Custom Search

predicting tops in historical price data

The TOPFINDER algorithm given in the previous article has been seen to be useful in "predicting" tops in historical price data. As   the saying goes: "Prophecy is extremely difficult, especially as regards the future!". So in the present article we will apply   TOPFINDER to a few stocks which have not yet reached their peaks. The future price action in these issues will thereby provide   demonstrations of either the power or the limitations of the method.

First, however, a few words are required regarding the mechanics of applying the TOPFINDER algorithm. First one must pick a   launch point, i, which identifies the day on which TOPFINDER is to start. Also the duration, D, must be chosen, where D is the   cumulative volume from the launch point to the top. D is actually determined by iteratively adjusting it to provide a best "fit" to   the price pullbacks subsequent to launch. To start this process, one must set D equal to some initial guess; I usually choose fifty   days worth of volume, i.e. the cumulative volume at launch minus the cumulative volume fifty trading days earlier. In the fitting   process used to determine D, generally an "eyeball" affair, give more weight to fitting the more recent pullbacks.

The actual computation of the TOPFINDER curve involves interpolation since we compute the difference between the current   value of cum(p*v) and the corresponding value e units of cumulative volume earlier where e = d*(1 - d/D). "d" is the cumulative   volume at the day for which the TOPFINDER curve is being computed minus the cumulative volume at lauch. e, therefore, will   generally fall in the middle of some trading day so one has to interpolate linearly between the average price at the close of that day   and the average price at the close of the preceding day.

In other words, cum(p*v) is only available from the data at a set of discrete cumulative volumes corresponding to the end of each   trading day. Yet we are treating cum(p*v) as a continuous function of cumulative volume in order to determine it at values of   cum(v) which do not generally correspond to one of these discrete points, and for this purpose we interpolate between the discrete   values of cumulative volume bracketing e. Computationally, this will generally require a macro in a spreadsheet implementation   of Midas, or some simple interpolation procedure if a high level language is used. Here I'll have to leave you to your own devices   since to help you set up such calculations would carry us beyond the scope of these articles.

Turning now to prophecy, the first figure shows TOPFINDER applied to LSI Logic as of the day of writing this article. (obv has   been omitted from the Midas chart so we can show more details of the price data). Shown are the primary support S1 (unlabelled   for simplicity), and two TOPFINDER curves, labelled T1 and T2. T1 is launched at cum volume=0, and thus is the topfinder   counterpart of S1. T2, on the other hand is launched at the point (cumvol around 2.5 million) where the price takes off from the   primary and where we ordinarily would therefore start S2. (Again to keep the graphics simple, I have omitted S2).