Effect of different filters : Combining filter

CONTINUING THE INVESTIGATION- EFFECT OF DIFFERENT FILTERS

Up to this point, our testing has established that traders should consider focusing bearish engulfing trades on periods when the market is in a downtrend. Knowing the market environment in which to use a trading pattern is critical. Now we will investigate the effects of different filters on the basic bearish engulfing pattern to see if there are ways to improve the trading results. 

DOLLAR VALUE

For some trading patterns, the dollar value of the stock can influence the trading results. In order to investigate the effect of stock price on the bearish engulfing pattern, we ran the backtest during the 12/31/04 to 04/28/05 period using a price filter. The results are shown in Table 3.2. Each set of data is based on a three day holding period.

TABLE 3.2 TEST RESULTS FOR 12/31/04 TO 04/28/05 WITH DIFFERENT PRICE LEVELS

For this test period, the percent of winning trades and the annualized ROI generally increased as the closing value of the stock rose. The number of trades decreased with each increase in price. In Table 3.2 there are 1,256 trades when limiting the bearish engulfing pattern to stocks above $20. This decreases to 64 trades when considering only bearish engulfing trades in stocks priced above $70 during the test period.

We would obviously expect the number of trades to decrease as the trade candidates are limited to higher and higher prices. Sixty-four trades in four months are still enough to be statistically significant, and the interesting thing about Table 3.2 is that the winning percentage consistently increased as we limited trades to higher dollar-value stocks.

The price of the stock is something we may want to consider when taking trades. If further testing uncovers better filters, we may not use price as a factor because multiple filters can reduce the number of trades in the test period to something that is not statistically significant. We want to examine several different filters and then select the ones that provide the most leverage with the lowest reduction in the number of potential trades available. In the case where a particular day presented me with more trades than I wanted to take, I could always use the price level to prioritize the trading possibilities.

VOLUME

Another filter that often affects trading patterns is volume. Volume measures the interest in a stock. Stocks that are moving up on increasing volume are showing that more people are willing to pay more money for the stock every day, just the kind of thing you want to own. Stocks that are moving up on less volume every day indicate that fewer people are willing to pay more money for the stock, and perhaps it is time to take profits.

The average volume shows the overall interest in a stock. High volume stocks have institutional support almost by definition. Lower volume stocks are often not followed by institutions because they must take large positions and cannot get in and out easily. In order to determine if the average daily trading volume affects the results of trading the bearish engulfing pattern, I ran the backtest for the 12/31/04 to 04/28/ 05 period seven times with a different minimum average trading volume requirement each time. The average volume is computed as the simple moving average of daily volume over a 21 day period. The results are shown in Table 3.3.

TABLE 3.3 TEST RESULTS FOR 12/31/04 TO 04/28/05 WITH DIFFERENT AVERAGE DAILY VOLUME REQUIREMENTS

The results are different from those of the price testing shown in Table 3.2. When we increased the minimum price requirement for taking a bearish engulfing trade, the winning percentage increased every time. When the average volume requirement for taking a bearish engulfing pattern trade was increased, the results were not smooth. Sometimes they increased between volume levels and sometimes they decreased.

Table 3.3 indicates that there is not a strong enough correlation between the average daily trading volume and the test results to use average volume as a factor in selecting trades. Backtesting is valuable not only for how it finds ways to leverage a trading pattern, but also because it shows you factors that you do not need to consider. In the case of the bearish engulfing pattern, we have found that results are better when trading higher dollar stocks and that the average daily volume does not have a significant effect on results.

FIRST DAY VERSUS SECOND DAY VOLUME

The next filter tested for the bearish engulfing pattern is the volume on the day the bearish engulfing occurs. Does it matter if the second day of the pattern occurs on above average volume? An example of a high volume bearish engulfing pattern is shown in Figure 3.9 where the second day of the pattern is marked with an up arrow and the high volume on the day is marked by a down arrow.

