Refining the Filter : Types of Filter and Holding Period

REFINING THE TESTING

When testing different patterns and various filters, I am looking for clear and convincing evidence that the pattern has worked well and that filters have an obvious positive impact on results. Backtesting shows the results for trading all of the patterns that occur during the test period and the ROI is annualized. That’s why traders will not see these exact numbers; they likely will not take every trade that presents itself and will usually not annualize their own results during short periods of time.

If a filter, such as the volume ones tested previously, has test results that show a few points difference in annualized ROI or a few points difference in the percentage of winning trades, it does not mean much. When a pattern or filter shows dramatic improvements in several different test periods, then I have more confidence the results are due to characteristics of the pattern or filter. An example of this is the dramatic difference found between test results of the hanging man pattern in bullish and bearish market environments. This large and consistent difference is unlikely due to chance and more likely due to the effects of the market environment.

Now that we know the best market environment in which to use the hanging man pattern, we should look at several other filters to see if there are parameters of the pattern itself that strongly influence results. If there are, then we can use them to improve trading results. If we do not find any, then we can just focus on trading the basic pattern in bear markets.

PRICE FILTER

The basic hanging man pattern during the 01/03/06 to 05/01/ 07 test period yielded 47% winning trades and an 8% annualized ROI loss. Adding a requirement to the basic pattern to filter out all stocks with a closing price below $20 bumped up the annualized ROI to a loss slightly under 6% and left the winning percentage about the same. Neither of these results is the type of clear improvement we are looking for. The results were similar when adjusting the filter to eliminate all stocks with closing prices less than $40. The price level of the hanging man pattern does not appear to be a filter worth pursuing.

VOLUME FILTER

Table 5.1 shows the results of testing the hanging man pattern with a series of minimum average volume requirements. The tests were run during the same period of 01/03/06 to 05/01/ 07. The first column in Table 5.1 shows the minimum requirement for the 21 day simple moving average of the volume; the second column is the annualized ROI; the third column is the percentage of winning trades. The results do not provide a clear indication that traders should factor minimum average volume requirements into the trading decision.

TABLE 5.1 HANGING MAN TEST RESULTS FOR MIN. AVG. VOL. DURING 01/03/06 TO 05/01/07

HIGHEST HIGH REQUIREMENT

When looking at the actual positions traded during several of these tests, I noticed that in a number of cases the hanging man seemed to work when it occurred as a recent high. An example of this is shown in Figure 5.12, in which a hanging man occurred in ATI on 12/11/06, as marked by the down arrow. In this case, the day’s high of the hanging man pattern was the highest high over five days.

FIGURE 5.12: HANGING MAN IN ATI ON 12/11/06 IS HIGHEST HIGH

Sometimes we notice things and it is just a fluke that several patterns worked with a particular characteristic. The only way to know if it is a coincidence or a real result is to test the characteristic in question. Figure 5.13 shows the results of testing the basic hanging man pattern during the 01/03/06 to 05/01/07 test period with the added requirement that trades are only taken if the high on the day the hanging man occurred was the highest high of the last five trading sessions.

FIGURE 5.13: RESULTS FOR HANGING MAN IS HIGHEST HIGH DURING JAN. 2006 TO MAY 2007

The highest high of the last five day’s requirement turned the hanging man pattern from a net loser during the test period into a net gainer. It also increased the winning percentage from 47% to more than 50%. The annualized ROI for the period was still below buy and hold; however, when a filter turns something from a net loss to a net gain and increases the winning percentage, it is worth noting.

During the testing process for the hammer pattern in Chapter four, we found that a filter for the widest range of the last five trading sessions improved the results. Using the same filter for the hanging man pattern during the test period of 01/03/06 to 05/01/07 produced the results shown in Figure 5.14. Once again, a filter that helps one trading pattern does not necessarily help a different pattern. Each trading pattern stands by itself, and filters must be tested with each different pattern.

FIGURE 5.14: WIDE RANGE HANGING MAN TEST RESULTS DURING JAN. 2006 TO MAY 2007

MACD OSCILLATOR AS A FILTER

It would be great if there were a couple of filters or, better yet, indicators that always improved trading results. There are hundreds of different indicators such as MACD, stochastic, relative strength, etc., that are available in charting packages. Many traders use these to time the market without having extensively tested them to see if they work. Some provide interesting results, many do not, but then that is a topic for another book.

As an example, some traders use the MACD or the MACD oscillator to time the market by trading long when the MACD oscillator is positive and trading short when the MACD oscillator is negative. In effect, they are looking for a bullish market when the MACD oscillator is positive and a bearish market when the MACD oscillator is negative.

We have found the basic hanging man pattern performs dramatically better in bearish markets than in bullish markets. So given that, some traders might think to use the MACD oscillator to time the market and determine when to take hanging man trades and when to avoid them.

