How a tail-shape rule traded HPE to a near dead heat with the stock

Selective rules of this kind almost always finish behind the stock they trade, yet over this window HPE's came out level with a passive hold while spending only part of the period in the market, and that result is what earns it a read. It is a backtest, the model's fixed rule set replayed over HPE's stored history, not a forecast or a recommendation; the live sample is still too thin to grade.

What this model counts is the arrival rate of far-out sessions in HPE, the days whose returns land well beyond the stock's ordinary span. While such days are infrequent it stays in cash, and it acts only once their frequency spikes to a level HPE rarely reaches, leaning long when the run of outliers tilts upward and short when it tilts down. The upward tilt dominated here, sixty of the sixty-two positions. Across those 62 trades the rule closed 62.9% as winners and accumulated +165.52% on an even stake.

How the model is built, end to end.

The figure the model leans on is how frequently outsized sessions are arriving in HPE's recent returns, a fourth-moment count that climbs when large days begin landing in quick succession and recedes when the tape returns to its routine. That count is read relative to HPE's own typical pace, so a value registers as extreme only beside this name's prior cadence. A run-of-the-mill pace leaves the model idle; only a clear acceleration in how often far-out days arrive unlocks a position.

An elevated count buys a look, not a trade ticket. Before it acts the model wants a paired directional read pointing the same way, so it never fires on one isolated shock. Should both agree, out goes an order carrying a preset stop and target. The trade then winds up by one of a handful of means, and for HPE the usual one was the position unwinding as the arrival count cooled back toward normal, short of either fixed barrier.

Follow the illustration as one sweep of the model over HPE: the arrival-count reading at one end, the gate it has to satisfy at the middle, the resulting position at the far end. No single piece is intricate; the whole design is forbearance, a sitting-on-its-hands until far-out days arrive in number and the paired directional read points the same way before any capital is risked. The rule set's lineage matters here too, because it surfaced as a lone entry within a machine-run idea search and kept its place by outlasting a clean backtest and an out-of-sample replay, the path from idea to deployment that underlies every figure on this page.

At a glance

These numbers come from a backtest of the model on HPE, scored against fixed acceptance gates, not from a live track record.

Walk-forward verification

A trade walked through

The walked example is a long carried just under two days. The outlier-rate reading climbed to an upward peak near 22 dollars, the direction read agreed, the model bought, and the position closed near 25 dollars for +13.71% once the reading settled back. It is a clean, ordinary win, a patient entry on a crowded reading and an exit taken when the move had played out rather than held for a hard target.

What the full trade record shows

Across its 62 HPE trades the model won 39 and lost 23. The exits split unevenly: 26 closed on the raw signal flipping, 15 ended when the reading settled back to ordinary, 12 reached the profit target, and 9 were stopped out.

A split this tilted toward signal-based exits is the mark of a model that mostly closes when its reason for the trade fades rather than at a hard barrier. Targets and stops together account for fewer than a third of the book, so most of the work is done by the reading itself turning ordinary again.

The biggest winner was a long that ran to its close for +22.24% over about two weeks in mid-2026. The patient winners tended to exit on the settling reading rather than a fixed target, taking what the crowded outliers offered before the effect drained away.

The worst trade was a long stopped out for -8.98% over roughly a week in September 2024; that single loss is also the equity curve's deepest drawdown, and it is the reason the drawdown gate is one of the bars the backtest missed.

No single trade carries the record. With HPE and a passive hold finishing neck and neck, the lesson of the book is that the model matched the stock without being exposed to it the whole way, trading a fraction of the time to land in the same place.

How does this compare to just holding HPE

Over the same window the model was tested on, simply buying HPE and holding it would have returned almost exactly the same. Comparing the two is how you tell whether the rule earned its keep or merely rode the tape, and here the honest reading is a tie: the model matched the hold while sitting in cash for much of the period. The tiles below put numbers on how close the two finished.

How well does the model reproduce its tape?

Walk-forward verification checks whether the saved rule path reproduces the expected signal behavior on held-out timestamps it was never fit on. It is a consistency and replay-integrity test, not evidence the model will earn money live. A clean reproduction means the deployed rules behave like the studied ones; it says nothing about whether HPE will keep generating outliers at the rate it did across the test window.

In live trading the model has stayed quiet so far, with too few signals to set against the backtest. Until more live trades pile up, the backtest is the only evidence on hand, and it should be read as exactly that, a study of how the rules behaved on saved history.

When this approach fails

The model's losses spring from the same place as its trades, the outliers. Its sharpest one is arriving early: outliers crowd in, the model buys, and HPE keeps sliding, which is how the deepest trade lost 8.98% on a long stopped out in September 2024. Because it is long nearly all the time, it has little protection against a sustained drop, and a single stop-out at that depth is also what set the equity curve's worst stretch. Its quieter cost is restraint itself, every calm patch it judged too ordinary to trade.

Three things deserve watching if this ever trades at size. The first is drawdown, since the worst stretch in the backtest was about 8.98% and it overshot the 5% bar the gates set. The second is the long-heavy stance, because the model has almost no way to profit from or hedge a lasting decline. The third is the distance between live and backtested behavior, the earliest sign the outlier-rate reading is no longer marking the same moves.

Risk and honest limits

On this run the model's automated checks recorded a caution rather than a clean pass. It is one more reason to read everything here as backtest evidence about the rules, not a verdict on the stock.

Lifecycle