Robustness in trading systems

Hi, I thought this will be a pretty interesting discussion.

For the last few months I asked myself, if it is desirable to have a system with a high win/loss ratio or to have a system with a high R:R. As far as my understanding goes, those two paramater are mutually exclusive, meaning you can either have a high win/loss or a high R:R, not both.

I was playing around with some "positive-expectancy" systems and started optimizing values to the point of overoptimization. That happened with recent half a year of data.
Now these optimized values were then applied to 10 years of data to be able to see the robustness of the system.
In short, the results were the following: If the optimized values favored a high win/loss ratio on half a year data, this high win/loss got completely eradicated when applied to 10 years of data, but still stayed positive in terms of net pips accumulated.
However, when the optimized values favored a high R:R on half a year data, the R:R only slightly changed, when applied to 10 years of data. Obviously, I was always much more positive in terms of pips, than with the high win/loss values.

What does all this mean? I can't talk for all systems out there, but I think most probabilities change over time. Take anything from PA & S/R to indicators, and that is also the reason why over the long-term most systems fail, cause of changing probabilities in the market.
However, if you were always trying to achieve high R:R with low win-loss, the performance of your account won't be dramatically altered, when the probability of a pattern changes.

That is why I came to the conclusion that it is not very desirable to search for high-probability, but rather for high R:R in trading systems. Of course, here we only talk about positive expectancy systems - systems with a statistical edge.

Any comments are welcomed!

I actually have it the other away around.
I've been playing with the searching for high reward vs risk and got obliterated in backtest pre 2008 vs now. This is why I always like to go for high probability because no matter how the function changes the the added losses won't drop significant below the profit threshold. I like to compare the "distance" to the breakeven level as a good indication of a good system... the further you are the more pounces you can take by the changing market.

Also I've found that the higher the R:R comes the more drawndown you will experience.

hm, interesting comment.
Obviously you are right with the less drawdown, when the win/loss stays the same, compared to a high R:R system.

Well, what I found interesting is the change of probabilities over time.
When market conditions change to a point that the probability of a pattern drops significantly as well, then the following happens:
if we have traded a high win/loss system, the number of losses will increase significantly. The previously tested out probabilities of the system will not work out at all anymore in the current market.
however, if we have traded a high R:R system, the number of losses will increase as well, but since it's a high R:R, our losses are comparably smaller and thus not affect the bottom line & reliability of the system as much.
(I hope my logic is sound here. If not, then please correct me.)

Of course, the premise here is that probabilities change over time. If they stay constant, then a high win/loss system is definitely favorable.

Now question is, if there are edges whose probabilities will not change over time.

But you have good arguments for your point of view and it is interesting that you found the opposite to be true.

I do believe that the expectancy is actually the result of both R:R and the win/loss ratio...Of course it takes the actual R:R (Average Winner and Average Loser) as it happened, not your theoretical R:R....

There is nothing wrong with having a high R:R as well as a high win/loss as well, it's just very rare but it doesn't mean that it's not there...

Imagine a classic trend following system...Typically it will have a high R:R and a low win/loss (Losers much more than winners)...Now, assume that you search for filters to reduce the losers a little bit without having any effect on the winners whatsoever...IF you manage to find such a filter, you will effectively enhance the win/loss ration of the strategy, though you might get a slightly worse R:R, but it will stay high (Something from 8:1 to 6.8:1 for example)...

The most important thing as you mentioned is that probabilities, or in other words, market conditions change over time...

If your system or strategy is robust enough, you would expect a minimum effect of market conditions changing on the expectancy, that is, the strategy WILL still remain profitable (Because of maintaining the positive expectancy) despite the changes in market conditions or in probabilities....


Regards,

Nader

yeah, I was actually referring to actual R:R as well.
Interesting theory about the filters, however, I haven't been able to find a filter that doesn't also reduce the winners as well. Of course, my search is far from over

thank you for your comment!

Hey Custos,

You may want to read this http://www.automated-trading-system....obust-systems/

Cheers,
-soso

thanks for the article, quite intriguing!

Well, you won't find a filter improving the win/loss while keeping the R:R the same....

All you can find, although rare, is a filter that can improve the win/loss more than it worsens the R:R so that the total impact would be slightly to your favor..


