Why doesn't it work more recently?

Hi All,

I have been reading as much as I can from the forum, and learning heaps. Thanks for everyone that contributes. I even then went out and learned MQL (already had C experience, so was pretty easy). But I am having one strange problem. I wrote an EA, and when I back tested, it worked really well... sort of. From May 2003 (where I started the testing) up until October 2006, it consistently made a very good profit. Since, then, it consistently made a very bad loss. The turn around was dramatic, too much to be a coincidence.

One other thing that may or may not be worth noting, so I will say it anyway... when running long tests, it seemed to just stop trading after a certain amount of time. I assume there are a limit to the number of trades it can record?

I imagine the first question will be "where did i get my data"? The answer is I went to tool, history center, and downloaded it (clicked download).

Has anyone had this problem before? Does anyone have any suggestions?

Oh, incase it is relevant, it was EURUSD, 30M, and the model was every tick.

if your lot size increases with account balance your EA will choke after it meets its max lot size ( in most cases 100 lots, varing by broker ) this happened to me while backtesting...I was baffeled for hours

Also MT$ is just glitchy sometimes. . . reinstall the program might clear it up.

And the market has changed drastically from 03 - 06..... things that were true then are fairytales now...

Maybe you know this already, but backtesting in MT4 is notoriously unreliable. Have forward tested it in demo? I prefer forward tested resuts over back.

Thanks guys. I didn't know about the max lot size, that would explain why it stopped trading at some point (yes, I was increasing lot size). As for the market changing, I have no doubt it has, but I don't think that is my problem here... it happened too suddenly. I can pretty much tell you the day where the system went from profitable to unprofitable.

As for MT4 back testing being unreliable, how unreliable. Have many people had systems that were profitable back testing and not forward testing? Or the reverse? Personally, I wouldn't have even bothered forward testing if it doesn't pass back testing. Is this a mistake?

the backtester is completely unreliable for intra-bar ticks because these ticks are system generated on what MT4 guesses might have happened. However if your EA just looks at the last close of bars for all functions, then the MT4 backtester will do just fine.

I have also seen the same thing, I think it is the market going though phases, for example moving averages used to work well but then the market became more volitile, last year martingale EAs did well, but this year...not so good as we seem to be in a phase of slow trending.

the increase in $$$$ on the spot market in the last few years is huge.

Thanks aicccia, that is handy to know, and will probably save me a lot of time in the future.

A few people have mentioned that the market is changing, but surely the market changes quite gradually, as in one month MAs work really well, then there is a transition period where the work better some months then others, then slowly, they don't really work all too well at all. (I am open to being corrected if this is not the case). This is not what happened in my EA, it was literally profitable most weeks up until a certain date, then was very rarely profitable.

Having said all this, it is a bit of a mute point given aicccia's input, because it did rely on ticks... but I am just concerned that my data is dodgy, and also mystified why it happened.

what date does it turn on you ?

I really just don't trust the MT strategy tester.

I'm not sure it models the ticks correctly, and I doubt it takes the spread into consideration to boot.

The MT strategy tester works OK as long as your EA is correctly written & you redownload your data at intervals. There is a post on forex-tsd about how MT tries to 'blend' data together from different sources (why I don't know).

what was the model integrity?

and also the modeling quality ?

I think you have to consider that with 'random' distribution
of wins and losses, there are bound to be certain times
where you get mostly wins, others where you get mostly losses,
and others where you get a mix

Something else to consider...

If your EA is time sensitive (only trades during certain hours) then it's possible that you've got a discrepancy in your historical data. I've heard it reported that data downloaded via the history center download button (as you did) comes straight from MetaQuotes as opposed to your broker. I haven't confirmed this, but if it's the case, you could have pricing based on two different timezones. If your EA trades based on prices at the London open, New York open, etc., then I'd expect a significant change in results at the point you switch from MetaQuotes pricing to pricing provided by your broker.

I hope that makes sense.

Also add on to telsa's post above~~

If the data from the metaquotes and your broker data in different timezone, you can actually got a "strangely blended" data for the history and that data may act like a future predictor - until the real data from your broker comes in.

Because datas from different broker are different, it is important to use you own broker's data. With a careful written program (like when you use 30m timeframe), you don't necessary need 90% modeling quality to get an accurate backtest result - as long as forward testing and backtesting produce the same result. That's why people said forward testing (on a demo of course) is important because you can then test it backwards to see if it represent the correct trades that you want it to make.

PS. the data that you downloaded from history center is from metaquots in case you don't know.

Hope it helps

Hi to all,

I am doing manual backtesting myself in MT4 demo using 4-hour timeframe. When I manually backtested, once I placed an entry order, if my trade is profitable EVERYTIME I exited at a candle close. In between my entry and exit time I checked if my stop loss has ever been touched using those 4-hour candles in between. I believe in my case, it should be reliable because if my stop-loss has been touched or past through, the candle will reflect that (as the candle will go past my SL point). If NONE of the candles touch or go past my SL (not just the close but both tails as well), then I am still in the trade and my profit taking exit will be valid. In this case the intra-ticks within those candles do not matter as long as I visually checked if the tails (far end) of every candle since my entry does not touch or go past my SL.

I believe it holds true for manual backtesting in my case or if anyone backtest's strategy is similar to mine. I am not sure if EA backtesting will be the same (any idea?). If the EA in MT4 takes account (will it?) of intra-ticks (so both end of the candle) in 4-hr time frame, then it's fine and backtest will be reliable. If the EA just take account of the 4-hour candle at the close only, then your SL might have been stopped out for intra-tick data with that 4-hour candle but backtest result might not reflect that and thus show inaccurate results.(1 hour time frame as well).

Your thoughts? Thanks.