This article describes one of Shishin’s research engines and the results it produced in a hypothetical five-year backtest. Backtest figures are not live trading results and do not predict future performance. It is educational, an explanation of how a systematic engine works, not personalised investment advice, and not a recommendation to buy or sell any security.
Genbu (玄武) is the quiet one of Shishin’s four engines, the quality specialist. It also has the lowest win rate of the four: across its 42 backtested trades it lost more often than it won, roughly two losers for every winner. And yet, per day it was allowed to trade, it was one of the most productive engines in the book. Both things are true at once, and the gap between them is the most useful thing Genbu has to teach.
When Genbu is allowed to trade
Like every engine in the system, Genbu is usually switched off. A breadth-driven regime classifier decides which single engine, if any, is handed each session, and Genbu is handed very few: it was the active engine on just 72 of the 1,258 backtested days, under six percent of them. It comes off the bench when the macro read is, in plain terms, look for quality, not story: durable margins, conservative leverage, the absence of disasters. The rest of the time it stands aside and lets the momentum, defensive, or recovery specialist take the day. Restraint is not a side effect here. It is the design.
What Genbu looks for
Inside an open gate, Genbu ranks the tradable universe with the same composite score the rest of the system uses, but tilted toward fundamental durability rather than the thrust off a base. It is hunting smaller companies whose quality is real but under-noticed: the kind of names that compound through neglect rather than through a headline. The score, not the ticker, earns the rank, and it decides both which of the eligible names to take and how large to size them. We do not publish the exact weights, for the same reason we do not publish any engine’s, but the direction is the point: Genbu is the engine that prefers a boring balance sheet to an exciting chart.
How it enters and exits
Genbu fills on the close, not the intraday print, and then manages the position with the same machinery as the breakout engine: it holds until the name either closes below its 12-day moving average (the trend has bent) or hits a volatility-scaled stop, 1.25× the 14-day ATR, floored at 2%, intraday. In the backtest 18 of the 42 exits were the trailing moving average and 24 were the stop, and the median holding period was just 8 trading days, the shortest of the four engines. A quality screen on the way in does not buy a quality engine a longer leash on the way out: when a Genbu name stops working, it is cut on exactly the same volatility-aware terms as anything else.
What its trades actually look like
Here is every one of Genbu’s 42 backtested trades, binned by outcome. It is the shape of a low-hit-rate, fat-tailed engine:
- 35.7% win rate. 15 winners, 27 losers, the lowest hit rate of the four engines. On hit-rate alone it looks broken.
- The median trade lost 7.3%. A typical Genbu position was a loser cut by the stop; the most common outcome bucket sits between −10% and −5%.
- The mean trade still made +9.7%, because the average winner (+43.9%) ran nearly five times the average loser (−9.2%). The biggest backtested winner, LUMN, returned +213%; LMND (+93%) and GRRR (+85%) sat behind it. The worst trade, LEU, lost 14.8%. The win is in the asymmetry, not the frequency.
Why a quality engine still loses more than half
The intuition most people bring to a “quality” engine is that it should be steady: a high hit rate, modest winners, few disasters. Genbu is the opposite, and on purpose. Screening for durability does not make a stock go up on your schedule; it raises the chance that, when one of these under-noticed names is finally re-rated, it re-rates a long way. So the engine is built to fail cheaply and often while it waits for that, lose 7-9% on the names that never catch, and hold the few that do. The payoff shape ends up looking like a momentum engine’s even though the selection logic is fundamental.
The other half of the story is restraint. Genbu produced 12.5% of the whole stack’s gains, about $929,000 of backtested profit, from only 72 days on duty. That is a large contribution for an engine that sits out 94% of the calendar. It earns its slot not by trading a lot, but by trading rarely and only when the macro is paying for quality. Whether an engine this selective is worth a slot at all is a fair question, and one we tested directly.
What we tested and didn’t keep
Two reasonable instincts about Genbu both failed. The first was that such a low-frequency engine might not earn its place, that the capital would do more inside the workhorse engines. We measured it: removed from the stack, the book was worse, so Genbu kept its slot. The second was that its edge should come from grading its candidates more finely, re-scoring the picks to prefer the “best” quality names. That added nothing measurable over our sample and was dropped. The lesson is the same one that runs through the whole system: the edge lives in when an engine is allowed to fire and how its failures are cut, far more than in ever-finer ranking. More tested-and-dropped ideas are in the experiments that failed.
Why Genbu is one of four, not the whole strategy
An engine that trades 42 times in five years and loses most of them could never stand alone: run by itself it would spend almost the entire calendar either idle or forcing quality trades into regimes that do not reward them. Its value only shows up alongside the breakout specialist that carries the trending tapes and the defensive and recovery engines that cover the rest. Genbu is the engine for the specific weeks when the macro says quality is what is being paid for, and it is explicitly not asked to be the engine for any other week. That division of labour is the entire point of four engines for four regimes.
Sources & methodology
The figures here are the Genbu slice of Shishin’s locked five-year backtest (hypothetical), counted per position with the late-stage partial trims merged back into their parent trade. How the book is measured, survivorship-free universe, close-fill convention, the leave-out-the-winners robustness test, and the per-trade significance battery, is documented in the five-year record, leave out the winners, why backtests lie, and statistical significance. The other three engines are profiled in inside Suzaku and the companion deep-dives, and the framework that switches between them is four engines for four regimes.