How do you gain, and measure, your trading system's statistical edge?

I first read about the edge ratio in the Book Way of The Turtle. The edge that the trading system's entry provides is measured using this method by measuring how the price behaves over a certain period of time, starting from the moment the trade is taken until a fixed point in time in the future.

Basically, when the entry criteria is met the number of pips the market moves in favour of the trade is divided by the number of pips the market moves against the trade from the moment the trade is entered until the fixed point in time in the future. It is this number that gives the entry's edge ratio.

If found that 80-day Donchian channels produced an edge ratio of about 1.40 to 1 over 20 days and that this produced a system that had a ratio of around 2.10 to 1 when it came to the system's pips won to pips lost ratio. This was improved to 2.29 to 1 when used with an exit that also had an edge.

The full results are here http://www.myforexdot.org.uk/Measuri...stemsEdge.html

So what is you trading system's edge and how do you quantify it?

Well... not heard much on this, will be interesting... it's normally the search for the holy grail...

Nicely done Davidee, good to see some clear uncluttered BS free thinking. Not a holy grail guess121, just profitable systems. They do exist. Just not sold as EAs that's all.

I'll be interested to spend a few hours see if I get similar results backtested. In answer to the original question, I use two things:
1) On entire trade (using same method of exit as entry), work out maximum drawdown: maximum profit potential . Typically I can get around 0.7:1 MD:MPP.
2) a 1:1 ratio at multiple values - from around 50 to a couple of hundred pips, eg 100p TP and 100p SL. System should be profitable on aggregate at multiple values. Obviously I'm working on shorter timescales than you.

Interestingly though, just because max profit potential > max drawdown, doesn't always make for a profitable system as determined by 2); sometimes it only puts you on the right side of a trade and the TPs become another issue.

There's something missing from that calculation: the biggest drawdown in pips or better, in percentage of the account if a trader would risk 1% on every trade. Anytime I look at a profitable system is not the reward that I'm interesting in but the drawdown.
I prefer the calculation in percentage because the volatility is not constant and 100 pips today for EURUSD are not the same 100 pips one year ago.
These facts make the exact calculation of the edge very difficult. Instead of some number which is meaningless I would prefer to see the profit curve.

LE. triphop, it seems that we share the same ideas.

Very true Sauron; these are really the steps to see if what I'm looking at is statistically significant before I put it into a system. Then bring on the profit curves...
I do it this way because I've screwed up calculations too often and to try and avoid curve fitting, I like to strip it down and work out if the inputs I am using are significant or not before system modelling. Ie is it an edge or a load of cobblers.

I use this http://en.wikipedia.org/wiki/Student...-sample_t-test.
If you search though some of my posts, I bang on about it somewhere else if you are interested.

We have a winner.

What are the profits month over month.

More importantly... what was the draw down during that period?

m

Thanks for all the replies.

This is not a complete trading system, rather it is a method I'm using to quantify the edge of entries and exits.

To be successful in a zero sum game like the Forex market you must have an edge substantial enough to leave you with a profit after the costs of trading. So what I'm asking is -

1. What is your edge?
And 2. How do you know it's really an edge? How big an edge actually is it? And how do you quantify it?

And since this board is about sharing info I included a link to my website to tell you about how I quantify my edge. That article wasn't meant to form a complete trading system...

Nevertheless, I will answer the question and attach the full results.

I have tested this on the EUR/USD currency pair only - when I'm developing a system I only use one currency pair and only ever test it on multiple currency pairs after I've finished as this helps me avoid subconsciously curve fitting. I actually get a truer picture to the system's value after I test it on other pairs not involved in the process of making it, and I only do this once I have finished.

On the EUR/USD, using data from MetaTrader from the beginning of 2000 to the end of 2009, the biggest draw down is 2864 pips and the biggest winning steak is 7974 pips. This comes came from trading the following system with no stop losses -

1. When the price closes lower than it has ever closed in 80 days go short, when the price closes higher than it has ever closed in 80 days go long.

2. Keep all trades on for exactly 20 days.

This is before I try and squeeze another edge out of the exit, which is actually a lot harder to do.

As with other trend following systems, the big losses usually occur when the trend is ending. When the market is making higher highs or lower lows we're adding positions, and the longer the trend continues the more profitable the system. But the trend can't continue forever so the trades we make last, before the trend ends, will usually be losers. Giving back some of your profits is an essential part of trend following as you are basically holding on until your system gets a signal that the market has turned and finally gone against you - and you don't get this signal until after the event. Incidentally, the biggest losing streak came after the biggest winning streak...

The single biggest losing trade was over 1800 pips at the end of the winning streak, stop losses usually degrade the performance of a technical trading system, but in this case perhaps they could have been used to add an edge to the system too?

I would use a stoploss = 2 * ATR or more.
If you don't use a SL then you can't really say how profitable your system is.
Risk for example 2% per trade and adjust your lot size depending on SL. Doing this you'll see that the number of pips has actually no meaning.

If the EUR/USD is at, say 1.3500, then a win of 7974 pips is about 59% of the account and a loss of 2864 pips is a loss of about 21% pips with 1:1 leverage.

Did you run some kind of analysis to come up with the 80 day and 20 day parameters?

http://i47.tinypic.com/34or4a1.gif

A frequency distribution is how I determined the 2-Bar high low breakout had an edge. It is measuring the maximum number of pips possible. Draw down is not taken into account.

Not really, no. I'd like to point out again that the tests for the edges I mentioned above are not intended, by themselves, to form a complete trading system. The tests were there simply to prove or disprove the existence of statistically significant edge which is large enough to be tradeable.

The edge(s) are the first building blocks in my trading systems. I work on gaining an edge long before I look into money management or anything else as without an edge no trading system will ever be profitable.


