Trading the Equity Curve

This is what I call a circuit breaker. It will hurt performance – always – but needed to reduce the risk of system failure.

I have never been able to get equity curve trading to work without seriously impacting performance. Plus, it’s another rule that impacts degrees of freedom along with additional lines of code that need to be maintained.

I think that it’s a gimmick people use to try and cheat drawdowns (not going to work!).

There are also better ways when determining if your strategy is broken or not.

However, I respect your decision to test and test again as this is the only way that one can come up with an independent assessment on validity of the idea.

What do you use currently to know that strategy might be broken..do you have any objective rule? Since you said u stopped trading the above strategy some time back, how did you identify that?

Hi Cesar,
thanks for a great post.
Have you ever tried dynamically changing your position size depending on how your equity curve is doing?
Another thought (completely untested) is what you could achieve if you treated your equity curve in a trend following manner i.e. you enter and exit as if it was a trending stock?
Cheers,
Ola

The ONLY way trading the equity curve can be beneficial is if you are trading a suite of uncorrelated models which are designed to perform in differing market conditions and you use whatever method you are using to switch strategies on and off as a method to allocate between them. It doesn’t matter which method you use to trade a system’s equity curve. If you are using it to simply turn a single system on or off in a binary fashion, it will ALWAYS underperform simply trading the system flat out through thick and thin.

I think equity curve trading is valid but only on strategies that have a normal distribution of Rs. I wouldn’t do this on trend following strategy as not trading could result in missing a entry which could be a massive winner. Perhaps not shutting off the strategy is a good approach but only reducing position sizing.

Hello, I just came across the post, I trade an short term algorithmic breakout system on a basket of commodities, which, after 9 months of fantastic results completely broke down in the waning 3 months of last year. I would think the method to toggle a system on and off would have to have a similar timeframe as the system itself. A 200 day moving average is a wonderful way for long term holders of equities to gauge the market regime, but isn’t very helpful for a system that trades daily and one who’s average holding period may be only a few days.
Filtering trades as well as filtering systems, after all, that’s what we’re attempting to do, must have similar and congruent timeframes or will result in too much give back. If the speed of the system signal lags the systems entry signals by too long of a period the benefit I believe would be lost.
Any thoughts??

Something I consider for some systems is to calculate the Kelly coefficient going back about 100 trades. This is plotted with the equity curve with its own Y axis. If this plot is moving down towards zero, it is time to watch out or reduce exposure. It could be used as a rough guide on position sizing too.

Just my 2 cents worth:

I use Howard Bandy’s CAR25 metric, where I use a triangulated distribution of the last 2 years of marked to market daily returns. If that value falls below zero. The strategy moves to a shadow trading mode where it monitors the system and turns it back on only when the CAR25 metric returns above zero. My automated systems also adjust the position size of each trade using his method to determine safe_f; The position size which for me limits MDD95 to 20%. Thus if a sytem deteriorates it will first take smaller trades then turn off if its not worth the risk. (book ref: Quantitative Technical Analysis, Howard Bandy)

Other things I have tried and thought were worth more investigation at the time. When using a equity curve Moving Average, instead of turning off the system either 1) reduce the position sizing by 50% or strangely enough increase it… (I found the equity curve of ‘some’ mean reversion systems to be in themselves mean reverting). Though this defeats the purpose of protection against a broken system and will ensure larger losses if the system is in fact broken!

Here is a “walk-forward” idea that I have thought about but never tested. For example;
1. Optimize the system for 1 year from the start of the testing period (=IS).
2. Look at the resulting out-of-sample (OOS1) results for the next 3 months.
3. If the OOS1/IS > 50% then trade it for the following next 3 months (OOS2). If <50%, don't trade it.
4. Log the returns for the OOS2 period.

Repeat step 1-4 several times by shifting forward 3 months. Calculate the equity curve and the CAR, Max drawdown etc from the successive OOS2 results.
Compare with the regular equity curve.

The idea is to stop trading at the first OOS1/IS<50% during the walk-trough. Start trading again when OOS1/IS>50%. Then compare the trading results using this approach vs trading the whole time (regardless of the OOS1/IS ratio).