Position sizing high win rate systems

[sinner]

Hey guys,

One thing I have noticed with high win rate systems is that the equity curve has the tendency to look like the stockmarket, i.e. slow steady climbs up punctuated by steep rapid declines followed by slow steady climbs up.

Whereas the opposite is true for low win-rate systems, with a tendency to look like the VIX, which tend to have curves which slowly decline punctuated by large upmoves.

e.g.

curves.png

Now I have come up with my own ways to deal with low win-rate systems, not so much using position sizing but simply not participating in those systems unless conditions are good for them.

But I would like to hear thoughts and have a discussion on how to better size trades in high win-rate systems to smooth out the curve. Because it's clear to me that a lot of alpha is "wasted" regaining new highs in equity after one or two consecutive losers, and it seems to me the primary reason is equal position sizing for all trades.

One simple idea I had would be to size down based on consecutive winners of a "simulated" system which takes all trades e.g.:

0 consecutive wins: 1*R
1 consecutive wins: 0.8*R
2 consecutive wins: 0.6*R
3 consecutive wins: 0.4*R
4 consecutive wins: 0.2*R
5 consecutive wins: 0*R

obviously the above "table" would vary depending on your "projected" norm of consecutive winners, if you project to normally have 10 consecutive winners before a loser then you might size down 10% per consecutive winner.

This is probably a bad idea somehow, which is why I thought we could discuss it. But off the top of my head, I don't mind sizing down on consecutive wins because one fully sized loser might wipe out 3 or 4 wins anyway.

[tech/a]

Win rate is one equation
What sort of average win size / trade?
Is risk a % of capital or a tech stop?

Decreasing of position sizing normally occurs after a loss.
Increasing after a win.

You can also move your stops to
B/E quickly so your losses become B/E s

Just randomly thinking aloud

[DaveMac]

Win rate is one equation
What sort of average win size / trade?
Is risk a % of capital or a tech stop?

Decreasing of position sizing normally occurs after a loss.
Increasing after a win.

You can also move your stops to
B/E quickly so your losses become B/E s

Just randomly thinking aloud

Plus, even a system with a 90% win percentage will experience a 5 trade losing streak at some point. So, I think Risk should not be so high that 5 losing trades will send you broke.

[Lone Wolf]

I see what you're trying to do and I'd like to know if it works. But I'm thinking it won't.

Immediately following a losing trade you raise the risk to 1 R. You continue risking 1 R until you win, at which point you slowly reduce trade size down to zero and wait for another loss. You're effectively increasing your risk following a losing trade on the expectation of a string of winners to come. At the same time you're limiting the potential of a winning streak by reducing trade size as it goes on.

If your system predictably had a loser every 5th trade and only ever one or two in a row then it would work. But if conditions turn sour and you get a string of losers you'll be hit hard. And it'll be hard to get it back to new highs again because even if you had 20 winners in a row the positive outcome will be crippled by the system scaling down to zero after 5 wins.

Just my opinion of course.

[waza1960]

One thing I have noticed with high win rate systems is that the equity curve has the tendency to look like the stockmarket, i.e. slow steady climbs up punctuated by steep rapid declines followed by slow steady climbs up.

Whereas the opposite is true for low win-rate systems, with a tendency to look like the VIX, which tend to have curves which slowly decline punctuated by large upmoves

Which is another way of saying that high win rate systems have a trade off against a lower win/loss ratio i.e. smaller wins against bigger losses and the reverse for Low win rate systems.
This is where I have spent a lot of time improving the win/loss ratio against % profitability so none of my systems will run live unless I exceed my custom metric of 100 this being win:loss divided by % profitable trades.....
This is where trade management comes into its own (Break Evens /multiple exits/different stops) so this is where I focus to smooth out my equity curve.

One simple idea I had would be to size down based on consecutive winners of a "simulated" system which takes all trades e.g.:

0 consecutive wins: 1*R
1 consecutive wins: 0.8*R
2 consecutive wins: 0.6*R
3 consecutive wins: 0.4*R
4 consecutive wins: 0.2*R
5 consecutive wins: 0*R

My gut feeling is that you are going to forego too much profit to reduce the larger losses.Much better to find smart ways to limit losses while maintaining %profitability IMO.
You could easily do a speadsheet simulation to test.

You can also move your stops to
B/E quickly so your losses become B/E s

+1

Plus, even a system with a 90% win percentage will experience a 5 trade losing streak at some point. So, I think Risk should not be so high that 5 losing trades will send you broke

Risk is a separate issue IMO

What about a more dynamic position sizing algo (assuming say backtested 5 losses in a row)increase positions after a loss for next three trades and then reduce position size...

My negative progression code tries to gain a similar edge in a different way so like exploring the posibilities here

[CanOz]

I'm waiting for Waza to chime in on this as somehow i feel he may have visited this thought before...

LOL@Waza...i had no idea you were posting when i posted that...i swear!

