The guide to forex profitability

forex profitability

The holy grail doesn’t exist but nobody can stop us searching for it, right? This blog exists just because so many systems are failing right as we speak. What makes a system a good system? Why do they fail? I’ve been searching for these answers for a long time but only since last year I’ve dedicated a lot of my spare time to investigate the market behavior. My scientific background (I have a degree in computer science and a lot of other math and statistic courses) helped me a lot in my quest. What I’m about to say can make the difference between trading blind and trading smart. I guarantee you that if you have the patience to read the article completely you’ll be rewarded at the end.

The questions

1. Why so many systems are failing?

They fail because they are not designed for real trading, they are designed to make money. For the vendor only. Let’s summarize things a little bit:

a. They are curve fitted to look good. In respect for profitability rules, the robot should be optimized for a sample period, for example 2007-2014, then it should be validated against an unseen period 2000-2007. If the relative drawdown, maximum historical drawdown and drawdown length are the same for unseen period, then the strategy is valid and reliable. It means that the robot performs exactly the same on unseen data.

But most vendor don’t do that. They optimize it for the whole period 2000-2014. The relative drawdown, maximum historical drawdown and drawdown length are much smaller, but the strategy is not validated! That’s why many forex robots are failing right after release. Because they are curve fitted.

b. Drawdown length is usually high if relative drawdown and maximum historical drawdown are small. Are you willing to bear 2 or more years without profit? I was, but not anymore. I rather have a valid system with larger drawdown but faster recovery period  then one with a very hard to bear drawdown length but smaller drawdown in equity.

2. What exactly  is drawdown?

Drawdown is nothing but a deviation from a market pattern. Patterns are hidden in the market, covered with noisy, apparently random movements. If everything would be just a coin toss then no system (automated or manual) survives longer than a few months. There are many systems surviving for years, so the coin toss approach is simply not true.

You can’t say nothing about the outcome of a coin toss, but you can definitely say a couple of things about market behavior:

a. Once it reaches a high or a low, it retraces. The noise = we don’t know when the high or low are reached.

b. Any level will eventually get broken. The price doesn’t stay long between bollinger bands. The noise = We don’t know when a level gets broken. We can’t distinguish fake moves from real ones.

c. Technical analysis works just because we think it works and behave accordingly. This is called self fulfilling prophecy. For example, the price reaches a level we consider to be the highest high. If many traders place short trades then the price goes down a little bit.

Ok, so drawdown is caused by the noise. The market is not deterministic in nature, but it’s definitely not random. Only quantum systems are truly random.

So, any system works well as long as the market conditions are in its favor, then it enters into a drawdown period or crashes completely.

3. Design flaws.

Wouldn’t you buy a forex robot which has a consistent winning streak of 30 trades and only 2 losing trades in a row? Most newbies are ready to bite it, but things are too good to be true.

If take profit is 10 pips and stop loss is 150 pips, is such a forex robot worthy? After 30 wins in a row, 2 loses erase all the previous profit leaving you at breakeven.

Exactly the same happens with small stop loss and large take profit systems, such as breakout systems. You win 100 pips on a single trade, then lose everything on a 10 losing trades, 10 pips stop loss.

The most convenient way is to design a system (and validate it) with take profit being approximately equal to stop loss.

4. How can we succeed?

I’m saying it once more: the holy grail doesn’t exist and nothing is certain in forex, but we could use statistical laws to have an edge against the market.

Let’s analyze a good system for example: it is optimized from 2000 to 2007 and tested against 2007-2014. The relative drawdown, maximum drawdown and drawdown length don’t change when tested against unseen 2007-2014 data, the profits are equally distributed so the system should be considered valid.

Stop loss is 41 pips, take profit is 45 pips.
Consecutive losses: 6
Consecutive wins: 9
Drawdown length: 300 days.

Is this a good system or a bad one? Well, I’m not winning to bear 300 days or drawdown, that’s for sure. But should I discard this system? No way.

I trust it to be valid because it passed my validity test (it behaves the same on unseen data as on optimizing period).

Here it its summary:

GBPUSD system report

GBPUSD system report

 

GBPUSD system backtest

GBPUSD system backtest

 

The equity curve is nice, but not very appealing. And still, this system is great. Why? Because with the right money management I can make it very profitable, I can dramatically reduce drawdown length from 300 days to just 45 days and and enjoy a 2% monthly.

How?

Stop loss is 41 pips, take profit is 45 pips.
Consecutive losses: 6

This is all I’m interested in: I’m using some sort of fibonacci recovery:

1 lot – lost
2 lots – lost
3 lots – lost
5 lots – lost
8 lots – lost
13 lots – lost
21 lots – won – get back everything and I’m at breakeven, because I trust the validity of the EA of having no more than 6 losing trades in a row on both periods: optimizing and unseen data period.

But what if the market changes and there will be more than 6 trades in a row? Statistically, it not that likely, IF THE STRATEGY IS VALID AND NOT CURVE FITTED, but it could happen.

I still apply this recovery method  but only after I lose 3 trades. This way I increase the number of trades I’m willing to lose to 9 and still have a nice and fast recovery without losing my account.

1 lot – lost
1 lot – lost
1 lot – lost
1 lot – lost
2 lots – lost
3 lots – lost
5 lots – lost
8 lots – lost
13 lots – lost
21 lots – won – get back.

GBPUSD system recovery on

GBPUSD system recovery on

This is the same EA, with recovery on. The equity curve looks much much better, right? The drawdown only increased by 300 pips and the drawdown length was reduced from 300 days to 45 days.

5. What’s the catch?

The catch is to design such valid systems. I stated before that artificial intelligence can’t be used for trading because there is too much noise in the market and neural networks can’t identify patterns.

This is true, but I’m telling you one thing: I’m not using neural networks and bayesian filters to separate patterns from the noise, I’m using them to identify the noise and not to trade during a noisy period! This approach is the most successful I could find.

Stay tunned for my incoming live account. I’m going to put these systems on a few live accounts.