What happens if you think you have struck gold? - page 3

 
I have backtested numerous EA's I have created. I am the author but I hire programmers to do the coding. I have a constant backtest going 24/7 on something and I consider myself very experienced in backtesting using MetaTrader. I have done backtests that resulted in over $250,000,000 in profit only to realize that my data was no good. Then I downloaded quality data, ran the same test and the account went to "0". You have to make sure your data is quality data. Also, I noticed that the test you ran had only a 25% Modling quality which means the results are suspect. I have also created five EA's that backtested with excellent results only to fail in forward testing. Live demo forward testing will really give you a better result but is still not the same as using your own money. Right now I have an EA I created that produced over $12,000,000 from 1/1/2007 until 7/10/2009 and am now forward testing it for proof that the backtesting was correct. I have learned to be very suspicious with backtest results that look like I have found the holy grail.
 
oldchartreader:
Also, I noticed that the test you ran had only a 25% Modling quality which means the results are suspect.

Your comments are otherwise okay. But the above statement is way too simplistic. Please don't take this the wrong way, but I don't think you fully understand how the % is calculated and what it really means.

Additionally, the requirement is dependent upon the EA and about exactly what data it needs to make its decisions.

- Some EAs can be accurately tested with very little (say D1 OHLC) data

- Other EAs will need actual tick data to have any chance of a representative test

If, as it sounds, your strategies dictate that you require very accurate fine-grained data, then I'd ask you - do you really want modeling of any "quality"?


CB

 
tootrue:

Any advice - this robot has gone from £10 000 to over £2million in 13 months


Panthers Claw

Don't get too exited just yet. Some general rules:

-If it works on bactest it MAY work on forward test...

-If it works on forward test it most likely will also work on backtest.

If/when you do a forward test (with a normal broker) I'll bet my hat that you'll be sorely dissapointed... the EA may not even show ANY profit.

EAs on backtest can be either deliberatly, or by accident, be made hugely profitable... there are many 'pitfalls' in backtesting.

E.g the EA can use bar0 of a higher TF, which means it is peeking into the future... works great on backtest where bar0 of all higher TFs are

available, but obviously it will fail miserably on forward test.

I've actually made such an EA and the profits whent off the scale...the Tester didn't have room for all the digits in the profit column!

I have now learned to fully ignore backtests as proof of anything, ONLY forward test or live trading matters.

An other thing I've found is that indicator based systems or strategies most likely will not be profitable in the long run... they may work so-so for a while

then they start loosing.

 

I don't like to dissapoint you but at least do a ten years backtesting before posting anything.


Yesterday I coded an EA that worked like magic for all 2009, but stroke down for 1999 data. Don't be too happy so that you won't be dissapointed

 
DayTrader wrote >>

Don't get too exited just yet. Some general rules:

-If it works on bactest it MAY work on forward test...

-If it works on forward test it most likely will also work on backtest.

If/when you do a forward test (with a normal broker) I'll bet my hat that you'll be sorely dissapointed... the EA may not even show ANY profit.

EAs on backtest can be either deliberatly, or by accident, be made hugely profitable... there are many 'pitfalls' in backtesting.

E.g the EA can use bar0 of a higher TF, which means it is peeking into the future... works great on backtest where bar0 of all higher TFs are

available, but obviously it will fail miserably on forward test.

I've actually made such an EA and the profits whent off the scale...the Tester didn't have room for all the digits in the profit column!

I have now learned to fully ignore backtests as proof of anything, ONLY forward test or live trading matters.

An other thing I've found is that indicator based systems or strategies most likely will not be profitable in the long run... they may work so-so for a while

then they start loosing.

I like your post/comments esp the last sentence. You are negating all the indicators or any oscilators only? Personaly I dont trust any oscilators. Price following indicators like Moving Averages based strategies too fail in the long run? Also if you are forward testing how many days of testing is enough, though this varies on the strategy/tf used. But what if there is a strategy on 5min, how long you need to forward test before u see the light bulb? :)

Thanks in advance for your reply.

 
Kent:

I like your post/comments esp the last sentence. You are negating all the indicators or any oscilators only? Personaly I dont trust any oscilators. Price following indicators like Moving Averages based strategies too fail in the long run? Also if you are forward testing how many days of testing is enough, though this varies on the strategy/tf used. But what if there is a strategy on 5min, how long you need to forward test before u see the light bulb? :)

Thanks in advance for your reply.

