The following article makes interesting reading on this subject:
Improving performance with a filter to avoid losing streaks.
What if there was a way to increase a system's net profit by about 5%, while decreasing the Max DD by about 40% ? Would you be interested? I sure would.
This is every system traders dream - filter out the drawdown phase while getting in at the bottom, enjoying all of a system's run up. But the reason it is just a dream centers around the extreme difficulty in coming up with an indicator which filters out a system's losers while hanging on to a big portion of its winners. Cutting the drawdown is usually pretty easy, but the savings in drawdown is more often than not at the expense of profits.
Attain has been testing hundreds of ideas and filters over the past few months, in order to improve upon trading system performance. We are firm believers in sticking with a system - but with several prominent systems such as R-Mesa suffering new Max DDs over the past six months - several clients have called upon us to help them improve their performance.
Most of our early work on overlaying some system filters was based on market characteristics - looking at market volume, breadth, volatility, etc. While there have been some promising developments in this regard - most of these market based filters did not hold up across the universe of systems we track and have data on, and that qualifier was paramount to our research. After all - if we're trying to create a filter to minimize curve fitting, we don't want to curve fit the filter.
So we moved into the very tricky area of using the equity curve itself as a filter. Some of our ideas for trading an equity curve revolved around using stochastics or relative strength indicators to pick tops and bottoms in the equity curve, but we found the most success using moving average cross overs of the equity curve to signal whether to be "ON" or "OFF" a system.
Much in the way you might trade a stock around its 200 day moving average, the MA cross over filter we tested would have an investor "ON" a trading a system when the 10 day moving average of system equity is over the 200 day average, for example; and "OFF" the system when the 50 day average has dipped below the 200 day average. This is basic technical analysis 101, but its use on the actual equity curve of the system has been limited at best.
We tried several different moving average lengths for the fast and slow curves, and came up with several combinations that tested "ok" on the hypothetical results of 81 trading systems currently in our testing database. What's "ok"? The Net Profit divided by Max DD risk adjusted return ratio rose by an average of 42%. What's that mean in layman's terms - the testing showed the systems making nearly 50% more money for every dollar of drawdown when applying the filter. The Max Drawdown fell by an average of 16%, and the net profit fell by an average of -130%; with 41% of systems showing improvement using the filter. Here are the top 10 systems in the test using only pre-release hypothetical data:
This table shows the percentage improvement in Drawdown, Ending Equity, and the Net Equity/Max DD ratio for the listed systems after applying an equity curve filter. These numbers are the results of testing only, and do not represent trading in actual accounts. Please see the disclaimer at the bottom of the page.
But the testing really hit home when we applied the filter to only those systems we have actual trading results for. Using the filter across 48 different trading systems we have actual results for, our filter improved the risk adjusted returns for 83% of them (39 out of 48); with the Net Profit/Max DD ratio rising an average of 401%!! That's 400% more return for every dollar of drawdown than you would get when trading the system normally. The rest of the numbers line up with the dream of reducing drawdown while increasing return, with the Max DD falling an average of 38%, and the Net Profit actually rising an average of 5.63%. Wow! Here are the top 10 systems in the test using only actual fill data:
This table shows the percentage improvement in Drawdown, Ending Equity, and the Net Equity/Max DD ratio for the listed systems after applying an equity curve filter. These numbers are the results of testing only, and do not represent trading in actual accounts. Please see the disclaimer at the bottom of the page.
What is going on here? Why the difference in the testing between the hypothetical results and the actual results? That stumped us for a while, as both sets included the same commissions charge and slippage. But then it hit us.
If the hypothetical, pre-release testing of many trading systems has been developed on the known market environment up to that time, it is more than possible that the system will encounter a different trading environment in the future. And if you're a cynic and believe the pre-release testing contains a certain degree of curve fitting (intentionally or accidentally - it doesn't matter), then you're thinking that there will be a lot of poor trading environments for trading systems in the future.
So while the pre-release testing had little or no "bad" environments, the future performance of the system will have several such environments, making it not only more important to try and filter out those bad times - but easier to identify them, as there will be more of them (and they will last longer). In short - there's not a lot of bad (or long) losing streaks in pre-release testing, as that would not wet an investor's appetite in the first place. But there is the very real chance that there will be some bad losing streaks in the future. Whether there are or not, it seems to make sense to have some hedge against them happening.
The moving average filter signals a system is entering a poor environment by seeing the current performance of the system not match up with the long term performance. That signal means the current environment may be a poor one for the system, and lets the investor sit on the sideline until that time is over (as signaled by the short term system performance moving back to or above its historical norms).
The 2nd table above shows 10 of the systems which had the biggest Net Equity/Max DD ratio increase after applying the filter. It should come to no surprise that several of the systems on the list have lost money in actual trading (BWT Rock/N Russell, R-Mesa eRL, Cipher, etc.) - as those systems have encountered losing streaks which needed to be filtered out. It's common sense, really; those systems needed the help a filter could provide. But a filter than can turn a losing system into a winner has a lot of appeal, obviously. After all, a system's equity curve moves up and down just like any other piece of market data, and you should theoretically be able to trade those moves profitably.
So is it time to dust off some old systems like Blue Wave's Rock'N Russell, or I-Master? It could be. The results of the filter are encouraging, and it seems to be quite logical. Systems don't break, remember, they just become more risky (have more or longer losing streaks) as they cycle in and out of "phase". So the trick is keeping some exposure to the system in an attempt to profit when the system is "in phase", while using a filter to keep you on the sideline while the system is "out of phase".
Want to trade using this filter, or have a system you want to see it tested on? Just pick up the phone and call us at (800) 311.1145. We're not giving out the exact parameters of this filter due to our excitement of the possibilities it may bring, but can of course implement it for Attain clients.
- Jeff Malec
The tables didn't copy - sorry!
Simon
