How the strategy evolved.
Most firms show you only the model they run today. Shishin keeps its prior models on the record too, so the progression is auditable: you can see what changed, when, and why, and check that each step was an improvement we can defend rather than a number quietly revised after the fact. The current model, the full evidence behind it, lives on the track record. This page is the lineage.
Four locked milestones
Each was the live model in its day. The same five-year window and survivorship-free database throughout, so the figures are comparable across the row.
The first locked baseline
The four-engine stack, re-run against the current survivorship-free database after the BULL engine was tightened (a concentration cap, an overextension trim, and a score floor). Each lever was validated on its own before it went in. This was the starting point the later models are measured against.
Same strategy, refreshed regime data, then frozen
The regime classifier’s breadth input was refreshed, which shifted which engine is deployed on some days. We re-ran the same clean four-engine configuration on the refreshed breadth, then froze that breadth so the record stops moving underneath us. The strategy did not change: the regime data did. This was the live model through 2026-06-24.
A day-15 partial trim, for a smoother ride
It keeps the same locked four-engine pools and frozen breadth, and adds one overlay: once a position has been held 15 trading days, it sells 40% of it and lets the remaining 60% ride to its normal exit. Trimming a slice of the most-extended winners clips the book’s most violent up-days, which is what lifts the Sharpe. This was the published model through 2026-10-15.
A fixed defender, and idle cash put to work
The current model, and two changes. First, a correction: the previous Byakko (defensive) engine’s stock leg used a bar from the same day it traded, information it could not have had at the time. We replaced it with a pre-registered rule that uses only the prior day’s data. On its own that lowers the record, to $6.27M at a Sharpe of 2.35. Second, an addition: when no engine is deployed, the idle cash now sits in an ETF sleeve, the top three of 27 London-listed UCITS ETFs by 12-month momentum, which goes to cash when the S&P 500 is below its 200-day average. The sleeve added $1.01M (11.4% of the gain), net of margin interest. The per-trade ledger reconciles to NAV to the cent.
What the restatement changed
The previously published model against the current one, on the same window, with the middle step shown: the corrected engines before the ETF sleeve is added. Read the three together.
| Measure | Prior · 2026-06 smoothed | Byakko v8 · no sleeve | Current · with sleeve |
|---|---|---|---|
| Final NAV (from $100k) | $7.54M | $6.27M | $8.95M |
| CAGR | +137.2% | +128.6% | +145.5% |
| Sharpe (annualised) | 2.44 | 2.35 | 2.56 |
| Max drawdown | -15.9% | -15.94% | -15.5% |
All three cover 2021-05-12 to 2026-05-14 on the same survivorship-bias-free database. Figures are hypothetical, net of fees. The prior model is kept here as the firm’s reference record, not as a live recommendation.
A correction down, an addition up
Two changes landed together. We show them separately, because one made the record worse and we would rather you see that than have it netted away.
The correction. Byakko’s stock leg, as previously published, chose its names with data from the same day’s bar, which was not yet available when it traded at the open. We rebuilt the leg on a pre-registered rule that uses only the prior day’s data, and re-ran the whole stack. With the corrected engine and nothing else changed, the record falls from $7.54M to $6.27M, and the Sharpe from 2.44 to 2.35.
The addition. About a third of the time no engine qualifies. That capital used to sit in cash. It now goes into the ETF sleeve, which subscribers receive alongside the engine signals: the top three of 27 London-listed UCITS ETFs by 12-month momentum, sized by inverse volatility, each with a stop, re-ranked monthly, and moved fully to cash when the S&P 500 is below its 200-day average. Over the backtest it averaged 51% of NAV and added $1.01M, after $9,872 of margin interest on the nights an engine entry overlapped it.
What the sleeve is not: a hedge. It holds trending sector, country, and commodity ETFs, which mostly move with the market, so it adds market exposure on days the engines are idle. Its own protection is the 200-day switch, not diversification. In start-date tests from 2021 through 2023 the median result was higher with the sleeve, so the gain is not tied to this one window, but it is a trend follower: in a sideways, whipsawing market it will give some of that back. Past performance does not predict future results.
A smoother ride, not a richer one
The most important line on this page, because it is the easy number to oversell. The smoothing's durable benefit is risk-adjusted. The extra NAV is not.
The durable gain is risk-adjusted. The day-15 trim lifts the Sharpe from 1.78 to 2.44 by clipping the book’s most violent up-days, and the per-year Sharpe rises in every one of the 4 full-year regimes in the record, the bear, the recovery, the broadening, and the bull alike. That is a structural property of taking some risk off an extended winner: it shows up consistently, in every regime, not just in the lucky stretches.
The extra NAV is a different story, and we will not dress it up. The uplift over the prior model ($6.30M to $7.54M) is window-specific: it is concentrated in 2024 and partly given back in 2025. Trimming winners helps in some years and costs in others, and over a different five years it could land either side of the prior model. So we do not present the smoothed stack as a higher expected return.
Put plainly: the claim is a smoother ride, not a richer one. The honest, repeatable improvement is the steadier path (the higher Sharpe in every regime), not the headline NAV. That is why we keep the prior model on the record beside it: so the comparison is visible, and the part of the gain that is window-specific can’t be mistaken for the part that is durable. Past performance does not predict future results.
The full evidence behind the current model, every trade, the statistical-significance battery, the drawdown record, is on the track record. The architecture of the four engines is on the strategies page.