Ways to reduce drawdown?

[Sir Burr]

Hi,

Been testing a new system and trying to reduce the drawdown on it.

I'm using a fixed $ position sizing with a fixed capital base (no pyramiding of profits). I have tested % risk position sizing and does not perform nearly as well.

I've found that increasing the number of trades taken will reduce drawdown (and profits) plus adding a profit % exit and a max loss % exit helps.

I set the % max loss based on removing the worst trades by maximum adverse excursion (MAE).

Tried an ATR stop but the above % stop seems better plus easier to impliment so removed it.

Any other general ideas on reducing drawdown please? Anything at all like using the equity curve or index filters etc etc

Cheers SB

 

[MS+Tradesim]

Relative strength comparison helped reduce drawdown significantly in my mechanical system. I use Jose Silva's URSC kit.

When you say you tried an ATR stop, how many different values did you model? eg. 2xatr(10), 2xatr(21), 3xatr(5)?? Both multiplier and the number of bars you use can make a large difference.

FWIW, I don't use the closing price to base exits on. Try different things which seem counter-intuitive. I found some interesting and pleasant surprises by avoiding the path most often travelled.

 

[howardbandy]

Hi,

Been testing a new system and trying to reduce the drawdown on it.

I've found that increasing the number of trades taken will reduce drawdown (and profits) plus adding a profit % exit and a max loss % exit helps.

I set the % max loss based on removing the worst trades by maximum adverse excursion (MAE).

Any other general ideas on reducing drawdown please? Anything at all like using the equity curve or index filters etc etc

Cheers SB

Hi Sir Burr --

I'll comment on the topics I've left in the quote above.

Drawdowns tend to be proportional to the square root of the time a position is held. Shortening the holding period will reduce the drawdowns. If you have a copy of Quantitative Trading Systems, look at page 290 for some charts that show drawdowns over time.

Adding a profit target will reduce the drawdown, providing your system is a short term system. (If your system is a long term system, then profit targets will hurt your performance.) In this, I am assuming that your system has both a signal exit and a profit target exit. Every exit coded into the trading system must occur often enough so that it does not fall into the "cherry picking" category. This is usually not a problem with profit targets, since they are usually set at 1 or 2 percent. But it is often a problem with maximum loss exits which are set to avoid specific conditions observed in prior crashes. So, be careful with the MAE exits.

Trading the equity curve -- Usually this means calculating the daily equity or closed trade equity and either exiting an existing trade or blocking entry to new trades. A "shadow" equity curve is often calculated and used to compare performance with and without equity curve filtering.

In order to successfully trade the equity curve, there must be a serial correlation -- either positive or negative -- between daily equity (if you are using daily equity) or trade profit (if you are using closed trades). In Quantitative Trading Systems, look at page 239.

If there is a positive serial correlation, then wins tend to be followed by wins and losses by losses. So the technique is to stop trading after one or more losses and wait until a win (that you did not take, but did watch) to begin again.

If there is a negative serial correlation, then stop trading after one or more wins and wait until there have been one or more losses to begin again.

Thanks,
Howard

 

[Sir Burr]

Relative strength comparison helped reduce drawdown significantly in my mechanical system. I use Jose Silva's URSC kit.

When you say you tried an ATR stop, how many different values did you model? eg. 2xatr(10), 2xatr(21), 3xatr(5)?? Both multiplier and the number of bars you use can make a large difference.

FWIW, I don't use the closing price to base exits on. Try different things which seem counter-intuitive. I found some interesting and pleasant surprises by avoiding the path most often travelled.

M+T,

Yes I've tried numerous ATR's but will have another look at that, thankyou.

"FWIW" Counter-intuitive sounds very interesting! I enter using a set price and exit with an extremly wide % stop and % profit plus timed. Can you give an example other than using open or closing price? Do you mean basing it on a set intraday price from an indicator? I'm talking here about daily or weekly systems.

