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.
Byakko (白虎) is the odd one out among Shishin’s four engines. The other three have a momentum-shaped payoff: they lose more often than they win and make it back on a thin tail of large winners. Byakko’s payoff is much flatter. Close to half its trades win, its typical trade is roughly a scratch rather than a clear loss, and its winners are modest. Its job is the one nobody else wants: to keep capital working when the broad market is falling.
When Byakko is allowed to trade
When the broad equity tape is weak, the conventional advice is to go to cash. Byakko’s premise is that there is almost always a corner of the market that is not falling: defensive sectors, healthcare and pharma, miners and commodity names, the occasional defensive ETF. The regime classifier hands Byakko the day when breadth has deteriorated, and it was on duty a lot, 322 of the 1,258 backtested days, most of them sessions on which a long-only momentum engine would simply have been bleeding. Its function is regime translation: capital stays in equities, but only in the equities that work when the index does not.
What Byakko looks for
Byakko has two layers. The first is a small basket of energy and gold ETFs (XLE, GDX and XOP, traded through their London-listed UCITS versions), held while the regime calls for defence. The second is a stock layer drawn from defensive sectors: healthcare and pharma, mining, energy, and a few others. The names it surfaces are nothing like the small-cap breakouts the other engines chase. It is looking for what is quietly holding up, or even rising, while everything else sells off.
The stock layer runs on a fixed, pre-registered rule, and every input is known at the previous session’s close. A candidate has to be above its 50-day average, liquid, and volatile enough to matter (an average daily range of at least 5%). One of three simple setups has to have fired: a breakout from a 20-day consolidation, a reclaim of the lower Bollinger band, or a reclaim of the 50-day average. It must not be stretched, with caps on its distance above the 20-day average, its RSI and its 20-day range. The broad market must not have just ripped higher, which is a cap on SPY’s 20-day return. And the stock has to close in the upper part of its 15-day range. Each position gets a flat 25% of the book.
That rule replaced an earlier one in October 2026, and the reason is worth stating plainly. The old stock layer ranked candidates with a composite score that, in the backtest, read the entry day’s own close, range and volume while filling at that day’s open. Live trading can never see that information at the open. Rebuilt on the prior day’s data, the old score showed no edge, so we retired it and restated the backtest. The new rule’s thresholds were set on 2021-22 data only and then frozen. One caveat remains: the choice of which filters to use came from exploring the full sample. Only their values are out of sample, and the rule is up for review after its first 30 live trades.
How it enters and exits
Unlike the other three engines, which fill on the close, Byakko’s stock picks enter at the open, the morning after their inputs are known. It also exits differently from the momentum specialists, and that difference is most of why its distribution looks the way it does. Its dominant exit is a fixed-horizon time exit, the close ten sessions after entry, rather than a trailing moving average: in the backtest 102 of its 206 exits were the timed exit, 87 were a volatility-scaled stop, and 17, all of them in the ETF basket, were a gate exit, the position closed because the regime itself flipped back to risk-on and the system rotated out of Byakko and back toward the breakout engine. The median holding period was about 10 trading days. A defensive trade is taken for a defined window, cut if it breaks, and handed back when the weather clears.
What its trades actually look like
Here are all 206 of Byakko’s backtested trades, binned by outcome. Set it next to any of the momentum engines and the difference in shape is immediate:
- 49.0% win rate. 101 winners, 105 losers, close to an even split and well above the momentum and quality engines.
- The median trade was roughly flat (−0.5%). The typical Byakko trade is close to a scratch, not the clear small loss of the momentum and quality engines.
- The mean trade made +3.2%, with the average winner (+13.5%) about twice the average loser (−6.6%). The payoff is flatter than the momentum engines: smaller winners, a higher hit rate. One name ran a long way, CGC (+184%); the next best, PPTA (+34%) and OCGN (+33%), were far smaller. The worst trade, ADPT, lost 11.7%.
The engine that earns the least, and matters anyway
Byakko contributed only 7.3% of the stack’s backtested gains, about $647,000, the smallest share of the four despite taking the most trades. That is the role. The momentum engines mint the big numbers in the regimes that suit them. Byakko’s job is to stop the drawdowns that would otherwise happen while those engines are correctly sitting out. An engine that keeps the book gently positive through the quarters when everything else should be flat is worth far more to the compounded result than its 7.3% line suggests, because the gains it protects are the ones that go on to compound. This is the same reason drawdown is treated as a metric, not an afterthought.
What we tested and didn’t keep
Byakko’s exit was the most heavily tested part of it. The obvious choice was to give it the same trailing-moving-average exit the momentum engines use, and we tried that, along with a hybrid of the two, against the fixed-horizon time exit. For a defensive engine the time exit won: the trailing exit kept it in defensive names too long, past the point where the regime had already turned. The stock rule was retested the same way in 2026: shorter trailing exits and longer holds both lost to the ten-session time exit. We also swept its stop and its sizing, and risk-based or tiered sizing did no better than the flat 25%. The conclusion was that a defensive engine wants to be taken off the table on a schedule, not trailed like a trend, the reasoning behind close-confirmed versus timed exits is in volatility-aware stops, and the wider list of tested-and-dropped ideas is in the experiments that failed.
Why Byakko is one of four, not the whole strategy
Byakko run alone would be a mediocre strategy: a defensive book earns little in the long stretches when the broad market is rising, and it only shines in the windows everyone else dreads. But that is exactly what makes it the perfect counterpart to the breakout engine. When breadth is strong, the momentum engine carries the book and Byakko sits out; when breadth breaks, the system rotates into Byakko and the momentum engine stands aside. They are not uncorrelated, they are regime-opposed, and committing fully to whichever one fits the day beats blending them into a muted average. That opposition is the whole argument for four engines for four regimes.
Sources & methodology
The figures here are the Byakko slice of Shishin’s locked five-year backtest (hypothetical), restated in October 2026 with the pre-registered stock rule described above, and counted per position. 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 engines are profiled in inside Suzaku, inside Genbu, and inside Seiryū, and the framework that switches between them is four engines for four regimes.