Millard, Channels & Cycles

There are always books that are to me, for lack of a better word, thorny. I’m convinced that there is something sweet in them, but I can’t get past the thorns. One of these books is J. M. Hurst’s The Profit Magic of Stock Transaction Timing, published in 1970. Innumerable traders say that this book changed their lives; so far I am not one of them. Hurst was clearly a visionary. Not only did he propose a theory of cycles, he also offered hints about what was later to be known as fractals. But I had no clue how to translate this into either an intellectual road map or a course of action.

Along came Brian J. Millard, who in Channels & Cycles: A Tribute to J. M. Hurst (Traders Press, 1999) set out to make Hurst’s work more accessible to traders without an engineering background. He starts with Hurst’s basic premise that higher frequency trading trounces (long only) buy-and-hold. Trade all the zig zags, long and short, and the contributions from the short side add significantly to the bottom line. The problem, of course, is that zig zag indicators work only in hindsight, so how can the trader identify a swing high or low more or less in real time?

Enter cycles and channels. The standard Hurst indicators use offset moving average channels of various lengths; these nested envelopes help to show how price cycles have evolved in the past. Unfortunately, since they are offset, they stop half a cycle short of the hard right edge. So programmers have had to develop algorithms to extend these channels to the present and into the near future. There are a host of commercial products that purport to do this. Although in this book Millard offers some homespun ways to extrapolate cycles into the future, he also sold software to do this--first Microvest 5.0 and Sigma-p, then Channalyze and CCS Visions. Since his death these are no longer available. But new programmers have stepped up to the plate, some using polynomial regression (for instance, the Arps Hurst Bands and perhaps Sigma Bands), others relying on Fourier transform spectral analysis (for instance, Clyde Lee’s Swing Machine). One problem with all this code is that you can’t backtest it because you have shifted the unknown back in time and made it appear to have been known. That is, it’s like trading based on tomorrow’s or next week’s Wall Street Journal.

One concept I finally understood from Millard’s book is that trends are additive. Since he considers it one of the most important ideas in his book I think it’s fair to say that I may be wiggling my way between the thorns. Here’s his example. A stock rises $2 in four weeks, then falls by $2 over the next four-week period, then does nothing over the following four weeks. A longer-term trend coexists during which the stock has risen by $8 over 40 weeks. It’s simple arithmetic to combine the short-term trends with the long-term trend. Over the long term the stock has risen by an average of 20 cents a week. So we can add this average long-term weekly rise (times four, of course) to the four-week rise of $2, for a total of $2.80. “The total effect of the two trends is to cause a rise in price of $2.80.” (p. 48) Over the next four weeks the long-term trend again adds 80 cents, but the short-term trend takes away $2.00, for a net fall of $1.20. For the final four weeks the short-term trend adds nothing, so we have a gain of $0.80 from the long-term trend. It’s important to note here that the initial leg of the combined trends rises faster than the rise in the component trends.

In an idealized chart of short-term cycles contained within a long-term channel we can see that, because of the additive nature of cycles, “the short term rises are enhanced relative to short term falls while the outer channel is rising. The opposite happens while the outer channel is falling.” (p. 123)



This is to my mind a very effective way of understanding the interplay between higher time-frame and lower time-frame price movements.