By: Thomas Demark
For many years now, one of the most popular trend-following methods has been the moving average. The simplicity of its construction and the ease of its interpretation contribute to its widespread usage and acceptance. Unfortunately, this tool's success is derived from a particular market's ability to trend. My research indicates that markets generally move in trading ranges and trend much less frequently. Historical observations suggest that, approximately 75 to 80 percent of the time, price of a particular security tends to move in a trading range. On the other hand, 20 to 25 percent of the time, price trends are either up or down. Furthermore, additional research shows that price accelerates in a downtrend generally about two to two and a half times faster than price in an uptrend. This phenomenon can be easily accounted for by the fact that whereas investors typically accumulate a position over a period of time, their recognition of a price decline is immediate and their tendency is to liquidate the entire position at one time.
The most common and basic calculation of the moving average is arithmetic and involves only adding together the closing prices of a security over a prescribed period of time, dividing this sum by the total number of entries, and then plotting that period's value on a chart coincident with that interval's price range. Unfortunately, this is the common practice but it is not necessarily the ideal or the correct one. Most market timers ignore or fail to recognize many questions that arise. Specifically:
My experience suggests that the results achieved by using the traditional moving averages are no better than those realized by using most other conventional trend-following approaches. Moving averages are easily calculated and understood and can be found on most quote service graphics displays and in almost all graphics software packages. Don’t confuse this universal availability with utility and trading success, however. I have found that, in this industry, there is no correlation between acceptance/usage and performance results – in fact, precisely the opposite is true in most instances.
Given some exceptions, despite extensive research, I have uncovered only a few circumstances in which moving averages can be applied and respectable results can be expected. Specifically, by definition, moving averages identify turning points in trend well after they have occurred. As I stated earlier, markets operate within a trading range most of the time. Occasionally, however, prices do break out of this pattern. My adaptations to moving average analysis have proven to be worthwhile in those breakouts because the risk of being whipsawed is diminished considerably.
Basically, the various moving average techniques I recommend all cope differently with the issue of trading range whipsaws. One version projects moving averages into the future; another averages highs, lows, and closes for a period of time to create a fictitious average price to compare with the moving average; and still another employs a moving average only when price breaks out of a trading range. A description of each is presented here.
Conventional trading analysis provides for the moving average entries to be aligned with the trading days such that the last moving average entry coincides with the last price entry. There is nothing sacred about this particular relationship, and this practice has always bothered me. I experimented with centering the moving average and realized some improvement in performance results. Instead of calculating the moving average and positioning the values beneath the most recent entry, as most traders would do, I experimented with the technique of having the projected average coincide with the current day's price. In a sense, you might say that 38 percent of the moving average band appears prior to the current price entry, and 62 percent is projected into the future. In other words, 62 percent of the moving average has been projected into the future. I found that this shift retained the pattern of the moving average and at the same time reduced the likelihood of whipsaws inherent in trading ranges.
Another approach for calculating a series of moving averages attempts to avoid the problem of trading ranges and of whipsaws by making certain that the short-term moving average exceeds the upside long-term moving average in the case of a buy signal, and that the short-term moving average exceeds the downside long-term moving average in the case of a sell signal. At the same time, both moving averages must exceed either a fictitious price peak or price trough that is an average of the most recent two days' highs and two days' lows. I have been the most comfortable using moving average of 5 and 21 days' duration. To demonstrate just how this method works, calculate the value of each moving average by summing the opens, highs, lows, and closes over each respective time period. Next, project both averages into the future: in the case of the 5-day average, project 3 days into the future; in the case of the 21-day moving average, project 13 days into the future. If the 5-day projected value is more than the 21-day projected value, then look to buy; if the 5-day projected value is less than the 21-day projected value, look to sell.
Performance results can be enhanced further by making certain that both averages exceed the hypothetical two day price high (buy) or low (sell). Further improvements are realized if both moving averages are declining or advancing together. Finally, by making certain that the 5-day average is greater than the 21-day average for a buy signal and less than the 21-day average for a sell signal, you should realize improved results.
To avoid a multitude of signals while locked in a trading range market, I created a moving average system that only became active when price recorded either a 13-day-high low or a 13-day-low high. Let me explain this concept further. If price advances and it records a low greater than all previous 12 lows, then a 3-day moving average of lows is installed and followed for a period of 4 trading days to identify a place to sell. Conversely, if price declines and it records a high less than all previous 12 highs, then a 3-day moving average of highs is installed and followed for a period of 4 trading days to identify a place to buy. The moving average is active for a period of only 4 days after the higher low or the lower high is recorded. As you can see, the application of the moving averages is dependent on the fulfillment of various prerequisites. Other variations of this approach can also be applied. In every instance, however, the key ingredient of any approach is its ability to remain dormant while price moves sideways. Once price breaks out of the trading range, the technique should be sufficiently sensitive to detect any movement that would precede a trend reversal.
For many years, I observed a central tendency for price activity to move within a band defined by a moving average that was identified by multiplying each day's price low by 110 percent and each day's price high by 90 percent. This band can be smoothed by multiplying an average of the previous 3 days' lows and highs and by increasing the band factors to 115 percent and 85 percent. When price exceeded this moving average band, overbought and oversold readings were generated. The percentages presented can be adapted to specific markets.
One technique I developed many years ago I call the TD Moving Average technique. It is designed to initiate buy and sell signals on the first day both of two moving average – a long term and a short term – turn up or down simultaneously for the first time. Generally, the short-term confirms – that is the day action is to be taken. In other words, the first instance they both move up or down versus the previous day's TD Moving Average readings is the trigger day. Typically, the two moving average periods I use are 13 and 55 days, but the latter period has been adjusted to use as many as 65 days.
I believe another approach has merit, but because of both software and data constraints, I have been unable to test it. This method involves identifying and averaging the median (middle) price recorded each day for a particular period of the day. I plan to experiment with variations of this technique now that I possess the software required; I am awaiting the necessary data.
My moving average techniques are unconventional. They have been designed to counteract the nemesis of all moving average approaches – trading range and sideways markets. I believe that these methods circumvent the obstacles confronting the average trader, and can be implemented to give the savvy trader a market edge.