Anyway, you asked about why I choose 80 days for a breakout... What I've found is this -

 

1. The longer the trading time frame the more significant the breakouts, but this is only true up to around 120 days where it starts becoming so long ago it's no longer relevant.
2. The worst time frames are the ones most commonly used, for example 20 days is probably worse than 18 days etc. 50 days and 55 days are also pretty bad as they're very common too.
3. It's only after about 60 - 70 days that the benefits of using a long term breakout become really significant.

Therefore any breakout period between about 70 days and 120 days is probably a good one to use for a long-term trend following system. So I just picked 80 out of the air for the purposed of that article.

What else could I do? I mean, I could have tested every day between 70 and 120 days on every pair and come to the conclusion that, say, 91 days was best for the EUR/USD, 102 days was best for the GBP/USD etc... But that would be 'curve fitting' and just result in great historical back-tested results that my system would be unable to duplicate in the future because I had inadvertently 'fitted' the system to closely to the past - in the markets history does repeat itself, but never exactly.

Why 20 days for an exit? Two main reasons -

Firstly, I couldn't use 2 or 3 days as 80 day channel breakouts are medium to long term systems.

And secondly, after around 15 days the edge really starts to work as that tends to be enough time for it to take effect. However the edge tends to lose it's significance after around 30 days. Look at the results of a 20 day time based exit vs a 40 day time based exit - the losses are a lot bigger but the edge barely moves so the losses start to grow closer to the size of the winnings. So 20 days because anything between 15 and 30 days will probably do.

I have a system that uses a similar method to those in the article in trending markets and another completely different method in choppy markets; the results are a lot better than the results of that test for an edge. It wins around 70% of the time too. I'll publish the complete system on this forum and here http://www.myforexdot.org.uk/FreeTec...ngSystems.html when I have the time. But in this thread I was really just looking for people to share their edges and their methods for testing the significance of those edges, so thanks for your contribution

And thank you for yours.

I find this thread very interesting, because of the use of Donchian breakouts. I use them myself, almost exactly as you do....

So after reading your blog, I tried to replicate the test with my own Metatrader. I did not get any 30,000 pips profit over 9 years.... it was more like 6,000, based on the 80 day entry if the price closed at a new high or low. The exit was 20 days.

So I scratched my head, wondering what your chart looks like in your backtests.


Anyway, I am satisfied with my own research of using Donchian successfully (I used a 40 day breakout myself) and my exits are far less complicated that what you are constructing. But I applaud your efforts... It was fun to read your research.




I'd like to add my two cents, do with it what you will:

The EURUSD has had a strong trending characteristic since it's inception. USDJPY, GBPUSD, and USDCHF do not behave the same (well, USDCHF is awfully close) so you can't expect your algorithm to work across multiple pairs. Don't beat yourself up over it if your EURUSD algorithm fails on other pairs. For example, GBPUSD is simply too choppy, and Donchian breakouts are full of false signals there.

My tests really showed 39000 pips from 2002 to 2009 without stoploss. Also 80 and 100 days were the best choises, with fewer and fewer pips for 60, 40 or 20.
EURUSD seems to be the great winner using Donchian, the other pairs behaving less succesfully. Risking 1% for every 200 pips you could have multiplied 6-7 times your account, not bad at all considering that the biggest drawdown would have been 20%. No brainer system, even my dog can understand it.
I didn't expect EURUSD to be the most trending pair. Nothing guarantees that it will remain like this but one can use all 4 majors.

Thanks for your reply tdion,

Are you sure you have back-tested exactly the same system I'm using and not a 'normal' Donchian breakout? What I mean is everytime the breakout makes a higher high or lower low do you add another position? Even when there are other positions that haven't closed due to being on for less than 20 days?

The reason I ask is I get around 6,000 pips if I enter only once when the channel is initially broken, but of course the losses are a lot smaller then too. Does your system 'pyramid', adding more lots to the open positions during a long lasting trend?

I'm using metatrader data and I've included the full results from 2000.01.03 - 2009.12.31 in the reply in post number 8 of this thread in the attached file.

The columns are, the date, the highest closing price in the previous 80 days, the lowest closing price in the previous 80 days, today's closing price, the closing price 20 days in the future, GoLongFlag - a 1 if today closed above the highest closing price in the last 80 days to signal that we put on another long position, GoShortFlag - a 1 if today closed below the lowest closing price in the last 80 day to signal that we put on another short position, and the result.

The results over this time period are 29020 pips lost and 61016.1

Can you compare my closing prices with the closing prices in your data so we can get to the bottom of this.

That's how I feel about it too, 6 times my account over 7 years is something I would be very happy with - especially at 1% for every 200 pips which is leverage of less than 1:1.

When I first started trading I started out spread betting, aiming for profits of 3% a day compounded and other ridiculous figures like that. The result was I lost my entire account, put more money in my account, then just lost it all again...

I believe one of the lessons traders have to learn is to understand what the market will and won't give you, and not to aim for anything more than that. I like systems like this because they convince me that 30% a year can be done, and I've learned to be happy with that.

So many traders aren't happy with results like that though, they want their $5000 account to become $5,000,000 in 18 months and what have you, but they can't mathematically demonstrate the existence of a statistical edge big enough to produce these gains that they are aiming for - I suspect these traders are where a large amount of Forex 'failures' come from.

So many traders aren't happy with results like that though, they want their $5000 account to become $5,000,000 in 18 months and what have you, but they can't mathematically demonstrate the existence of a statistical edge big enough to produce these gains that they are aiming for - I suspect these traders are where a large amount of Forex 'failures' come from.

19 posts and not a single mention of risk adjusted return? Pathetic what this place has become...