Regards,

Nader

exactly, that's also what I've been thinking.

Goes without thinking


Regards,

Nader

Looking to expand:

http://img441.imageshack.us/img441/2737/76569208.png

X - is the profitability of the trades (number of wining trades/total number of trades)
r - is the reward/risk ratio

- the graph shows that to expand the profit factor Pf by the same amount we need to up the profitability by y or reward/risk ratio by x
Since y<x it's shown that the better solution is to go for a bigger profitability then reward/risk ratio if one wants to make more money

It's shown that the connection for profit factor Pf and profitability P is:
http://img534.imageshack.us/img534/987/ppp1v.png

that is while increase in the ratio r will increase profit factor by a proportional (linear) increase, increase in profitability P of your trades will increase substantially the profit factor Pf

maybe not the robustness in trading systems but still something to remember

A bunch of graphs and a conclusion doesn't make the conclusion true.

On top of it there's a bunch of PROVEN fund managers/traders that say that the way to go is to look to increase the average profit instead of the winners %.
And it makes sense since the markets tend to trend (sometimes).

This is just a expand of the http://www.priceactionlab.com/Litera...fitability.pdf. You are welcome to calculate and see it for yourself. I'm sorry I was thinking this link was the one that was previously posted; I was referencing to this, sry for confusion. I wasn't saying what was easier to get: profitability or reward/risk or how the market behavior should be used in your advantage.

Your other sentience...
And they tend to range (most of the time).

Fair enough.

But range or trend, trying to skew the winning % in your favor is the same hopeless thing - trying to predict the market direction. I don't think you'll get too far, at least over the long term.

Skewing the average profit in your favor is done by the trend followers and is real. Of course it comes with a downside, equity volatility, but then who says that trading should be easy.

hm, the calculation is interesting, but I think the premise is wrong. It is based on probability staying the same over time. Then of course it would make sense to try to tune up the winning rate.
But once probability changes, so does the winning rate.

However, it is interesting in terms of profitability at one point in time. I guess taken over a longer period with changing probabilities, the R:R would be favorable. But that's just my own thought right now.

Also another thing: once we have found an edge, it is difficult to "tune" it up to get more winnings than the edge can provide by itself. However, it is easier to tune up the R:R and try finding optimal parameters to allow the highest profitability possible from the edge.

Just my own thoughts here.

Sorry to say this, but the terms used in the paper like Profitability, winning trades, losing trades, amount of winning trades, amount of losing trades, etc...are a little bit confusing...!!!

Anyone agree with me or is it just me?


Regards,

Nader

Probably. I was looking as how there is a lot of MM ~ HFT that is basically shifting the balance from R:R ratio to the profitability. Going for 85%+ but with 1:2 reward:risk because of the volatility. There is a stat. merit for doing that. I do see that you can't get far with upping the "tune" trades you have without sacrificing the trade number. And that is more counterproductive then to try to modify your R:R to get the optimum.

I think we need to go back to the building blocks…


What does a trading strategy give you in terms of inputs? I believe the following:

- When to enter? Entry rules
- How much to buy or sell? Position Size
- When to exit? Exit rules (Both stops and profit takes)
- Tactics or Trade Management (Like pyramiding, adding to your position, partial profit taking, trailing stops, etc…)

Now, let’s say that you do have all those rules or building blocks…You apply them and backtest and get the results….What will the results look like?

The results will show you the following, among hundreds of other statistics:


1- Total number of trades
2- Total number winning trades
3- Total number of losing trades
4- Highest Profit in dollar terms (The trade which produced the highest profit in dollar terms)
5- Highest Loss in dollar terms (The trade which produced the highest loss in dollar terms)
6- Average Profit per trade (Sum of profits generated by all winning trades divided by their number)
7- Average Loss per trade (Sum of losses generated by all losing trades divided by their number)

So, what are the win/loss ratio and the Risk/Reward ratio? How can we calculate them from the above results…??


The win/loss ratio refers simply to the ratio between the number of winning (Profitable) trades and the number of losing (non-profitable) trades…Simple…So that should be no. 2 divided by no. 3…


Numerically, if the test results produced 120 profitable trades and 360 losing trades, the win/loss ratio is simply 120/360 = 0.33333….


