How a Hurst-regime model read BE as trending, traded it long, and still trailed a huge run
It judges whether BE's tape is trending or mean-reverting using a persistence measure in the spirit of the Hurst exponent, then trades the regime it detects. Across this window it read BE as a trender and went long throughout, winning 59.57% of 47 trades for +425.33%, yet a stock that returned more than thirty times its money left it far behind, with a 20.30% drawdown.
BE delivered one of the largest runs of any name here, and the question is what a model that first classifies the regime does with a tape like that: across this window it read BE as a persistent trender, traded long throughout, and banked a strong return that the stock still dwarfed. Read what follows as a study of that fixed rule set replayed over BE's saved history, not a forecast or a recommendation; the live sample is still too small to grade.
One question precedes any trade: is BE's tape trending or mean-reverting? The model answers with a persistence measure in the spirit of the Hurst exponent, a gauge running high when moves tend to carry on and low when they tend to snap back. A trending verdict points it with the move; a mean-reverting verdict would point it against. All window long the classifier kept tagging BE a trender, so the entire book of forty-seven trades ran long, just under three in five came good at a 59.57% win rate, and a uniform stake apiece brought the total to +425.33%.
How the model is built, end to end.
The reading the model lives by is persistence, how much BE's recent moves tend to carry on versus snap back. A high persistence reading marks a trending regime in which the model trades with the move; a low one marks a mean-reverting regime in which it would lean the other way. The classifier is the whole front end, deciding which playbook applies before any trade is sized. On BE the reading sat in trending territory for the duration, which is why the book came out one-directional and long.
The regime call grants permission; the direction follows from it. Having classified BE as trending, the model opened long with a stop and target set in advance and stayed with each position on those rails. After entry the trade ends in one of a few places, and on BE the spread was fairly even: the profit target closed the most positions, with the stop and the timed exits each accounting for a sizeable share, the pattern of a trend-following book that lets winners reach their targets while cutting the moves that stall.
The illustration above walks the model across BE in a single line: the persistence reading, the trending-or-reverting label it produces, and the long position that label permitted here. The point of the design is to pick the right stance before trading, with the move when the tape persists and against it when the tape reverses. The model's lineage belongs in the frame too. It began as one candidate in an automated search and reached deployment only by clearing backtest and walk-forward checks, the route from idea to live rules behind every figure here.
At a glance
| Gate | Actual | Threshold | Status | Threshold source |
|---|---|---|---|---|
| win rate | 59.57% | >= 70.00% | fail | canonical registry standard |
| max drawdown | 20.30% | <= 5.00% | fail | canonical registry standard |
| sample size | 47 | >= 30 | pass | canonical registry standard |
| total return | 425.33% | >= 100.00% | pass | canonical registry standard |
| expected return | 9.050% | >= 5.000% | pass | canonical registry standard |
| Metric | Value |
|---|---|
| Total return | 425% |
| Win rate | 59.6% |
| Max drawdown | 20.3% |
| Expected per trade | 9.05% |
| Trades | 47 |
These figures come from a backtest of the model on BE, scored against fixed acceptance gates, not from a live track record.
Walk-forward verification
| Metric | Value |
|---|---|
| Walk-forward match | 100% |
| Verified timestamps | 1,739 |
| Signal correlation | 1.00 |
A trade walked through
The walked example is a long held about seven days. With the persistence reading marking BE a trender, the model bought near 55 dollars and the move ran to its target near 71 dollars for +30.09%. It is the classifier working as designed, a trending regime correctly read and a long carried to its target as the move extended.
| Metric | Value |
|---|---|
| Direction | long |
| Entry price | 54.91 USD |
| Exit price | 71.43 USD |
| Hold time | 6.8 days |
| Return | +30.09% |
What the full trade record shows
Across its 47 BE trades the model won 28 and lost 19. The exits spread fairly evenly: 18 reached the profit target, 16 were stopped out, and 13 closed on a time exit.
| Exit reason | Trades | Share |
|---|---|---|
| Take-profit | 18 | 38.30% |
| Stop-out | 16 | 34.04% |
| Time exit | 13 | 27.66% |
A balanced split like this fits a trend-following book: when the regime call was right and the move extended, positions reached their targets, while the stop cut the trends that stalled and a steady share timed out as moves matured slowly.
The biggest winner was a long that ran to its target for +35.95% over about three weeks in mid-2025, opened while the persistence reading sat firmly in trending territory. With long average holds, the standout winners were patient trend rides rather than quick scalps.
The worst trade was a long that turned over for -17.02% in mid-2024; that trade also carried the equity curve toward its deepest drawdown of about 20.30%, the figure that broke the drawdown gate.
No single trade carries the record, and with BE up more than thirtyfold over the window, the book's lesson is the gulf between capturing a trend in slices and simply owning the stock through one of its largest runs.
How does this compare to just holding BE
Over the same window the model was tested on, simply buying BE and holding would have returned many times more. Lining the two up shows whether the rule earned its keep or merely came along on a historic tape. The honest read is that it came along, capturing a slice, and the tiles below put numbers on just how far ahead the stock finished.
| Metric | Value |
|---|---|
| Model total return | +425.33% |
| Buy-and-hold | +3221.22% |
| Difference | -2795.89% |
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 not built on. It is a consistency and replay-integrity test, not proof of live profit. A clean reproduction means the deployed rules behave like the studied ones; it says nothing about whether BE will keep registering as the trending regime the classifier read across the test window.
| Metric | Value |
|---|---|
| Match rate | 100.0% |
| Correlation | 1.000 |
| Alignment | Quiet |
In live trading the model has been quiet so far, with too few signals to set beside the backtest. Until more live trades accumulate, 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 failures come from the regime call itself. When the persistence reading marks a trend but the move stalls or reverses, the model is committed the wrong way, which is how the deepest trade lost 17.02% on a long that turned over in mid-2024. Because the classifier kept BE in trending-and-long territory throughout, it had no short stance to fall back on during pullbacks, and a misjudged regime on a stock this volatile can run against the position well past the stop.
| Metric | Value |
|---|---|
| Losing trades | 19 |
| Worst single-trade return | -17.02% |
| Worst in-trade drawdown | -20.30% |
Three things are worth watching if this ever trades at size. The first is drawdown: the worst stretch in the backtest was about 20.30%, deep enough to break the drawdown gate. The second is regime-call risk, since the whole book rests on the persistence reading being right about whether BE trends or reverts. The third is the gap between live and backtested behavior, the first sign BE is no longer classifying as the trending regime it did in the study.
Risk and honest limits
On this run the model's automated checks logged 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
Where we are
These figures are a backtest, not a live track record. As real trades accumulate, a live-performance section can be added; until then, read every number here as evidence about the rules on saved history.
Sources
This article is based on Stonewell One research, including backtesting, walk-forward verification, deployment monitoring, and model-risk review.Trade-level entries, exits, and holding times come from Stonewell One's backtest of BE over the April 2024 to May 2026 replay window.The model is compared against simply owning BE over the same window.