Analyzing a Pullback System

It is important for the trader to have multiple techniques in his toolbox, since different techniques are favored in different market conditions. I run over a dozen different scans every evening and select the ones best suited to the current market environment. In order to know if a scan is suited for the current market environment the trader must make an investment in thoroughly understanding how each system is affected by a variety of market parameters. Back testing does not guarantee future performance, but I would much rather be using a system that back tests well than one that does not.

Pullback systems are popular because they produce results. They are based on the idea that trends continue, and the best way to trade them is to look for a pullback in the trend because it generally represents a lower risk entry point. Managing risk is one of the keys to prospering in the trading business. One of the keys to managing risk is to thoroughly understand a system and how it is affected by various market conditions prior to trading it.

Prior to trading a system I want to have a clear understanding of issues such as:

· How does the system behave when the Market is in an uptrend?
· How does the system behave when the Market is in a downtrend?
· How does the system behave when the Market is moving sideways?
· How do price and volume affect the system?
· How does trigger day volume affect the system?

One of the systems in my trading toolbox, MAPS, looks for stocks in an uptrend that have not been below the 30 day simple moving average for at least 30 days, and have currently pulled back to within 1% of the 30 day average. The system then triggers if the stock makes a higher high the following day. The exit strategy is to hold the stock for three days, then sell. This exit approach not only works well, but makes it easier for traders who cannot watch the market carefully during the day. Most of my systems are based on end of day data so I do not have to be glued to the screen all day if I don’t want to be.

An example MAPS stock is shown in Figure 1. SLG was found by running the basic MAPS scan on 11/10/04. SLG had been above the 30 day average (shown in red on the chart) for at least 30 days and then pulled back to within 1% of the 30 day average on 11/10/04. The following day SLG triggered, by making a higher high, when it moved above $54.19. Three days later it hit a high of nearly $59.

I look for a system to have three key characteristics in the market environment in which it is intended to be used prior to considering it for trading. First, winning trades must occur more often than losing trades. Second, the average gain of the winning trades must be larger than the average loss of the losing trades. Third, the system should show better returns than just trading a market index like the SPX. If a trading system has these three characteristics then it has a positive expectancy, and is potentially interesting.

My first step in looking for these characteristics is to run a back test during a market environment that should be favorable to the system. If it does not look interesting under favorable market conditions then I look for another idea. If the system tests well in favorable market conditions then I look at the effect of various filters to develop an understanding of how the system works, and then back test it in a variety of different market conditions. To receive time tested trading setup’s, an analysis of current market conditions, and tips from a full time trader with 20 years of experience take a look at the Timely Trades Letter.

Figure 2 shows a bullish period in the NASDAQ that I used for the initial testing of the MAPS scan. Back testing MAPS during the bullish period of 03/14/03 to 01/23/04 yields interesting results as shown in Figure 3. The annualized ROI is better than trading the S&P index during this period. Note that this figure can be somewhat misleading because it assumes that one takes all trades during the period, something that your account size may not allow. However, it serves as a good figure of merit.

Figure 3 also indicates that 62% of the trades are profitable and the average winner gains 2.4% while the average loser loses 1.81%. If you’re winning most of the time and the average win gains more than the average loser lost, the system has potential and is worth investigating further. There are other factors that come into play, since average wins and losses can sometimes be misleading, but with the data so far the system does look interesting.

After verifying that the system has promise in one bullish period, I needed to verify that it also works in other bullish periods. A system that is only tested in one period may be curve fit to the data, or may just be a fluke. Testing in multiple bullish periods addresses these issues.

Testing MAPS in the bullish period of 09/20/01 to 01/02/02 resulted in 144 trades with an annualized ROI of 112%, 61% winning trades, and the average winning trade gained 3.5% while the average losing trade lost 2.11%. MAPS shows interesting results in this period as well. I also tested MAPS in the bullish periods of 10/09/98 to 01/29/99 and 03/11/02 to 08/09/02. The results for these bullish periods in the market were also very encouraging.

I then tested MAPS in two different Bearish market periods and obtained the results shown below. During bearish periods MAPS shows better results than just trading the market index (SPX), but the results are not strong enough for me to actively trade MAPS during market declines.

Parameter………………….. 03/11/02-08/09/02…….. 01/30/01-04/09/01

Annualized ROI………………………. 14.54 ……………………26.31

SPX (buy & Hold)………………… <51.71> ………………..<89.63>

Number of Trades ……………………303 …………………………64

% Winning Trades …………………….52.48 ……………………..54.69

Avg. Gain of Winners………………… 2.34 ……………………….3.08

Avg. Loss of Losers …………………..2.31 ………………………..3.16

I use NASDAQ trend lines along with an analysis of key support/resistance areas, and volume patterns to determine whether or not to use the MAPS system. If the NASDAQ has been in a decline and breaks a key descending trend line, I will start using the MAPS system. If the NASDAQ has been in a bullish phase and breaks below a key trend line I will stop using the MAPS system until the market environment becomes favorable again.

