Evaluating mechanical timing systems against hold baselines

Why the control comes first

Evaluating a mechanical timing book starts with costs and a hold control, not with a finished chart. The archive workflow treated a broad equity index as one tradable instrument, omitted dividends, applied fixed tax rates that switched at a one-year holding period, and deducted commissions so an active book would not be favored merely by ignored costs.

Score the passive-hold baseline

The passive-hold baseline is a multi-year long-only control scored with the same instrument, commissions, and holding-period tax rules used for active books. The archive scan used hold periods longer than two years from 1972 through 2004 and scored each hypothetical round trip from that day's average of high and low in the entry year to the same average in the exit year, after a $30 round-turn commission and a 15% long-term tax.

Different entry and exit year pairs inside that multi-year hold grid produced different after-tax compounded rates. Calendar timing was therefore treated as material even when no intra-year trading was allowed.

Separate the ceiling from a live book

The mechanical reference rule bought when the slope of a six-day smoothed close flipped up and sold when it flipped down, with fills at the next open. Because the smoother needed later closes, that path was an unrealizable historical ceiling rather than a live procedure. The later prices inside the smoother are lookahead: the historical test marks turns that were not knowable at the fill.

A delayed-confirmation variant executed at the open on the third session after the reference turn so two later observations could validate the slope change before the mechanical timing book acted. The active assumptions withdrew estimated 28% short-term tax from the account each quarter, reused proceeds on the next buy signal, paid $30 per round turn, held cash at zero interest between signals, and still excluded dividends.

Replay the locked procedure

After the reference design was locked, the same three procedures were walked onto a later window from July 2003 to March 2005 and onto two single-stock series. That walk-forward check keeps the mechanical timing book fixed. A low-volatility advance favored the passive hold, while a more volatile name created more in-and-out swings even when its period trend was weaker.

Trend and volatility readings were the filter for which mechanical procedure to apply. That is regime match: choose hold versus in-and-out rules from the instrument's prevailing trend and volatility rather than applying one book to every name.

Live patterns as stand-ins, not ceilings

The evaluation framed oscillators, candlestick patterns, and other chart structures as candidate live signals meant to approximate the reference turning points. A candlestick-pattern signal is a repeatable open-high-low-close structure used as a live, falsifiable stand-in for a turning-point hypothesis.

On end-of-day data, shortening the smoother well below six days increased whipsaw trades, while lengthening it reduced the number of turns and left some swings unused. Editorial reading: a live pattern has to mark turns that delayed confirmation could have taken, in a market that regime match has already judged suitable for a mechanical timing book rather than a passive hold.