I have been playing with the back-tester again. First I wanted to see whether buying and selling at random gave a 50/50 chance of making a profit, or was consistently worse than that. The answer is that it is about 50/50.

Then I decided to test a simple system that buys a new high and sells a new low of the S&P 500. I wanted something more consistent than my previous strategy, in which all the profit came from the shorts. This is not a system that I plan to use, more an exercise in understanding what kinds of strategies work better than others. The system works as follows:

  • Buy a new 30-day high.
  • Sell a new 15-day low, but only if the market is below its 200-day moving average, volume is above average, and volatility is not excessive.
  • Stop loss is twice average daily range from the entry point; profit target is the same distance as the stop loss.
Here are the results:
This is really not too bad a system. It has good performance in most years, and only one bad drawdown when it gets whipsawed by the volatile top of the equity bubble in 2000. I could build in my bubble indicator (a combination of autocorrelation and P/E) to avoid equity bubbles, but the point of this excercise is not to data mine until I have the perfect system. It is to get some insight into the market by building a simple system using rules that I thought would work before I started. Here are some points based on this analysis:
  • It is easy to get a success rate of 70% or higher by placing profit targets closer than stop losses. However, losses are of course relatively larger. Systems with this feature still make money, but not as much.
  • Increasing profit targets reduces the success rate without obvious benefit to the system.
  • The system makes good profits when the market is trending, because it is allowed to exit a trade and immediately get back in — so it gets into trends early and tends to stay in them. This is an advantage of breakouts — they are an early signal of a change in trend.
  • Stop losses need to be wide. A stop any closer than twice the average daily true range for the past 30 days greatly reduces the efficacy of the system.
  • Surprisingly, volume, volatility and trend conditions do not improve the success rate of longs. For shorts, it helps to short only when volume is high, volatility is not too high, and the trend is down. Shorts also work better with an earlier signal — a new 15-day low, as opposed to a new 30-day high for longs. I think this is basically because equities (and markets in general) tend to drop quickly but rise slowly.
  • It is slightly better if you don’t require the market to close at a new high or low — e.g. making a new high during the day and then falling back is a valid signal and works slightly better than a close at a new high.
The main conclusion: new high/low systems do work!
I am going to play with the system in some commodity markets, but for now I have spent enough time on this today.