An 80% win rate collapsed to 31% the first time we tested it forward
Strota builds intraday trading rules on past sessions, then runs them forward on days the fitting never saw. The most selective long rule in the set fires only when a stock is trending up, sitting at the high of its day, trading above its volume-weighted average price, and showing a lopsided order book. On the data it was built from, that combination won 80% of the time. On 22 June it finally appeared in quantity out of sample, on exactly the kind of rising market it was designed for. Thirteen setups triggered. They won 31% and lost money, an average of 0.153 R apiece. R is the trade's own unit of risk: one R is what the position was sized to lose if its stop was hit.
The whole session went the same way. Thirty-nine trades resolved, every one of them long, because in a bullish tape not a single short setup triggered. Twelve of the 39 finished green, a 31% strike rate, for an average loss of 0.091 R per trade and 3.56 R across the book. That is the worst of the three forward sessions logged so far, and it happened on the market backdrop the strategy is supposed to like.
The method here is a walk-forward test. Freeze the selection rules, apply them to sessions that played no part in building them, and grade every alert by what the rule itself dictated: enter at the trigger, stop where the setup says, otherwise carry the position to the close. Three such sessions exist. A trending day on 17 June produced 53% winners and +0.199 R per long. A neutral grind on 18 June gave 47% and +0.149 R. The gap-up of 22 June is the third, and the first on which the selective stack fired more than once.
Why a rising market delivered the worst of the three is the interesting part. The morning gapped up and then faded. A cluster of breakout longs fired in the first twenty minutes of trade, into that strength, and closed flat to red. What the engine needs, it turns out, is a trend day, not merely a green open. The two look identical at the bell and behave in opposite directions afterwards, and nothing in the selection logic distinguishes them.
Six trades took a full stop, 6 R of damage between them, and what connected them was familiar. They were chased. The stopped-out names had been bought long after their moves began. One was already up 8.7% on the session at entry; others 6.4%, 6.3% and 5.5%. Every earlier study in the series that measured entry extension found the same slope, and this day reproduced it without ambiguity: the further above its base a stock was bought, the worse the trade ended.
One filter did more than underperform. It inverted. The cohort with heavier size resting on the offer than on the bid, which had tested at 66% while the rules were being written, won 29% and lost 0.159 R per trade. Its mirror image, the bid-heavy group, won 42% and eked out +0.047 R. This is now the third forward session in a row below the in-sample figure, and the sequence points one way: 55%, then 46%, then 29%. Order-book skew, as measured here, is noise dressed as signal.
The day's single best trade broke two of the strategy's own gates. It gained more than 11% on the session, gapped and held about 10.5%, and finished locked at its upper circuit for +2.34 R. The rules would have skipped it twice over, once for sitting in an early and unproven stage of its move, once for gapping more than 3%, a size that earlier work had classified as exhaustion. What separated it from the technical gap-ups that faded was a genuine catalyst: an order win, in the news before the open. Real news sustained; chart-only gaps did not. That is the one new idea the session produced, and it rests on a single observation.
Take that name out and the arithmetic turns uglier still: the day becomes 5.90 R of loss at a 29% strike rate. Concentration cuts both ways. An in-sample session earlier in the series owed +30.7 R to one dominant winner; this one owed the entire difference between a bad day and an awful one to a single catalyst stock. A book that hinges on one name in either direction is not a high-probability system. It is a right-tail lottery with a modest ticket price.
Three habits did survive. Holding to the close remained the only exit worth having, with almost every position squared off in the last minutes of the session and no intraday exit signal firing usefully. The separation between clean breakouts and chased ones held, the larger winners coming from entries taken near their bases. And the short book, gated off after earlier sessions bled through it, cost nothing this time for the simple reason that nothing triggered.
Read across all three forward days, the honest description of this system is not the one its builders wanted. It harvests trend days, gives much of it back on everything else, and depends on a handful of outsized winners that often have news behind them rather than a chart pattern. The high-win-rate story attached to the selective stack was an artifact of the stretch of history it was fitted on. A single day cannot prove a rule works. It can, when the rule promised 80% and delivered 31% on its first real test, do a great deal to disprove one.
The limits deserve stating plainly. This is one session and 39 trades, a single bullish-but-faded day, which is nobody's idea of a regime sample. Nothing short was tested at all. Thirteen selective setups is still a small count, although it is the largest such sample out of sample to date and the first on a rising day, which is precisely what makes its 31% the most informative reading available. And every figure above is idealised: risk units taken from the logged alerts, with no brokerage, taxes or slippage removed. Net-of-fees rupee performance is tracked separately, and it is a different and lower number. What is described here is what happened on three past sessions. Nothing about it is a forecast of a fourth.
More money stories
- Buying weakness worked, buying strength lost: results from a broad signal sweep
- Crypto's momentum formula loses money on Indian stocks. A six-month version doesn't.
- Two of eight sessions made the entire profit — and the short side lost money
- Low-volatility stocks, not momentum, produced a book positive in every completed year
- We tried to make a momentum strategy win every year. The momentum had to go.
- Fading morning gap-ups paid in liquid stocks and lost in illiquid ones