SB

 

[Sir Burr]

Thankyou Howard, I was hoping you may post too.

If you have a copy of Quantitative Trading Systems, look at page 290 for some charts that show drawdowns over time.

Interesting charts on page 293, mine is it's a shorter term system 1-4 weeks.

Every exit coded into the trading system must occur often enough so that it does not fall into the "cherry picking" category. This is usually not a problem with profit targets, since they are usually set at 1 or 2 percent.

Wow 1 or 2 percent, is that of capital? I'm talking about a ~75% profit exit on the stocks. Profit exit is rare, about 10% of trades (# of trades = 2000) and max loss 5% of the trades.

I've just had a quick look at page 239 about "Equity Curve Feedback" and will test for positive or negative serial correlation. Also, shouldn't be too hard to test by removing the trades (if that's the way to go) in Tradesim.

Thanks again
SB

 

[bingk6]

Hi SB,

FWIW, just a couple of suggestions.

1) Design your system to be be a shorter term system (Howard already mentioned).

2) Incorporate into the system a profit taking exit that is based around a rapid increase in pricing in a short space of time. Eg I use the ROC indicator. At the time in which the entry is initiated, you note down the value of ROC(C,100) = x say. Then your exit could be triggered when ROC(C,100) = x + 50. The benefit of this exit is that the exit will be initiated only after a sufficiently rapid increase in prices in a short space of time (you can tweak the ROC parameters to suit your requirements), without taking out your longer term slower trending trades. The other thing also is that you are not limiting yourself to a fixed % profit as well. You're trying to capture the "irrational exhuberance" with this exit. With Amibroker, this form of dynamic exit can be easily coded and backtested.

3) I use the MAE and MFE scatter charts in Excel to look for areas to place profit targets as well as stop loss settings. However, I caution placing too much emphasis on it simply because it reeks too much of curve fitting.

4) I recommend including a Amibroker custom metric in your reports to indicate the type of exit that terminated the transaction (profit target, system exit, trailing SL etc etc) just so that you are well aware as to how often each of the different types of exits have been triggered. If the ratios don't look right, the remedial actions can be obvious.

 

[happytrader]

Hi Sir Burr

If you are trading short term you need to 'make your money when you buy' To do this you need to be extremely specific with your entries. Intraday charts are useful for this.

Cheers
Happytrader

 

[Sir Burr]

Appreciate your points Bink - first thing I'll do is setup Amibroker custom metrics (never tried that!).

If you are trading short term you need to 'make your money when you buy' To do this you need to be extremely specific with your entries. Intraday charts are useful for this.

Yes I found that too, taking buy signals at specific prices improves my system quite a bit.

Cheers SB

 

[nizar]

Some interesting ideas on this thread, thanks for starting it SB.

Does anybody know how to determine the serial correlation of your trade profit and equity curve using MS/TradeSim ?

The ROC exit idea is an interesting one I need to investigate as well.

At the moment, I only use ROC as part of my entry.

 

[Sir Burr]

In order to successfully trade the equity curve, there must be a serial correlation -- either positive or negative -- between daily equity (if you are using daily equity) or trade profit (if you are using closed trades). In Quantitative Trading Systems, look at page 239.

Hi Howard,

Sorry I'm a bit thick! Would you mind explaining how to use that code starting on page 341 (plotruntest.afl) please? I think that is why I didn't go through that the first time I read through that section. Wasn't sure if it was run with system code (don't think so) or used with an equity curve. Can it be used with weekly data?

Thanks SB

 

[Temjin]

If there is a positive serial correlation, then wins tend to be followed by wins and losses by losses. So the technique is to stop trading after one or more losses and wait until a win (that you did not take, but did watch) to begin again.

If there is a negative serial correlation, then stop trading after one or more wins and wait until there have been one or more losses to begin again.

Thanks,
Howard

While I'm not sure about the negative serial correlation part, I have observed the phenoman in the positive serial correlation one.

