1991issue C121-2
Stop bounds versus added system parameters
A stop-loss or maximum-adverse-excursion bound can keep a loss decision fixed before a trade is placed. Adding that bound on the same sample, without removing another control or increasing sample size, introduces a new parameter and can make simulated profits look larger without showing that the procedure is more robust.
- Adding a stop after simulated profits look too small, without removing another control or increasing sample size, introduces a new parameter and reduces statistical freedom.
- That extra control can make simulated profits look larger and losses look smaller even though the procedure has not been shown more robust.
- A stricter robustness check replaces some other buy or sell decision tool with the stop-loss bound rather than stacking the bound on the existing rules.
- Holding the underlying buy and sell rule fixed and specifying only the worst loss to accept is presented as avoiding an added entry parameter.
A bound on pain, not a rewrite of the signal
Maximum adverse excursion is a loss-control measure of the worst move against an open position, used to bound acceptable pain without rewriting the buy or sell signal. A stop-loss is a pre-set bound on loss or exposure, applied before entry and while the position is open, treated here as money management rather than as a new entry rule.
The archive characterizes maximum adverse excursion as a loss-control or money-management technique rather than as a trading system.
An added stop is an added parameter
Adding a stop after simulated profits look too small, without removing another control or increasing sample size, introduces a new parameter and reduces statistical freedom.
That extra control can make simulated profits look larger and losses look smaller even though the procedure has not been shown more robust.
The gap between simulated results and later live results is framed as a problem of how many parameters are used relative to sample size. If an extra rule is added to the entry decision, the sample size needs to increase.
Editorial reading: the audit is whether the bound is doing money-management work or acting as another free control on the same sample.
Replace a decision tool instead of stacking a bound
Robustness-testing is a check that a procedure still holds when sample size matches the number of free controls and when a stop is substituted for another decision tool instead of stacked on top.
A stricter robustness check would replace some other buy or sell decision tool with the stop-loss bound rather than stack the bound on the existing rules.
Holding the underlying buy and sell rule fixed and specifying only the worst loss to accept is presented as avoiding an added entry parameter.
A first loss does not settle later trades
A loss on the first trade of a newly specified procedure does not show whether later trades will work.
All readings on this track · 15 readings
- 1987Evaluating a black-box pyramiding routine with adverse excursion
- 1991Set the first stop from a capital-scaled MAE histogram
- 1991Opening gap fades bounded by excursion and time stops
- 1991Stop bounds versus added system parameters
- 1991Bound losses with MAE, stops, and drawdown limits
- 1992Multi-year evaluation of MAE-bounded mechanical rules
- 1992Moving-average add-ons could not be separated by maximum adverse excursion
- 1992Evaluating maximum-adverse-excursion stop reversals with short time stops
- 1992Failed range trades as breakout-system tests
- 1998Fitted moving averages for trend add-on entries
- 1998Monthly changer rules specified as one mechanical procedure
- 2002An excursion cutoff test for stops and profit exits
- 2006Constructing peak-excursion filters for stops and size
- 2006Cost-aware excursion filters for stops and holding period
- 2017Staged stops, drawdown limits, and mechanical risk survival