Testing Trading Rules Over Different Time
by John Sweeney
After two Sundays of dinking around (Settlement, STOCKS & COMMODITIES November and December
1991), I'd finally converted a chance idea to something with remotely promising prospects something on
the order of a three-to-one return on margin. I'd done that by emulating the basic ideas in four trading
rules (Figure 1) plus two stops. As has been my experience, the stops were the most effective in
improving the results and the stops were selected by the maximum adverse excursion (MAE) analysis
(Figure 2) to cut off the big losers while allowing enough room for winners to be successful.
Many people look at Figure 2 and say, "Hey! We've got more losers than winners ! " Right. That's
normal. You want to look at the dollar income from that, as in Figure 3, because we know the test was
profitable. This figure gives a better sense of whence the profits come. System Writer Plus accurately
picked up two gap days when stops were ineffective, so we have an unusually high dollar loss on two
trades, despite which our approach is profitable.
Even if profitable, the system raises questions. There are fewer than 30 trades, the magic number needed
for reliability. Just one big winner earned a ton of money. What about other time periods? What about
other tradeables? Couldn't the system be refined to be even more profitable by eliminating the losers?
To make the last first, refining things at this point is impossible. We're testing on too short a period to
reliably sift through and spot problems in the logic. Better at this point to test on several other periods to
find these problems, refine the system if possible and then retest on all the different periods. This iterative
process, while tedious, usually highlights gaps in my trading concepts.