Well, two of the biggest mistakes in EA codeing (which looks great on backtest) is using future data as mentioned above, and ignoring the fact that the Tester will trigger an order as soon as the price is right... even if its just a short tick-spike...

on real trading you'll see that most of the spikes are NOT going to open or close your order ( off-quotes, re-quotes, slippage, etc). I've many times seen my orders not close even though the price has been 10.0 pips beyond TP for several minutes, then

a retrace or reversal and the order is still not closed. This is especially bad for short term scalpers. A fairly short forward test (say 50 trades) will show if there is a terrible (programming) flaw in the strategy, but really you need to forward test

for months and 100's of trades to get a reasonable good picture of things.

As for indicators I've basically ditched'em all, maybe except for EMA and Pivots. As we all know only price followers works ok on strong trends, and only oscillators works ok on ranging. Sad truth is that price is a combination of the two, except for short

periods where either range or trend is ruling.

If your main goal is not programming but wanting to be PROFITABLE then :

-REMOVE ALL indicators from the chart, they will only mislead and destroy!

-Lower TFs contain a lot of random noise, so work on minimum 1H TF, 4H and 1D are even better

-Study the CANDLE-chart carefully, MOST IMPORTANTLY identify support and resistanse levels where you can find them... yes NO indicator here...just eyeball the chart and draw H-lines.

-Study the candle patterns at points of interest, especially just before reversals

-Study breakouts and why/where they ocure.

You just might be amazed at what you'll find!

Sorry this is bad news for indicator freaks but I have been there and spent hundreds of hrs on programming and indicator testing... MY conclusion is that its not the way forward to profiable trading.

 
DayTrader:
...Sorry this is bad news for indicator freaks but I have been there and spent hundreds of hrs on programming and indicator testing... MY conclusion is that its not the way forward to profiable trading.


DT
I'll second all that - I hardly use indis at all for entry, though they may be part of eg divergence filtering

A growing problem for indi-based EA's is that more people are using STP/ECN type broker feeds which have further very significant negative effects on indicators

-BB-

 
BarrowBoy wrote >>

DT
I'll second all that - I hardly use indis at all for entry, though they may be part of eg divergence filtering

A growing problem for indi-based EA's is that more people are using STP/ECN type broker feeds which have further very significant negative effects on indicators

-BB-

I don't use indicators at all for my strategy; just some college level statistical and probalistic math.

Mike

 

It will be better if you only post your results when you struck gold with real live accounts. Based on my experience, any golden back tests and even demo tests could easily fall apart when it comes to live test.

This is because brokers with dealing desk or banks on ECN accept orders and close orders completely different from demo tests, not to mention history back tests.

So if your EA scalps just a few pips in short amount of time, it may not work when it comes to live test.

So any of you above show live trading results?

 

In response to what was addressed to me:

.

1. I'm currently testing my EA live on 3 brokers with slightly different settings. Higher spread brokers seem to break this strategy -- like it would do to most scalping strategies.

2. The modelling quality is 25% because it's M1 data, it's impossible to get better than that. Other timeframes use M1 data to get to 90% modelling quality but it's the same data nonetheless.

3. The only indicator I use is an EMA the rest is all price action. The EMA is not even necessary, it just complements the strategy and raises the profits and lowers the drawdown ever-so-slightly.

.

I would be very interested in some quality backtesting data, perhaps one that has almost every tick. If you could point me towards something useful I'd appreciate it. The only real data I have now is like the last week or two off my brokers and it turned out profitable by a big margin with only 100:1 leverage.

.

Also, know that my strategy is mostly about money management and not fine tuned entry points. With this method I can afford to lose 30+ times in a row to break even after a half-win.. and any trade that is opened has a chance of being a jackpot which regularly quadruples your balance on a 100:1 leverage account (this would then take hundreds of losers in a row to break even). With 500:1 leverage, I've seen up to 12x the balance on a single trade in backtesting. Money management is the same on a backtest or live so it's always at least a possibility that this could happen live. I'm not saying I'd count on it happening but there's always this chance that it could -- however small it may be.

.

Jon