[sinner]

Which is another way of saying that high win rate systems have a trade off against a lower win/loss ratio i.e. smaller wins against bigger losses and the reverse for Low win rate systems.

Yes. In terms of market structure another way of saying it is "short volatility" or "long volatility".

This is where I have spent a lot of time improving the win/loss ratio against % profitability so none of my systems will run live unless I exceed my custom metric of 100 this being win:loss divided by % profitable trades.....
This is where trade management comes into its own (Break Evens /multiple exits/different stops) so this is where I focus to smooth out my equity curve.

Good to know how you're handling it right now.

My gut feeling is that you are going to forego too much profit to reduce the larger losses.Much better to find smart ways to limit losses while maintaining %profitability IMO.
You could easily do a speadsheet simulation to test.

I do plan to do some testing very soon. What I don't like about the idea of using "smart ways to limit losses" is that they generally impact robustness, smart things don't tend to be robust.

Risk is a separate issue IMO

Yep. I am talking about sizing down from 1R, not 100% of equity.

What about a more dynamic position sizing algo (assuming say backtested 5 losses in a row)increase positions after a loss for next three trades and then reduce position size...

Yeah this is the sort of idea I was looking to discuss, but the size down pattern was the only thing I could think of off the top of my head. I've also had some small success in trying to avoid trading if vols aren't trending down but this has it's own issue around what happens when the down trend finally ends.

My negative progression code tries to gain a similar edge in a different way so like exploring the posibilities here

Glad to know I'm not the only one.

[sinner]

If your system predictably had a loser every 5th trade and only ever one or two in a row then it would work. But if conditions turn sour and you get a string of losers you'll be hit hard.

Assume that the system has an overlay that avoids this situation, I am really trying to deal with those 1 or 2 consecutive losers in this case, even if the underlying has a string of 10.

And it'll be hard to get it back to new highs again because even if you had 20 winners in a row the positive outcome will be crippled by the system scaling down to zero after 5 wins.

Just my opinion of course.

Yeah I also saw this as a potential issue but figure this 'worst case' scenario just means the system is sitting in cash, waiting for a loser to happen in the simulation, right? That seems alright to me.

[sinner]

Decreasing of position sizing normally occurs after a loss.
Increasing after a win.

If I bet 100% of equity, then I'm naturally decreasing size after a loss and increasing on wins. This is the exact sizing I'm trying to avoid (although it works well for low win rate systems).

You can also move your stops to
B/E quickly so your losses become B/E s

Just randomly thinking aloud

Well that's two people who've suggested it now (you and waza) so I'll add it to the list.

Thanks guys!

[peter2]

I agree with the trade management modification. Make the losses smaller and get the trade to BE asap. If the trade doesn't go in your favour immediately limit the loss.

If your system has a high win rate and above average expectancy then you have more position sizing options available. I suggest you investigate the distribution of your R multiple results to find out if the winners are grouped. If so then take advantage of this by risking more after a winner. A common model adjusts the risk (pos sizing) using a set % of core capital and adds a % of the profits. This aggressive compounding works well if your winners are grouped and the W% >55%.

Once you use an aggressive pos sizing model you must do something to minimise the max DD. Review your R-multiple distribution and identify the average R DD. It might be ~4. Your main goal is to limit the DD caused by a larger losing sequence. Once your DD is worse than -4R, reduce risk to minimum % without adding any profits. Stay at minimum risk size until your DD is again within "normal" parameters. You can then resume adding a % of profits to your risk.

This R multiple brake prevents a large DD that might otherwise take you out and has minimum lag in getting back on track.

[howardbandy]

Greetings --

The answer is much more complex than any of the posts have suggested. Formulas and rules of thumb do not give accurate, safe-to-trade answers.

My book, "Modeling Trading System Performance," discusses this in detail. It includes an Excel add-in that runs Monte Carlo Simulations so that you can see the distribution of profit and distribution of drawdown, and calculate the position size that maximizes account growth while keeping maximum drawdown to a level in accord with your personal risk tolerance.

The data can be anything -- actual trades, paper trades, backtest results, hypothetical distributions. It is completely independent of the trading system or the development platform. All it needs is the trade list, or a probability distribution representing the trade list. Details on how to format the input are explained in the book.

Go to the book's website and read the free chapters. You will get an idea of the process.
http://www.modelingtradingsystemperformance.com/

For more about the design, testing, validation, and analysis of trading systems, go to my website and read the papers under the menu "Articles." It is all free.
http://www.blueowlpress.com/WordPress/

Best regards,
Howard

[tech/a]

Understood
But I thought he was looking for ideas
To test?

