1988issue C071-7
Two-sided optimization for volatility-scaled breakout rules
A long-only switch can be written as buy, sell, and wait inequalities, stretched by a volatility estimate, and locked only after the search sample already contains both a major advance and a major decline.
- Write the mechanical-trading-system as one procedure: buy after a percent-from-extreme rise from the post-exit low, sell after a drop from the post-entry high, and stay flat when neither inequality holds.
- A volatility-scaled-threshold and a persistence-count let volatility-position-sizing raise the entry hurdle, change how long a break must persist, or shorten the exit distance before exposure is taken.
- System-optimization here is a sequential-parameter-search that retunes one coefficient at a time until the set stops moving.
- Accept the locked set only if the search window is a two-sided-sample that already contains both a major advance and a major decline.
Write entry, exit, and waiting together
A complete long-only switch can be written as two inequalities: buy when the distribution-adjusted quote exceeds the lowest price since the last sale by X percent, and sell when it falls below the highest price since the last purchase by Y percent. The mechanical-trading-system stays flat when neither condition holds.
That pair is the percent-from-extreme trigger. One historically searched pair of those percentages was a buy rise of 8 percent and a sell decline of 3.5 percent.
Stretch the rules with a volatility estimate
The same skeleton can replace constant X and Y with linear functions of a volatility estimate, so a larger estimate can raise the entry hurdle or shrink the exit distance. That replacement is a volatility-scaled-threshold.
The volatility estimate can be the fractional change between successive quotes, averaged over a day-count chosen by the same search. Entry and exit can further require the percentage break to hold for a consecutive-quote count, and that persistence-count can itself be a linear function of the same volatility estimate.
Used this way, volatility-position-sizing is applied before entry and throughout the position. It can refuse a new exposure or cut an open one when swings widen.
When a cash distribution falls between two quotes, a price factor built from the pre- and post-distribution quotes is applied so the volatility increment is not an artifact of the payout. That dividend-adjusted-volatility step keeps the measured fractional change on price action rather than the payout.
Retune one parameter at a time
A ten-parameter set can be searched by cycling through the parameters one at a time, retuning each until file payoff stops rising, and repeating the cycle until no parameter still moves. That sequential-parameter-search is the system-optimization used to lock the full set.
Each single parameter can be located with an increment search that steps by a delta, reverses and halves that delta when payoff worsens, and stops when the increment is small. A worked search needed ten payoff evaluations and seven distinct passes through the quote file.
Accept a fit only on a two-sided sample
Fitting the rules on a four-year window that contained no meaningful decline produced a 1987 sell that first printed on 9 October, confirmed on 16 October, and still filled only at the 19 October close under the fund's execution rules.
The same constructor later required the search sample to contain both a major advance and a major decline. Under that two-sided-sample rule, a 1979 to 1986 fit, cut off a year before the 1987 break, exited on 14 September, while a four-year fit on a more two-sided series had already exited on 17 August.
All readings on this track · 25 readings
- 1988Two-sided optimization for volatility-scaled breakout rules
- 1990Weekly range volatility as a tactical regime filter
- 1993Heat first, then contract count on a weekly close-only breakout
- 1994Shared sigma for trend entries, retests and stops
- 1995The promised result as one testable trading procedure
- 2001Convert a survival ceiling into a volatility filter
- 2002Bound the next loss before the bear trade
- 2004Dollar-loss tolerance versus percent-of-equity stops
- 2004Bounded risk before entry: stop, size, and leverage
- 2006Position sizing as the pre-trade bound on loss and exposure
- 2006A trend-regime matrix for volatility position sizing
- 2011A premarket range checklist for opening-hour breakouts
- 2011Stop-defined share count on a clustered intraday short
- 2012Mechanical system design with strategy, signals, and size
- 2012Precommit size before a one-day policy shock
- 2013R-multiple expectancy, drawdown, and volatility position sizing
- 2014Constructing trend-aware slope-divergence rules
- 2014Pretrade drawdown fences and volatility sizing
- 2015Convert pip value into cash before a forex lot size is chosen
- 2016Expectancy as an execution skill rather than a backtest score
- 2017Volatility parity equalizes the ticket; implied volatility is a regime dial
- 2018Wide-range breakout rules from chart volatility
- 2019A structure-based stop, then resize or pass
- 2020A cash rule that zeros a trend book on extreme volatility and a short losing window
- 2025R-multiples before you size up