Nested smoothing recipes versus unaudited oscillator experiments

A two-layer critique of quantitative filters

Editorial frame: this note teaches a two-layer critique of quantitative filters. First write nested-exponential-smoothing so its lookbacks cannot hide inside vendor syntax. Then refuse to let a storied stochastic-oscillator or a tidy equity path stand in for an experiment whose position counts and contract multipliers have not been reconciled.

The archive facts describe a historical workflow: a requested nested construction on closes, a published rate-of-change note that supplied only the second step, a clash between a performance table and an equity-path figure, an author reply about how positions were counted, and a later remark that a full-size equity-index futures contract did not reproduce the published smaller-contract results.

Write the nested transform in the open

One requested construction first applies a 13-interval exponential moving average to closes and then applies a 21-interval rate of change to that smoother. That is nested-exponential-smoothing: a two-step transform that first averages ordered closes with an exponential smoother and only then measures change on that smoother, not on raw prices.

The same nested transform can be written as the difference between the current 13-interval exponential average and its value 21 intervals earlier, or as the ratio of those two smoother values. Either writing is a rate-of-change-on-smoother: a comparison of the current exponential-average value with the same average a fixed number of sampling intervals earlier.

A published rate-of-change note was described as supplying only the second step. Embedding the exponential average inside that note is therefore a separate composition. Editorial reading: platform syntax for that composition is not unique, so a vendor shorthand cannot replace the written 13-interval and 21-interval recipe.

Reconcile the book before the path

A reader comparing a published performance table with its equity-path figure reported that the table's position total of 40 did not match the number of trade marks on the chart. Position-count-convention is the rule that decides whether a reported trade total counts one side of the book, both long and short books, or chart marks that are not positions.

The study author replied that 40 counted one side of the book and that the experiment comprised 80 positions, split as 40 long and 40 short. The same author located the table's maximum equity decline in the window from 1 July 1998 to 1 April 1999 on the equity-path figure, arguing the path and the table were not contradictory once that window was inspected.

Replication on a full-size equity-index futures contract was said not to match the published smaller-contract results because the contracts differ at 250 currency units per point versus 50, and because margin is taken as a percentage of the position. Contract-point-value is the currency amount earned or lost per index point, which changes leveraged results even when the price path is the same.

Editorial reading: an equity path that can be aligned to a drawdown window still does not finish the experiment if the position-count-convention is only explained after a mismatch is noticed, and if a change in contract-point-value is enough to keep a full-size replication from matching the published smaller-contract results.

A documented oscillator is not the experiment

A 2004 memorial notice credited a single practitioner with developing the stochastic-oscillator and dated that outlet's first article introducing the oscillator to a May/June 1984 issue in its second volume. In this evidence set the stochastic-oscillator is a bounded range oscillator that locates the close inside a recent high-low window. It is treated as a long-documented model, not as a self-proving forecast.

Editorial reading: documentary credit and an early article date explain why the oscillator is storied. They do not reconcile position counts, they do not fix contract-point-value, and they do not turn a tidy equity path into an audited experiment.