SQX Strategie gets other results on liveaccount as in SQX by himself

This is rather normal, you´ll always get such differences, it´s because of the live spread being a little different to the simulation with static spread - so a stop loss might get hit in the backtest while it was not live. Most important is that they do not affect profitability in the long run. You can assure this by using the data of your broker (instead of generic data) and by letting the strategy pass monte carlo tests (especially in relation to spread / slippage / randomize history data by a few percent). If it passes those tests fine, such differences do not really matter in the long run.

Thanks for the feedback. But I think the answer is a little bit more complicated because I don't only notice this phenomenon now but have observed it very often on all EAs I had.

My confidence is accordingly a little shaken.
But I decided to do a real broker test with the EAs now.
All EA´s run at the same time with 3 different brokers, among other things also Dukascopy. After 2-3 months all data are placed next to each other and compared how reliable the backtest of SQ actually is....especially in the real test with Dukascopy you will see how trustful SQ is.

Voted for this task.

This absolutely happens with all EAs, regardless from which trading strategy generator. Been in this for 10 years, never was different. It is IMPOSSIBLE to get 100% the same results live as on the backtests. The best thing to do is to "monte carlo" your strategies for at least spread / slippage / skip 5% of trade entries and exits / and if they survive that, these normal live trading differences won´t break the strategy as well and it´s ready for live trading, even if differences occur.

The only way this is a valid "bug" is if the trades taken on your live account were executed in a way that the algorithm you made did not specify. I haven't made a single strategy that hasn't had losing months. Unless your strategy wins every month its been tested (you should make sure you do a final test across the entire range of data) then its bound to have losing days/weeks. The key to seeing if your strategy is bad is if those losing days/weeks are outside of the norm of the drawdown you measure your strategy to have. So a better justification of your strategy not working would be if you experienced abnormal drawdown or if the trades were executed improperly as stated before. You also have to make sure your tests account for variables such as commissions and spreads accordingly and test everything on real tick data in the end in order to see what the truly accurate tests give you back for results to compare to live accounts. Simply stating that the algorithm no longer makes money is vague and misleading without any context. It would also be more helpful to look at how an entire portfolio of strategies performs live when trading simultaneously instead of just looking at GBPJPY which is one of the most volatile pairs you can trade with one of the highest spreads that can drastically change based on market conditions. I would recommend more context is provided before concluding something is wrong with SQX. 

Hello Marcel,

I am closing this task as it is probably not a bug in SQ. 

The other thing is these kinds of strategies (using multiple symbols or timeframes) are very hard to debug. 

We would need some more materials to work with. We haven't discovered any problems in the attached strategies so far.

A lot of fixes has been made since the task was opened and I believe SQ backtest engine is reliable enough. 

We are doing automated tests containing over 1000 strategies before every build to make sure we haven't introduced new bugs. 

Feel free to add a comment if you have any doubts or have new information on this topic. 

Best regards,

Tomas