FIGURE 3.9: HIGH VOLUME BEARISH ENGULFING IN FDX ON 3/15/05

In order to answer this question, I ran the backtest during the 12/31/04 to 04/28/05 period and added a filter describing the relationship of the volume on the second day of the pattern to the 21 day simple moving average of the volume. The results are shown in Table 3.4. The standard three day holding time was used for all positions.

TABLE 3.4 TEST RESULTS FOR 12/31/04 TO 04/28/05 WITH DIFFERENT DAY TWO VOLUME RATIOS

The bearish engulfing pattern is looking to short an uptrending stock after a down day occurs that has a body whose range completely covers or engulfs the previous day’s range. One would think that taking trades when this down day occurs on large volume would be good since the large volume might also indicate that the down day (the second day of the pattern) was more important. The test results of Table 3.4 indicate that, at least for this test period, this is not the case.

There are many ideas that seem logical for trading patterns. Some of these work, many do not. This is why testing trading patterns before using them is important. It is much better to find out what really works instead of trading on beliefs and assumptions. The data in Table 3.4 indicates that, in general, above average volume on the second day of a bearish engulfing pattern hurts rather than helps results.

If instead of looking for a volume increase compared to the average daily volume, we simply look for larger volume on the second day of the bearish engulfing pattern than we had on the first day of the pattern, the results are much more promising. The data in Table 3.5 shows the results of running the backtest seven times during the period of 12/31/04 to 04/ 28/05 using a constant three day holding period for each test. 

TABLE 3.5 TEST RESULTS FOR 12/31/04 TO 04/28/05 WITH DAY 2 VOLUME LARGER THAN DAY 1 VOLUME

The first line in Table 3.5 shows the results of the standard bearish engulfing pattern during this test period. The second line adds a requirement to the standard pattern that the volume on the second day of the pattern be at least 70% of the volume that occurred on the first day of the pattern. The third line in Table 3.5 shows the test results for the standard pattern with the additional requirement that the second day’s volume be at least 90% of the first day’s volume.

It is interesting to note that when the second day’s volume is at least 130% of the volume on the first day of the pattern, the annualized ROI in the tests doubles over that found when using the basic pattern. The relationship between the first day’s volume (the volume on the white bar up day) and the volume on the second day of the pattern (the black bar down day) is clearly important to results.

Another suggestion by some traders is to only take bearish engulfing pattern trades when the first day of the pattern is a recent high in the stock. The idea here is that the bearish engulfing pattern would occur at recent highs. An example of this pattern is shown in Figure 3.10, which shows the bearish engulfing pattern that occurred in UNH on 04/08/05. UNH had been in a brief run up, and the first day of the bearish engulfing pattern occurred at the high for the run. The second day of the pattern marked the beginning of a retracement in UNH and trading this pattern would have been quite profitable.

FIGURE 3.10: UNH BEARISH ENGULFING, FIRST DAY OF PATTERN A RECENT HIGH

FIRST DAY AS A RECENT HIGH

As noted before, traders need to be very careful when shown a few examples of a trading pattern that “works.” A few examples provide just enough information to get you into trouble. I am more interested in how the pattern has performed over several hundred or a thousand trades than in seeing just a few successful examples. Backtesting the bearish engulfing pattern with the requirement that the first day of the pattern be a recent high yields the results shown in Figure 3.11.

FIGURE 3.11: TEST RESULTS FOR FIRST DAY OF PATTERN BEING A RECENT HIGH

Figure 3.11 indicates that adding the requirement to the basic bearish engulfing pattern that the first day of the pattern also be a recent high as shown in Figure 3.10 reduces performance. The basic pattern performs better than the version with this additional requirement.

The entire bearish engulfing pattern testing to this point has focused on testing filters external to the pattern itself, such as price, volume, and the position of the pattern in the stock’s trend. Based on this testing, we have found that the bearish engulfing pattern works best when the general market, as measured by the NASDAQ, is in a downtrend and does not work well when the general market is uptrending or in a trading range. We have also found that the pattern works better on higher-priced stocks and with a holding period of three to four days.

