Why forex robots fail?

Why forex robots fail

Put your seat belts and stay with me, this is going to be an interesting article because it might answer one of your big questions: why my best commercial forex robot is falling like a rock in spite of great backtests, Monte Carlo and Walk Forward analysis? I put all my trust and money in it and now it goes down?? Damn with those analysis, they all are nothing but scam and empty words! Or is it? People are emotional beings driven by fear and greed. They always want to blame someone for their failure, lack of logic, discipline and knowledge and when there is no one around they blame God or Satan. When I was a kid I used to ignore my fathers advice who told me not to play football with older boys. Of course I did exactly the opposite, those boys took my ball, slapped me then  run away laughing  leaving me in one of the most deepest despair a 9 year old child can bear. My father didn’t say a word when I asked him to run after them in order to get back my beloved ball. At that time I blamed him but I learned my lesson. Never get into something you can’t handle based only on pure hopes and expectations. Before going into further details, let me remind you a few things:

1. Don’t invest what you can’t afford to lose, this way you’ll always be relaxed during drawdown periods. My monthly wage is $5,000, I don’t invest more than $500/month. Those are my spare money. Forex is a risky business, only 5% of traders are making money and most of them are psychopaths.

2. Don’t fall for hypes. Don’t buy anything without a proper analysis but once you buy something, make sure you take the full responsibility.

3. Don’t buy something based on faith, hopes and expectations. Leave those for church and rely on logic only when you trade. I read once that during the winter of 1884 a group of people sold everything they have just because their religious leader said that Jesus will surely come that night. Guess what? He didn’t come, they were left penniless but there were still waiting for Jesus! Don’t do the same, you can still wait for Jesus in the comfort of your house 🙂

4. Don’t buy forex robots unless you know exactly how they work, the strategy must be fully transparent! Less rules, the better! More rules = the robot is curve fitted and will probably fail in real live trading.

Now, let’s perform an analysis on one of my private bots, this one is called DayPivots because it trades only near support and resistance lines and have only 3 available parameters. It’s one of the simplest and effective strategy I can think of.

Here is 13 years backtest:

DayPivots backtests

DayPivots backtests

Maximal drawdown: 1053 pips
Total profit: 20,400 pips
Drawdown length: 393 days

Ok, so I have to be prepared to bear at least 393 days without profit and see my account losing more than 1053 pips. I can do that, but how reliable this strategy is? Does it work without crashing without any optimization on a highly correlated pair? Let’s try it on USDCHF using exactly the same parameters I used on EURUSD.

DayPivots USDCHF

DayPivots USDCHF

It survives but it doesn’t look good. Ok, maybe I have to choose another set of parameters. Please remember that backtests represents the best of all possible worlds! Which means, it will get worse in real live trading.

Is it curve fitted? Yes, it is! Because I have chosen the best outcome after I optimized it from 2000 to 2013. In order to avoid that I have to choose a time period in which the market changed many times, choose the best performing parameters then test it against other periods. If it does well, then the strategy is sound enough to be used, if not, the strategy should be trashed in spite of good results.

Here are some random parameters from 2006 to 2010. Please note this is not the best outcome, it’s among one of the worst!

DayPivots 2006-2010

DayPivots 2006-2010

Well, doesn’t look too pretty but it survives. Let’s see how those settings are doing over a much larger unseen period, like 2000-2013.

DayPivots 2000-2013

DayPivots 2000-2013

Wow, it does pretty well over 13 years using parameters resulted from 4 years optimization, the drawdown is the same, profits are equally distributed, this has a great potential! It may suffer quite a drawdown over the years but at least I know almost for sure I can rely on this strategy. Keeping the same parameters, let’s see how it behaves on USDCHF over 13 years. Does it survive?

DayPivots USDCHF 2000-2013

DayPivots USDCHF 2000-2013

 

That’s a nice surprise! Without any optimization, using the same parameters, it does even better than the original pair!

Now, let’s optimize it from 2007 to 2010 (3 years) and choose the best parameters:

DayPivots 2007-2010

DayPivots 2007-2010

And see how it behaves on the longer run on unseen data from 2000 to 2013

DayPivots final 2000-2013

DayPivots final 2000-2013

Wonderful! Same drawdown, dobule profit, equal distribution of profits and this is not just a lucky coincidence because the same thing can be achieved with any other set of parameters (it survives on the long haul on unseen data).

Ok, so I will definitely keep this robot, but how about those 393 days of drawdown?

One of the most fundamental principle of trading says: “Don’t keep all eggs on the same basket!”. I strongly suggest you to read this article.

Why most commercial EAs fail?

1. They don’t really fail, your patience fails and you trashed them even if they behave as they should.

2. They fail because they are curve fitted and most of them are.

3. They fail because the backtests are not relevant (they don’t trade every signal).

4. Tick scalpers fail because of high slippage and spread, not to mention different feeds from different brokers.

5. Most vendors don’t backtest the EA properly, they use zero spread for backtest or hardcoded data to cover prolonged drawdown periods.

6. Lack of robustness, I have just shown in this article what that means.

If you like my article please share the link 🙂