Why a tail-fattening rule caught only a sliver of MU's run

MU was one of the period's biggest movers, a more than sixfold run, and the question the backtest poses is a familiar one for a choosy model: faced with a move that large, how much could a rule this selective actually keep, and what did the waiting cost it. What follows runs MU's saved history through the model's fixed rules as a study of those rules, not a forecast or a recommendation; the live sample is still too thin to grade.

The quantity this model isolates is how much probability mass has migrated into the extreme tails of MU's recent returns, the portion of the distribution claimed by its largest moves rather than its everyday ones. It abstains while that portion stays slight and engages once it swells to an unusual share, picking up exposure to the upside when the loaded tail sits high and to the downside when it sits low. Bar a single read the loaded tail sat high, on twenty-nine of thirty positions. The 30 trades returned a 53.33% win share, and the aggregate stood at +182.88% on equally weighted entries.

How the model is built, end to end.

What the model depends on is the probability mass lodged in MU's extreme return tails, a fourth-moment quantity that mounts as the largest moves seize a bigger portion of the distribution and falls away once the tape reverts to small, everyday sessions. The model weighs it against the portion this stock usually carries, so only against MU's own history does a level rate as extreme. A meagre portion leaves the model dormant; a true loading of mass into the tails is what clears the way for an entry.

A loaded tail nominates a trade; it does not dictate one. The model proceeds only after a parallel directional check lines up, which stops it from chasing any single large session on its own. With the two in line it books an entry whose stop and target are fixed up front. From there the exit comes one of a few ways, and for MU the most frequent was the trade closing as the mass bled back out of the tails, with stops a close second, a costly pairing that marks a model often caught too soon.

Take the illustration as the model traced along MU from one end to the other: the tail-mass reading to begin, the gate set across its path, the resolved trade to finish. Not one stage is fancy; the entire point is patience, a holding-back until probability mass loads into the tails and the parallel direction check lines up before any money is committed. The rule set's lineage deserves a mention as well, since it was singled out of a large field of machine-suggested options and earned its keep on backtest and walk-forward evidence, the line from idea to deployment that sits under each number reported.

At a glance

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

Walk-forward verification

A trade walked through

The walked example is a long held about twelve days. The tail-weight reading climbed to an upward extreme near 363 dollars, the direction read agreed, the model bought, and the position reached its profit target near 457 dollars for +26.16%. It is the model at its best, a patient entry on a fattened reading and a clean exit at the target once the move had run.

What the full trade record shows

Across its 30 MU trades the model won 16 and lost 14. The exits split three ways: 13 closed once the reading thinned back out, 9 were stopped out, and 8 reached the profit target.

A split with stops taking nearly a third of the book is the signature of a model that fattened into violent extremes and was often early: many positions were knocked out before the move arrived, while the survivors closed either at the target or as the reading thinned. The near-even win count tells the same story.

The biggest winner was a long that ran to its target for +26.77% over about three weeks off a low base in September 2024. The patient winners reached their targets rather than waiting on the reading, taking the move while the fattened tails still pushed in their favor.

The worst trade was a long stopped out for -13.65% inside a single day in April 2025; that single loss is also the equity curve's deepest drawdown, and it is the reason the drawdown gate is one the backtest missed.

No single trade carries the record. With MU running more than sixfold over the window, the lesson of the book is everything the model left untouched by waiting for the tails to fatten before it would act, holding a sliver of a very large move.

How does this compare to just holding MU

Over the same window the model was tested on, simply buying MU and holding it would have done far better. Putting the two side by side is how you judge whether the rule earned its keep or merely tagged along, and here the honest answer is that it tagged along, holding a small piece of a sixfold run. The tiles below put numbers on how wide the gap was.

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 proof the model will earn money live. A clean reproduction means the deployed rules act like the studied ones; it says nothing about whether MU will keep fattening its tails the way it did across the test window.

In live trading the model has been quiet so far, with too few signals to set against 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 same place as its trades, the tails. Its sharpest one is being early: the tails fatten, the model buys, and MU keeps dropping, which is how the deepest trade lost 13.65% on a long stopped out in April 2025 after a single day. Because it is long nearly always, it has little defense against a sustained slide, and a stock this volatile can keep moving against a position longer than the reading expects. Its quieter cost is selectivity itself, every stretch it judged too ordinary to trade in the middle of a sixfold run.

Three things are worth watching if this ever trades at size. The first is drawdown, since the worst stretch in the backtest was about 15.26% and the stop-heavy book means losses can land in clusters. The second is the long-heavy exposure, because the model has little way to profit from or hedge a sustained decline. The third is the gap between live and backtested behavior, the first sign the tail-weight reading is no longer marking the same moves.

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