a solid fix for this.

Cheers,
Martin

Hello again mstreck, and thank you for your post! You're very welcome for the responses, it is my pleasure to assist you!


Does the issue with historical data time stamps when back testing on the Trading Station platform exist only with the time stamp or does it affect candlestick data as well?

I have checked with several departments and the issue you are reporting only affects the time stamp; it does not affect the open, close, high or lows of the candles. Although the time stamps may be off the data is still accurate and correct.

As always please do not hesitate to reach out to me directly whenever any questions come to mind, if you need further assistance, or just to let me know how you are doing. Remember I am always here to assist you!


Thank you again for your post mstreck! Please feel free to post with additional questions or contact me directly at julius@fxcm.com

Originally Posted by mstreck

Thank you Julius for your quick answer to my questions.

Unfortunately, I use quite a lot of time correlations in the entry logic of my primary strategy which uses multiple time frames, i.e. between time frames and also within a time frame.
This makes it almost impossible to verify the code in a backtest. The other issue is that I use in the same strategy daily pivot points as an additional entry filter of price levels, and there I have the issue that the indicator produces wrong values for the first day of the week as a consequence of the bad time stamps.

There may be a chance for me to produce a work-around, but for that I need to be sure of the following:
Do the candles itself represent correct data in the sense of that there is no additional or less candles used in the price streams, or in other words, are 'only' the time stamps around a weekend wrong, but everything else is correct (again, no missing candles?
I have the impression that the candle series looks OK, but I have not extensively verified this, only for a small portion of a H1 chart.
I am not sure whether I can work-around the time stamp issue, but as it looks like a repetitive pattern, there may be a way to do so, I am just not sure if it will be worth the effort.

As you say I have to wait until end of this year, it may be worth for me to do so, since I do not trade manually, only automated, and I am not willing to take the risk to go back to Strategy Trader, although this may be an option until there is a new release of TS II / MarketScope with a solid fix for this.

Cheers,
Martin

Just to verify that the backtest was properly using historical data, I decided to run back a few days from where a trade was automatically made for me in the strategy I have running. When running the backtest, the trade that had been made for real didn't happen in the backtest. Is there a problem with the historical data to where it doesn't have everything needed in a 5 minute chart to accurately run a test?