The Risk/Reward ratio is the measure of your system average risk versus it’s average return…So that should be no. 7 divided by no. 6….


Numerically, if the test results produced a $5,000 average profit per trade and a $1,000 average loss per trade, the Risk/Reward ratio is $1,000/$5,000 = 1:5 or 0.2…This means that potentially, for every dollar you risk on average, you are expected to profit 5 dollars on average…


In the paper attached by Rikers which is written by Michael Harris, Harris defines something called the profit factor, which is the ratio between “The amount of winning trades” to “The amount of losing trades”….It’s not really clear what he means by “The amount” here, but I assume it’s the total profit in dollar terms and the total losses in dollar terms…He calls this ratio the “Profit Factor”….


In our example, 120 winning trades and an average of $5,000 of profit per trade means that total profits is 120 x $5,000 = $600,000….360 losing trades and an average of $1,000 loss per trade means that total losses is 360 x $1,000 = $360,000…


The profit factor according to this is $600,000 / $360,000 = 1.6666….


So as we can see, the profit factor is a completely different measure which looks at the overall monetary profitability of the system….If the profit factor is greater than 1, the system is profitable, if it’s less than 1, the system is not profitable…Pretty a direct way, although it is usually obvious when a system is profitable and when it’s not…


Now to the interesting part, what happens when we change one of the system components?



Let’s assume that we slightly change the entry rules while leaving everything the same…What will be the impact on the results? Which metrics are going to be affected and which not?


In fact, all the test results might change….You might get better results or worse, you never know…If you changed the entry to be a more sensitive one, you might get more signals, consequently, the number of trades will increase and thus the highest profit and the highest loss will change and also the average profit and loss will change…The total profit and the total loss of the strategy will change as well…(If the strategy is of positive expectancy, it might still come ahead unless you changed the entry rules in a way destroying the edge)…



If you kept everything the same and only changed the exit rules (Stop losses and Profit takes), you will have a different impact on the test results this time…


The total number of trades WILL stay the same since the entry rules remained the same…However, since the exit rules changed, the winners might no longer be 120 and the losers might no longer be 360…You are now messing with the win/loss ratio…


The highest profit and loss as well as the average profit per trade and the average loss per trade will also change accordingly…




Keeping the entry and the exit rules the same and changing the position sizing will even yield some interesting stuff…The total number of trades WILL stay the same again, the total number of winners WILL stay the same as well at 120 and the total number of losers WILL stay the same at 360…What we have here is an impact on profits and losses (And other measures like the equity curve among others)….You are now messing with the Risk/Reward ratio…


The highest profit and loss will change…The average profit per trade and the average loss per trade will change accordingly of course…



You can go ahead and change multiple variables at the same time, like, changing the exit rules and changing the position sizing at the same time…So you are altering both the win/loss ratio as well as the Risk/Reward ratio…It can be a bless and it can be a nightmare…It might not have an effect whatsoever on both, but will have an effect on drawdown and equity curve…You won’t be able to tell unless you do the changes and test and then compare the results…



Conclusion:


Changing the exit rules of a strategy (All other factors being the same) directly and only impacts the win/loss ratio…
Changing the position sizing rules of a strategy directly (All other factors beign the same) directly and only impacts the Risk/Reward ratio…

So, if you have a basic strategy showing a very good win/loss ratio relative to it’s logic (A trend following strategy with a 0.4 or 0.5 win/loss ratio for example), KEEP IT like that and DO NOT mess with the entry or exit rules…If you want to enhance this strategy performance, play with the position sizing rules and try to optimize it to the better….

While if you have a basic strategy showing a very good Risk/Reward ratio, something like 1:10 or 1:15, KEEP IT like that and DO NOT mess with the position sizing rules…If you want to enhance it’s performance, play with the exit rules and try to optimize it to the better….

Sorry for the long post….Opinions much appreciated….

Regards,


Nader

You don't know what winning trades means? Not even losing trades? Are you maybe in the wrong forum?

What else will I read visiting these forums...God help us...

I still think you should find an optimum, which allows you to profit as much as possible while keeping the strategy as robust as possible.