Figure 5.15 shows the test results for the same 01/03/06 to 05/ 01/07 test period used in the initial backtest shown in Figure 5.4. Only taking trades when the MACD oscillator is less than zero during this period decreased both the annualized ROI and the percentage of winning trades. For this particular pattern and time frame, the MACD oscillator is not an effective timing tool. Once again, it is dangerous to assume something will work without first testing it.

FIGURE 5.15: HANGING MAN RESULTS USING MACD OSC. TIMING JAN. 2006 TO MAY 2007

One idea frequently used in trading patterns is that of confirmation. One form of confirmation is to look for a trading pattern, and then to not take the trade unless the stock is moving in the desired direction the next day. Since the hanging man is supposed to mark the end of an uptrend, we could look for the stock to be down the day following the occurrence of the pattern as confirmation. An example of this is shown in Figure 5.16, where the hanging man pattern occurred in WCC on 05/08/06 as noted by the up arrow. The next day, WCC closed down, which provided the confirmation, and the stock dropped 14% during the next three trading sessions.

FIGURE 5.16: HANGING MAN IN WCC WITH CONFIRMATION

The idea of confirmation can be tested by looking for hanging man patterns on day one, then requiring that on day two the close be lower than the close on day one. During the 01/03/06 to 05/01/07 test period, trading only confirmed hanging man patterns resulted in worse results than just trading the basic hanging man pattern. The test results are shown in Figure 5.17, and they indicate that the annualized ROI loss increased to more than 16% and the percentage of winning trades decreased to just more than 45%.

FIGURE 5.17: HANGING MAN WITH CONFIRMATION TEST RESULTS

DIFFERENT HOLDING PERIODS

Up to this point, all tests have used a five-day holding period. The position is entered at the opening the day after the hanging man pattern forms, held for five days, and then sold. Table 5.2 shows the results of using different holding periods. As holding periods decreased, the annualized ROI bounced around slightly but did not show a clear trend. The percentage of winning trades increased slightly every time the holding period was decreased.

TABLE 5.2 HANGING MAN HOLDING PERIOD TEST RESULTS DURING 01/03/06 TO 05/01/07

While the data shows a slight improvement in the percentage of winning trades for shorter holding times, experience tells me it is not significant enough to focus on. There are a number of different things that can influence actual trading results as compared to backtesting, such as the number of trades taken, slippage, the trader’s risk tolerance, etc.

Because of these factors, I do not incorporate a parameter into my actual trading unless backtesting in multiple time frames shows clear and compelling evidence that the parameter strongly influences results. For example, trading the pattern only in bearish market environments or when the pattern occurs at the highest high of the last five days has been shown to strongly impact results and is worth factoring into my trading. A couple of points difference in the winning percentage are not enough to be useful on a practical basis.

MOVING AVERAGE TEST FEATURING THE 200 DAY

When shorting stocks, some traders like to look at tickers that are trading below their 200 day moving average. They feel that since some mutual funds ignore these stocks, there is less demand and they tend to drop easier. Like a lot of things we are told, it sounds interesting; however, in trading it is about what works.

Figure 5.18 shows the test results for all hanging man patterns that occurred below their 200 day simple moving average during the 01/03/06 to 05/01/07 test period. Using the 200 day moving average as a filter significantly reduces the percentage of winning trades, and still results in a loss for the test period.

FIGURE 5.18: TEST RESULTS FOR HANGING MAN PATTERNS BELOW THEIR 200MA

If instead of looking for hanging man patterns that occur below the 200 day simple moving average we look for patterns that form above short term moving averages, the test results are different. Table 5.3 shows the results of testing hanging man patterns between 01/03/06 and 05/01/07 and only taking trades when the stock closed above one of the short term moving averages listed.

TABLE 5.3 HANGING MAN MOVING AVERAGE FILTER TEST RESULTS FOR 01/03/06 TO 05/01/07

Taking trades that are above the 20 or 30 day moving averages does not show a significant difference when compared to tests without a moving average filter. The results for the shorter 5 and 10 day moving averages are more interesting and show an improvement in annualized ROI. Stocks that are above a long term moving average can have a variety of patterns. They could be basing or pulling back or moving up rapidly. Stocks that are above a short term moving average are more likely moving consistently up. They may be primed for a pullback when the hanging man pattern occurs.

Being above a short term moving average may find patterns similar to the highest day of the last five days’ requirement, which was found to improve results as shown in Figure 5.13. When the filters for highest high in the last five days and above the five day simple moving average were combined and used together, the results were not significantly different than those of Figure 5.13. This is another example of how traders should not look to combine a bunch of filters that improve results. Filters interact and overlap and each combination should be tested on its own.