AMOUNT OF OVERLAP

Now we will look at how slight variations in the bearish engulfing pattern itself affect trading results during the market downtrend of 12/31/04 to 04/28/05. The basic bearish engulfing pattern just requires that the second day’s black body overlap the first day’s white body—any amount of overlap is OK. The next test looks to see how the results are affected by requiring a larger overlap.

Figure 3.12 shows a bearish engulfing pattern that occurred in SNDA on 12/31/2004 and is noted by the up arrow in the chart. The top of the second day’s black body (marked by the up arrow) extends at least 15% of the day’s range above the top of the previous day’s body. In the basic bearish engulfing pattern definition, the second day’s body just needs to be above the top of the first day’s body. The new definition we are going to test requires that the top of the second day’s body be at least 15% of the day’s range higher than the top of the body in the first day of the pattern, as illustrated by the SNDA pattern in Figure 3.12.

FIGURE 3.12: SNDA BEARISH ENGULFING OF 12/31/ 2004 WITH UPPER BODY OVERLAP

Figure 3.13 shows the test results during the 12/31/04 to 04/ 28/05 NASDAQ downtrend using the modified bearish engulfing pattern. This new requirement of “top overlap” improves both the annualized ROI and the percentage of winning trades.

FIGURE 3.13: BEARISH ENGULFING TEST RESULTS WITH “TOP OVERLAP”

“Bottom overlap” is the requirement that the bottom of the second day’s body extend at least 15% of the day’s range below the bottom of the first day’s body. Modifying the basic bearish engulfing pattern to require just bottom overlap does not improve results and in fact reduces the annualized ROI when tested in the 12/31/04 to 04/28/05 NASDAQ downtrending market. 

Based on this testing, it appears that a strong initial up move on the second day of the pattern that then reverses—resulting in a down day—is more beneficial to trading results than “bottom overlap.” I also tested whether a long upper shadow on either day one or day two of the pattern affected the backtesting results. Neither one improved the annualized ROI during the test period.

COMBINING FILTERS

Table 3.6 summarizes the results of testing the bearish engulfing pattern. It is rarely advisable to just combine all the things that improve results, since this typically reduces the number of trades to a statistically insignificant number. However, one can sometimes combine a couple of filters and still have a significant number of trades and an improved trading pattern.

TABLE 3.6 SUMMARY OF TEST RESULTS FOR BEARISH ENGULFING PATTERN DURING NASDAQ DOWNTREND 

Combining the best two filters from Table 3.6 would seem logical, but let’s test it. The sixth filter shown in Table 3.6 required the volume on the second day of the bearish engulfing pattern to be at least 1.3 times the volume of the first day of the pattern. Using this filter with the basic bearish engulfing pattern resulted in an annualized ROI of 65%. The third filter shown in Table 3.6 required the stock to have a price more than $50. Using only this filter resulted in an annualized ROI of 51%. Combining these two filters resulted in an annualized ROI of 48%, as shown in Figure 3.14. Combining two strong filters actually reduced the annualized ROI.

FIGURE 3.14: TEST RESULTS COMBINING THE TWO TOP FILTERS OF TABLE 3.6

Filters interact in ways that are not obvious, and traders need to test combinations of filters just as they test individual filters to determine which ones are the best to use. At this point we have determined that the bearish engulfing trading pattern works best when used in periods when the NASDAQ is in a downtrend and when selecting patterns that have volume on the second day greater than 1.3 times the volume on the first day. The next step in testing the pattern is to verify that this filter is also beneficial during other downtrending periods in the NASDAQ.

Table 3.7 shows the test results for the basic bearish engulfing pattern and the pattern with the additional requirement that the volume on the second day of the pattern be at least 1.3 times the volume of the first day. Note that during each of the periods shown in Table 3.7, the market was in a downtrend and a buy and hold strategy would have lost significant money.

TABLE 3.7 BEARISH ENGULFING PATTERN TEST RESULTS IN DIFFERENT MARKET DOWNTRENDS

The results shown in Table 3.7 are interesting in that during four different market downtrends, the basic bearish engulfing trading pattern was a profitable trading system when the market moved down. It also shows that in each of those periods, adding the volume filter improved the trading results.