The Decile Tracker’s First Month: Value Led, Momentum Reversed

On July 15 we took every stock MoatMap ranks, all 21,830 of them across 30 markets, and sorted them into ten buckets by StockRank. Decile 1 held the names our model liked most, decile 10 the names it liked least. Then we froze the lists and waited.

The first measurement period closed on August 10, and the results are now live on the Decile Tracker. This post walks through what happened, factor by factor and country by country. But the summary belongs up front, and it is deliberately unexciting: value behaved exactly as a value model hopes, momentum did exactly the opposite, and one 26-day period is far too short to conclude anything from either.

Why We Track It This Way

Every quant shop shows you a backtest, and every backtest looks wonderful, because a backtest is a story told after the ending is known. The decile tracker is the other kind of evidence. The buckets are formed on live rankings, on a published date, before the returns exist. Once a vintage is scored, the rows are frozen: the only way a published number can change is a logged restatement. Nothing is fitted, trimmed, or quietly re-run until it looks better.

The cost of doing it this way is patience. A walk-forward record accumulates one month per month, and there is no shortcut. This first vintage is a single data point. We are publishing it because the discipline of publishing every month, including the awkward ones, is the entire value of the exercise. If we only showed you the staircases that slope the right way, this page would be a backtest with extra steps.

Two scopes appear throughout. The headline numbers use the investable scope: stocks above a $200M market cap with at least $1M of daily traded value, 12,932 names this vintage. The full universe including microcaps is shown on the tracker for completeness, and as we will see, it mostly demonstrates why the filter exists. All returns below are in US dollars, averaged per decile, and normalised to a 30.44-day month. For context, the S&P 500 returned about 2.4% over the same window: a broadly rising market, which is worth remembering when every number below looks green.

The Month in One Table

The StockRank blends three factor pillars: Quality, Value, and Momentum. The tracker scores each pillar alone, each two-way combination, and the full composite, so when the composite moves you can see which leg did the moving. Decile 1 is always the model’s favourite bucket, decile 10 its least favourite, and the spread is D1 minus D10: positive means the ranking pointed the right way this month.

Investable scope, average USD return per decile, normalised to a 30.44-day month, July 15 to August 10, 2026.

Value: A Clean Staircase

If you drew the chart a value model dreams about, it would look like this month’s value deciles: +6.7%, +6.5%, +5.9%, +4.2%, +4.2%, +3.4%, +3.4%, +3.5%, +2.5%, -1.9% from D1 to D10. The cheapest decile earned 6.7% for the month, the most expensive lost money in a rising market, and the eight buckets in between fall away in order.

What makes this staircase worth taking half-seriously despite the tiny sample is that it holds on three measurements at once. The averages decline monotonically, the medians decline monotonically, and the hit rate, the share of stocks in each bucket that simply went up, slides from 69% in D1 to 45% in D10. When a mean, a median, and a hit rate all agree, the result is not being carried by a handful of outliers. It is the whole bucket shifting.

Momentum: The Mirror Image

Momentum produced the same staircase upside down: -2.2%, +0.2%, +1.1%, +2.2%, +3.0%, +5.0%, +6.2%, +6.0%, +7.5%, +9.4% from D1 to D10. The stocks with the strongest trailing price trends lost 2.2% while last year’s laggards gained 9.5%. The hit rates confirm it is broad rather than outlier-driven: 42% of the highest-momentum bucket rose against 75% of the lowest.

A reversal month is not a surprise in the way it might seem. Momentum is the factor with the best-documented long-run record and the ugliest short-run temper: it works by riding trends, and when a market rotates hard, as this one did into July’s beaten-down names, momentum takes the loss in concentrated form. Single months like this appear regularly inside every long momentum sample in the literature. The honest reading is neither “momentum is broken” nor “ignore it, it always comes back,” but simply: this is the shape of the factor’s bad months, and we happened to launch the tracker in one.

