The second pattern

This signal is based on whether the inside bar closes high or low relative to the day’s range, rather than above or below the previous day’s close. A close in the upper 35 percent of the day’s range qualifies as a strong close and triggers a short trade, while a close in the lower 35 percent of the day’s range qualifies as a weak close and triggers a long trade. The rules are:

Enter long on today’s close if:

1. If today’s high is below yesterday’s high.

2. If today’s low is above yesterday’s low.

3. If today’s close is in the bottom 35 percent of the day’s range.

As formulas, these rules are:

1. If High < High[1];

2. Low > Low[1];

3. (Close-Low)/(High-Low) <= .35;

The rules are inverted for short trades. It shows some recent signals.

The setup was tested on daily data in the EUR/USD pair from Dec. 28, 1998 to Dec. 29, 2008, with the same account size and trade fees as the first test. Again, trades were exited after two days and the pattern was tested with and without the 40-day trend filter.

It shows the results. The pattern’s unfiltered results are better than the first pattern’s, and in almost every aspect, the filter version again performed better, especially on a riskadjusted basis. The filter cut the number of trades in half (and halved commissions and market exposure) while net profit increased slightly. Also, the balance between the maximum number of consecutive winning trades and the maximum number of consecutive losing trades tilted toward the latter without the filter; the longest winning streak far outpaced the longest losing streak with the filter.

In short, the pattern’s basic tendencies appear to be enhanced when long signals are traded when the market is up and short signals are traded when the market is down. Other methods of identifying the trend or immediate trade context have the potential to produce better results than the rudimentary rule used in this test.

One interesting detail in both patterns’ tests is the filtered versions produced slightly larger average losing trades than the unfiltered versions. But the filtered versions’ much higher winning percentages and slightly larger winning trade values more than made up for this deficit.

It compares a few key statistics for the filtered version of the second setup in different portions of the analysis window: Dec. 28, 1998 through Dec. 28, 2001; Dec. 29, 2001 through Dec. 29, 2004; Dec. 30, 2004 through Dec. 29, 2008. In terms of the number of trades and winning percentage, the results were fairly consistent from period to period, with the exception of the negative return for long trades from the December 1998 – December 2001 window.

Short trades outperformed long trades — in terms of consistency and reliability, not net profit — in two of the three periods. This is a good thing, as it indicates the setup is not merely hitching a ride on the back of the EUR/USD’s upside bias during the majority of the analysis period. Also, it makes clear the most recent four-year period (December 2004 – December 2008) was not the best period for the setup, although it was very positive.

For such a simple, robust setup, the results aren’t bad, and there’s plenty of room to extract more value from the pattern. First, losses were not controlled — all trades were exited after two days, win or lose. Testing indicated additional profit potential existed beyond this time horizon, but the probabilities of success decreased as time passed — i.e., winning percentages and reward-risk ratios declined, although they remained favorable. However, taking partial profits at the two-day point (or at a price level corresponding to a high-probability profit target in the first days of the trade) and protecting the remainder of the position with a stop-loss could allow for additional profits with minimal risk.

It shows the results of combining the two filtered versions of the patterns. The results are — predictably — something of a compromise.