Designing a trading strategy is tough work and that is if we include only technical factors as opposed to adding in fundamentals and seasonal cycles.
There are many variables to keep in mind when undertaking the design of your own strategy:
- Which indicators will you use?
- What price points will you use?
- Time horizons including your outlook, stops, and targets
There is a reason that our company has been around for so long and produced many who now trade for a living. This is hard work but our experience in trading and system design helps lessen that load and gets you on the road to profitability quicker than going it alone.
The sheer amount of combinations that can be put together to form a viable trading strategy is overwhelming.
Just when you think you have found a good strategy, the work just begins with your back testing and that includes finding out – if it is better than a random selection of variables with a random data set.
Data Mining For A Trading Strategy
Why would you want to test out your strategy against random data?
If I string together a few hundred strategies using a huge selection of variables chosen randomly as you will see in a moment, a true edge will outperform another set of random variables applied to random data. Statistically, since we are using a broad search, there should be out performance with a good strategy just by chance.
As an example: If our basket we are drawing from has 10 variables and we randomly pick 3 and test against real data, then we take another selection of 3 and apply to random data.
With the power of computers, it has become less burdensome to be able to test literally millions of combinations to find a strategy that has promise.
There is the issue of curve fitting the data to make the tested data look more appealing however that makes the edge, not truly an edge. Avoiding curve fitting is a must.
Another issue is that of random luck.
When you take a few variables and apply them to a data set, the results may look extremely good and prompt you to believe that it is a trading strategy worth using.
The question is, if the variables you pick are random and apply it to real data, how do you know that it is not simply “chance” that your trading strategy looks good?
Compare random variables to random data.
This is an equity curve of a tested long only trading strategy for S&P using a data range of Jan 1 2004 – May 21 2014
Other data that you’d want to keep in mind:
- Win rate 75%
- Profit factor 3.36
- CAGR (compound annual growth rate) of 10.10%
The strategy uses the follow variables in terms of each actual trade (this is not a strategy recommendation but feel free to test it):
- Open of 6 bars ago is less than or equal to the close of 9 bars ago
- High of 6 bars ago is greater than the low of 9 bars ago
- RSI setting of 2 – today’s close is great than or equal to 5 and
- RSI setting of 2 – today’s close is less than or equal to 15
- Our stop loss is 2 X 20 period ATR
- Max holding time is 10 days
- Enter and exit on next open after trading signal generated
Forgiving the last 6 months of 2008, this equity curve shows some potential. It would be very easy as a trader to discount the issue in 2008 and begin to commit real money to this strategy.
How Do It Do Against Random Data?
To consider our strategy having a true edge, we need to see it compared to something to see that it outperforms in the comparison. Given we are using a a huge database of variables and taking a random set of variables, statistically we should see some out performance.
A way to do that is to compare your strategy to a random baseline and expect to see your strategy performing stronger than random. We need to see that our possible edge does better than something found by random chance.
The solid blue line is the original equity curve and the red lines are random signals and random data. Our strategy that originally looked good, did not outperform random signals and data. We would want to see the solid blue line well above the red lines.
Testing Your Strategy – Better Than Random Signals?
Keep in mind that for this example, I used random variables plus real data and random data. A trader that is using a trading strategy that was not chosen randomly, can test that strategy by taking the strategy listed above and test against their own variables.
If the strategy is not better than random, what trading edge are you taking advantage of? None.
This is similar to people trading support and resistance levels. I wrote about the danger of support and resistance and random lines on a chart and the question you should pose to any trading approach is – is it better than random?
This is only one test that you’d want to consider when designing your own trading strategy albeit one of the more important ones.
Start with a hypothesis, assemble a range of variables to consider to serve that hypothesis, and test against various data sets.
Look for a strategy that outperforms random samples and you may have found a trading edge that will serve you for years to come.