Stop-placement methodology described below is the production approach used in the Shishin backtest. The adaptive variant outperformed fixed-percentage variants in our specific universe and window; that does not constitute a general claim that adaptive stops are universally superior. The right stop for any specific strategy depends on its universe, signal, and holding period.
The retail default stop is a flat percentage, seven percent below entry, eight, five, whatever the author of the trading book happened to settle on. It is wrong for almost every underlying. Volatility is not constant across names. A stop that does not adjust to the underlying’s recent range either gets noise-hit on the volatile names or sits unreachable on the quiet ones until the loss has already compounded.
The intuition
Consider two stocks. One has an average daily range of roughly two percent. The other’s daily range is closer to six. Both get the same fixed-percentage stop , seven percent below entry.
The first stock almost never reaches the stop on noise. A seven-percent adverse move on a two-percent-ADR name is a multi-day directional move, usually a real thesis break. The stop fires at roughly the right time. Good.
The second stock reaches the stop on an ordinary oscillation. A seven-percent move on a six-percent-ADR name is within the range of the underlying’s normal daily noise. The stop fires repeatedly on noise, before any thesis has had a chance to play out. The cumulative loss to noise hits exceeds the cumulative loss the stop was designed to prevent.
Same number, different stocks, completely different outcomes. The fixed stop is the wrong tool for both.
The adaptive alternative
A volatility-derived stop reads the underlying’s recent average daily range and sets the stop distance as a multiple of it. The math is straightforward: high-ADR names get a wider stop in dollar terms but the same equivalent “normal noise away from entry”; low-ADR names get a tighter stop but again expressed in units of their own noise.
The result is that every position carries the same risk profile in noise-adjusted terms. The expected false-hit rate is roughly constant across the universe. The stop fires when the move is unusually adverse relative to that name’s own oscillation, not relative to an arbitrary fixed percentage.
Why an upper cap matters
A pure ADR-multiple stop has a failure mode: the most extreme-volatility names in the universe end up with extreme stop distances. One bad trade on such a name can absorb a meaningfully large fraction of available capital before the stop fires.
The production stop carries an upper cap above which no stop can ever be placed, regardless of how volatile the underlying is. The cap exists to keep the worst-case loss per trade contained. Above the cap, the system treats the name as un-tradeable for that setup, the volatility is high enough that we cannot underwrite a stop we are comfortable holding. We do not enter the position.
The cap level is part of the engine’s parameters; we do not publish it. What we will say is that it sits well above what you would expect on a normal-volatility name, and well below what an uncapped formula would produce on the most extreme outliers in the universe. The cap is a guard rail, not the binding constraint on most trades. These per-trade envelopes are also the only stop layer in the system, there is deliberately no portfolio-level kill switch above them.
The receipt
We ran a stop-placement sweep across the full five-year backtest window during the production-hardening cycle for the Byakko engine. The sweep tested fixed-percentage variants at multiple distances against the volatility-derived alternative with various multipliers and caps.
The volatility-derived variant with an upper cap won. Not by a small margin. The improvement in final NAV over the best fixed-percentage variant was large enough to be the kind of structural choice you commit to and then leave alone. Bigger than parameter noise, big enough that we treat the family of approach as settled and the remaining work as fine-tuning within the family rather than re-litigating the family itself.
We are not publishing the sweep table. The exact multipliers, caps, and lookbacks that produced the winning configuration are part of the engine’s edge. A competitor with that table could replicate meaningful fraction of the Byakko engine’s edge in an afternoon. The principle, volatility-derived with an upper cap beats fixed percentage, is the part of the receipt worth publishing.
What this does not mean
It does not mean adaptive stops are universally correct. For strategies with very different holding periods, very different universes, or very different signal characteristics, the right stop might be time-based, structure-based, or even absent (some index-like strategies hold through arbitrary drawdowns because the exit is a re-balance, not a stop).
It does mean that for an equity-momentum strategy running across a heterogeneous universe with widely varying per-name volatility, the fixed-percentage stop is almost certainly wrong, and a measurable improvement is available by adapting to the underlying. The institutional discipline is to find out which framework applies, test it honestly, and ship the winner with the receipts.
Adaptive stops are ours. They were not assumed; they were earned in the sweep. The reason we publish the principle and not the parameters is that the principle is what readers should take away. The parameters are what we trade.
Sources & further reading
- Wilder, J. W. (1978). New Concepts in Technical Trading Systems. Trend Research.