Exponentially Smoothing The Daily Number of
This article presents an indicator that is the one-day rate of change of a triple exponential smoothing of
the daily number of declines, an oscillator similar to TRIX (covered by Jack Hutson in the early years of
STOCKS & COMMODITIES). Raff explains how the decline line, the number of stocks that have dropped in a
given period (most often seen as a day), as the basis of this oscillator, could have warned of the
impending stock market tumble in 1987.
The method looks good — but not good enough. It's deceptive: it works well enough, but in the period
that Raff tried the method out on, it would have made only about 50% of a buy-and-hold strategy
(although it would have shielded the user from violent declines in the period). He concludes that the
method can be improved.
by Gilbert Raff
I found myself reading with piqued curiosity when in the May 1991 STOCKS & COMMODITIES John
McGinley, the editor of the "TechnicalTrends" newsletter, was quoted as saying, ". . . the last 500 points
of the crash of 1987 were, I think, uncallable. I held a contest in 1988 in which anyone who could
produce an indicator that called the crash would get a free year's subscription to 'Technical Trends.'."
Always one to rise to a challenge, I found such an indicator and am happy to report it here for traders.
(Please also see my article in the October STOCKS & COMMODITIES for a more long-term tool that
predicted the 1987 stock market tumble.) Daily advances and declines on the New York Stock Exchange
(NYSE) clearly provide much data about market internals and have been analyzed in many ways over the
years. One particular example is in the form of the advance-decline line (Figure 1). Some analysts use
trendline breaks in this line as an indication of changing direction, but in my experience, this is neither objective enough nor consistent enough to help the trader very much.
Further, daily advance and decline data tend to be very noisy — that is, the data have a great deal of
random fluctuations (Figure 2). How, then, to use the data effectively?