[skc]

Sinner, I run a pretty high win rate system. Over ~1500 pairs trades the average win rate was ~80%. Obviously I go through patches of draw down, but the drawdown are more than manageable and hasn't prompt me to think of alternate position sizing methods to reduce those. To me, if a system works and still has plenty of scalability, then the more profitable action is to focus and learn about increasing size, as opposed to optimise the equity curve. But that's just my

I will try to run some analysis over the next few days if I have time, but on first glance your suggested variation in position size after consecutive wins isn't going to be productive (just a gut feel). I have seen big runs in my equity curve that happens over 10, 15, 20 consecutive wins. I certainly wouldn't sacrifice those for the sake of reducing drawdown. Without those runs, there won't be much to draw down!

Another note... the longer I trade the more I realise - the secret of a high win rate system is all in the stop placement. You can't play around with stops without affecting win rate. Moving stops to breakeven will result in more losing trades (assuming it's a loss after brokerage costs) and smaller average wins - there's no way around it.

Next observation is that my drawdowns are never from consecutive losses. The notion of consecutive losses seem to get banished around a bit - but I bet you that most systems are losing because of overall negative expectancy rather than consecutive losses. Think playing blackjack against the casino - I can spend all night without losing more than 3 hands in a row but still walk out of the casino with empty pockets.

Note that my trading, while system-based, has plenty of discretion in taking the entries and exits. Also, each of my individual trades are essentially non-correlated - compared that to a long only system where trade performance is highly dependent on overall market direction. One would better off spending time looking at a macro system on-off switch in that case.

[sinner]

Guys, I did some quick tests last night on a few of my 'generic alpha' systems with high win-rates. These aren't "real" systems but rather modules which plug together to form a proper system.

I wanted to test the idea of moving stops to B/E ASAP, because it's pretty easy to test, pretty easy to run an adaptive overlay on, etc.

Findings: Essentially, this technique is not very useful for high-win rate systems. Because? Because in the cases where you can move a "stop" to B/E, we are generally already talking about a profitable trade. The losers on the other hand never even give you a chance to move stops to B/E or only give a very brief chance. So the impact on the equity curve is minimal. Some slight overperformance on some instruments and slight underperformance on others.

Another note... the longer I trade the more I realise - the secret of a high win rate system is all in the stop placement. You can't play around with stops without affecting win rate. Moving stops to breakeven will result in more losing trades (assuming it's a loss after brokerage costs) and smaller average wins - there's no way around it.

Yep, bigtime! One thing that helped me "think through" this concept was the use of options. What if I "buy" my stops in advance, as soon as the trade is placed? Well then you can see stops *cost*. In the end it doesn't matter how you count them, stops cost money. So if you use them they need to "pay for themselves" in either equity or at the very least significant vol reduction.

I will try to run some analysis over the next few days if I have time, but on first glance your suggested variation in position size after consecutive wins isn't going to be productive (just a gut feel). I have seen big runs in my equity curve that happens over 10, 15, 20 consecutive wins. I certainly wouldn't sacrifice those for the sake of reducing drawdown. Without those runs, there won't be much to draw down!

Would really appreciate that! Like I said, I posted the idea mainly because I figured it was probably a bad idea one way or another. But I'm looking for a generalised solution, not married to this idea. In my case it's pretty unlikely for this system to see either a very long string of losers OR winners.

My book, "Modeling Trading System Performance," discusses this in detail. It includes an Excel add-in that runs Monte Carlo Simulations so that you can see the distribution of profit and distribution of drawdown, and calculate the position size that maximizes account growth while keeping maximum drawdown to a level in accord with your personal risk tolerance.

I guess this is another (maybe the only?) way to deal with the issue.

Once you use an aggressive pos sizing model you must do something to minimise the max DD. Review your R-multiple distribution and identify the average R DD. It might be ~4. Your main goal is to limit the DD caused by a larger losing sequence. Once your DD is worse than -4R, reduce risk to minimum % without adding any profits. Stay at minimum risk size until your DD is again within "normal" parameters. You can then resume adding a % of profits to your risk.

I kind of like this idea even though it's quite complicated in the scheme of things.

[skc]

Would really appreciate that! Like I said, I posted the idea mainly because I figured it was probably a bad idea one way or another. But I'm looking for a generalised solution, not married to this idea. In my case it's pretty unlikely for this system to see either a very long string of losers OR winners.

Ok here are the numbers.

Raw data

310 trades, 250 wins (80.6%), avg win = 0.3366, avg loss = -0.5592. Net profit +50.60.

Adjusted as per your original posts

195 trades (115 trades were adjusted to 0*R - i.e. not traded), 155 wins (79.5%), avg win = 0.2257, avg loss = -0.4103. Net profit +18.57.

Total loss averted = -17.14. Total gains forgone = 49.18

PT equity curve with size adjustment.JPG

So overall not a great result by doing the adjustments. May be it can be tailored a little bit more, but if the adjustment becomes more smooth (say reduce from 1R to 0*R over 10 consecutive wins), the chance of it making an impact on the loss is also reduced.

[sinner]

You're an absolute legend skc!

Sorry for the belated reply, started working on a pretty good new contract recently that's been taking up all of my time.