One known strategy is to increase position size when you have winning streaks trades up to a certain amount. Like 1% to 1.5% to 2.0%. Likewise, you decrease your position size as soon as you have reached a predetermined losing streaks and maintain that position size until you have made a winning trade again. i.e. 1% to 0.8% to 0.5% to 0.5% and then back to 1% when you win the previous one.

When I have more spare times, I will do a Matlab simulation on such position sizing strategies and see if it works out better than fixed % position sizing.

 

[CFD]

This doesn't work as you might think. What happens is when you have a loss (after a winning streak) you have the largest amount on it and when you have a win (after the losing streak) you have the smallest amount on it.

 

[nizar]

One known strategy is to increase position size when you have winning streaks trades up to a certain amount. Like 1% to 1.5% to 2.0%. Likewise, you decrease your position size as soon as you have reached a predetermined losing streaks and maintain that position size until you have made a winning trade again. i.e. 1% to 0.8% to 0.5% to 0.5% and then back to 1% when you win the previous one.

How do you backtest using strategies?
I guess you could call it non-fixed percent risk.
I don't think TradeSim can do this.

 

[Temjin]

This doesn't work as you might think. What happens is when you have a loss (after a winning streak) you have the largest amount on it and when you have a win (after the losing streak) you have the smallest amount on it.

That is why there is a limit on the highest and lowest position sizing during a winning or losing streak. Like no more than 2% on winning streaks and 0.5% on losing streaks. Just an example.

How do you backtest using strategies?
I guess you could call it non-fixed percent risk.
I don't think TradeSim can do this.

It is possible with any softwares that allow custom coding. Metatrader, Amibroker, etc.

As for Matlab, it is just simply a simulation and not necessary backtesting an actual system. Yeah, you can call it non-fixed percent risk. I know a fairly successful private managed future (though only 2 years of track record) that utilise this strategy, though obviously, all their risk management are "propertiary". hehe It specifically mention the gradual reduction of position sizes as a specific drawdown has been reached. Not sure about increasing position size for winning streaks though.

 

[howardbandy]

This doesn't work as you might think. What happens is when you have a loss (after a winning streak) you have the largest amount on it and when you have a win (after the losing streak) you have the smallest amount on it.

Hi CFD --

What doesn't work as you think it might?

Howard

 

[>Apocalypto<]

Don't Trade!

 

[julius]

At ~75% p/trade & 1 - 4 weeks holding it sounds like a breakout system? What is the % trades profitable?

Assuming somewhere around 50% it takes a long time for any signs of auto correlation to become statistically significant - if you turn the system off prematurely you can/will miss a large portion of annual profit. Over these time frames I think this kind of consideration could start to undermine your confidence in the system.

It might be worth searching for the market conditions that lead to system drawdowns. ie. over night volatility etc.

Profit targets can improve/decrease performance by changing the sequence of the signals traded, similiar to a monte carlo analysis, just something I'v observed.

 

[Synergy]

I've thought about using a % gain as an exit but I find that the % gain doesn't really relate to anything other than my entry, which varies significantly from any baseline average (breakout system). I had a fiddle with a % above baseline exit, but didn't have much luck. To me it seems more logical that a straight % gain exit.

 

[MS+Tradesim]

M+T,

Yes I've tried numerous ATR's but will have another look at that, thankyou.

"FWIW" Counter-intuitive sounds very interesting! I enter using a set price and exit with an extremly wide % stop and % profit plus timed. Can you give an example other than using open or closing price? Do you mean basing it on a set intraday price from an indicator? I'm talking here about daily or weekly systems.

SB

My mechanical system uses exits based on median price. I have not found this approach anywhere though no doubt if anyone has found it as profitable as I have then they may also be keeping it close to their chest. It's certainly not a catch-all as I've now turned my system off until the markets start displaying some more stability.