Backtesting can be used to develop a clear understanding of how different filters and parameters affect a trading system. I use backtesting to test the effect of price, volume, market conditions, and a variety of other parameters on each of the systems in my toolbox. MAPS is based on a pullback to a stock’s 30 day simple moving average. I ran several EDS scans to determine if there was anything magic about the 30 day average, or if other moving averages would work as well.

The results of this investigation are shown in Table 2. As the moving average used for the pullback in MAPS is increased from the 15 day to the 35 day average the annualized ROI increases, as does the percentage of winning trades. When the moving average used for the pullback in MAPS is increased beyond the 35 day, the annualized ROI drops off. Based on this data, the two best moving averages to use are the 30 and the 35 day. The 35 day average produces slightly better results with a significant drop in the number of trades, so I use the 30 day moving average in practice.

MA Value……… ROI % ………winners % ……….%losers ……….Risk/Reward ……..# Trades
15 ……………………44.5 …………..56.8 ……………..41.5 ……………..1.58 ……………..1520

20 ……………………56.3 …………..58.7 ……………..40.0 ……………..1.83 ……………..1961

25 ……………………54.8 …………..59.5 ……………..39.2 ……………..1.86 ……………..1827

30 ……………………70.3 …………..62.34…………….36.28 ……………2.31 ……………..1155

35 ……………………72.4 …………..65.9 ……………..33.1 ……………..2.47 ……………..652

40 ……………………65.1 …………..65.6 ……………..33.3 ……………..2.37 ……………..303

45 ……………………38.1 …………..65.8 ……………..32.9 ……………..1.66 ……………..155

Table 2: MAPS Sensitivity to the Pullback Moving Average:
Test Run 03/14/03 to 01/23/04 with 3 day holding period

I used backtesting to determine how the results for the MAPS system were affected by the price and volume of the stock. Filtering out stocks below $5.00 had a minor impact. Filtering our all stocks below $10.00 dropped the average annualized ROI by a few percent and had little effect on the percent of winning trades. Filtering out all stocks below $15.00 had similar results. The MAPS results were not strongly affected by these price filters.

Volume has a much stronger effect on the results. I filtered out all stocks with average daily volume below 100,000 shares and then increased this filter in increments of 100,000 shares to the point where I was filtering out all stocks with average daily trading volume below 600,000 shares. At each increment the annualized ROI and the percentage of winning trades increased. The MAPS system is relatively insensitive to the price of stocks, but is significantly improved by trading higher volume stocks. When I have a choice of stocks to trade, I pick the higher volume stock.

MAPS triggers based on the price rising above the pervious day’s high. I checked to see what affect the volume on the trigger day might have. Requiring that the volume on the trigger day be greater than the volume of the prior day had little effect on the annualized ROI, or the percentage of winning trades. Requiring that the volume on the trigger day be greater than the 20 day average volume improved both the annualized ROI and the percentage of winning trades. When I see several MAPS triggers in the morning, I pick the ones that are on track to have greater than average volume for the day.

The stock’s volume for the current day can be estimated by recognizing that the trading day has 13 half hour periods in it. A straight line approximation provides a multiplication factor for each half hour period. For example, multiply the volume at 10:30 EST by 6.5 to estimate daily volume, and by 3.2 at 11:30 EST. Volume doesn’t move in a linear manner during the day, it is generally heavier during the first and last two hour periods and lighter during the middle of the day. As a practical matter I have found that the linear approximation works well enough.

Table 3 shows the results of researching the effect of the depth of the pullback on MAPS results. In order to test this, the degree to which the stock must approach the 30 day moving average was varied from 0.6% to 1.7%,with the results shown below in Table 3.

Approach%…… ROI % ……% winners ……% losers ……Risk/Reward ……# Trades
…0.6 ………….78.93 ………..61.88 ……….36.46 …………2.58 ………………543

…0.8 ………….74.93 ………..62.4 ………..36.17 …………2.48 ……………….835

…1.0 ………….70.3 …………62.34 ……….36.28 …………2.29 ………………1155

…1.2 ………….62.09 ………..60.90 ……….37.59 …………2.10 ………………1527

…1.5 ………….54.82 ………..60.08 ……….38.48 …………1.94 ………………2157

…1.7 ………….52.93 ………..59.52 ……….38.86 …………1.90 ………………2594

Table 3. MAPS Sensitivity to Depth of Pullback Moving Average
Test Run 03/14/03 to 01/23/04 with 3 day holding period

The closer the stock pulls back to the 30 day moving average the better the ROI, and percentage of winners is. However there are fewer trades for the closer pullbacks, so I use the 1% number in my system and realize that when there is a choice I should look at the set up that has pulled back closest to the 30 day moving average.

Using backtesting to analyze the effects of various filters on a system can greatly improve your understanding of how the system works and when it functions best. The results of this process indicate that the MAPS scan shows promise when used in bullish market environments. Results can be improved by focusing on high volume (above 300,000 shares average daily volume) stocks whose volume on the trigger day is above the 20 day average. MAPS is one of the scans in my trading tool box and provides a frequent source of trades in a strong market.