Quality, and the Combinations

Quality landed in between. Its top decile beat its bottom by 2.4 points, which is the right sign, but the middle eight buckets are close to indistinguishable. Quality earns its keep as a partner: the Quality + Value pair was the best composite of the month at a 5.3-point spread, with its top decile up 7.8%.

The combinations mostly behaved like their ingredients. Anything carrying a momentum leg inherited momentum’s month: Quality + Momentum was the worst pairing at minus 7.1 points, and Value + Momentum could not quite overcome the drag either. The full StockRank composite, one part each of Q, V, and M, came out U-shaped: +5.3%, +4.8%, +4.5%, +3.9%, +2.6%, +2.0%, +1.6%, +2.3%, +4.0%, +7.5% across the ten deciles.

Look at where the U actually bends, though, because the shape is more specific than “it didn’t work.” Through the first eight deciles the ranking discriminated the way it should: returns step down broadly in rank order from 5.3% in D1 to the 1.6% to 2.3% range in D7 and D8. The break is confined to the bottom two buckets, where D9 and D10 jumped to 4.0% and 7.5%. That is exactly where a momentum reversal should show up in a blended rank: the stocks the composite likes least are disproportionately the broken-trend, cheap-for-a-reason names that this month’s rotation bid up hardest. So the composite ordered most of the universe correctly and lost the month in its tail, which is a more precise, and more watchable, observation than the headline spread alone. Whether that tail behaviour is a one-month quirk or a recurring cost of carrying the momentum leg is one of the specific questions the coming vintages will answer.

What the Microcaps Are For

The full-universe numbers on the tracker look dramatic in spots: the composite’s bottom decile shows a 20.9%-per-month average in the unfiltered scope. The medians tell the real story. That same bucket’s median stock barely moved, and its hit rate was among the lowest of any decile. The gap between a huge mean and a flat median is the signature of lottery-ticket microcaps: a few illiquid names multiplying while most of the bucket goes nowhere. That is precisely the return profile a real portfolio cannot harvest, and it is why the investable filter, not the full universe, is the scope we headline. The full scope stays published because hiding it would be its own kind of curve-fitting.

Fourteen Countries, Fourteen Verdicts

The tracker also slices the vintage by listing country for every market with enough measured names. The dispersion in month one is a useful reminder that “the factor worked” is always a statement about somewhere in particular.

StockRank composite, investable scope, sorted by spread. Countries with fewer than ~100 measured names are omitted from this table but visible on the tracker.

A few observations, offered loosely. The ranking pointed the right way in Germany, Taiwan, Canada, Hong Kong, and Australia, and pointed the wrong way most sharply in South Korea, Switzerland, and Malaysia. The United States, the single largest slice, was almost exactly flat, which is coherent with the global picture: the US is where this month’s momentum reversal was most concentrated, so the composite’s legs cancelled hardest there. China’s negative spread came with its bottom decile up 9.6%, a loose-liquidity rally in exactly the speculative names a fundamentals model ranks last.

And the strongest caveat in this post belongs here: several of these country cells hold barely a hundred measured names spread across ten buckets. Ten or twelve stocks per decile is not a sample, it is an anecdote with a denominator. The country table will become interesting somewhere around vintage six. Today it is published for the same reason everything else is: because we said we would publish all of it.

What One Month Can and Cannot Tell You

It cannot tell you the model works. It cannot tell you the model fails. With a single 26-day observation there is no t-statistic worth printing, and we will not pretend otherwise: the tracker’s own summary page will show significance measures only once there are enough vintages for them to mean something. A value staircase this clean is encouraging, and it is also exactly what a lucky month looks like. A momentum inversion this sharp is uncomfortable, and it is also exactly what an unlucky month looks like. The two readings are symmetrical, and choosing between them today would be mood, not analysis.

What one month does establish is the machinery. The vintages form on schedule, the returns freeze, the restatement path is logged and public, and the results publish whether or not they flatter us. The August 10 vintage, 22,653 stocks this time, is already formed and sealed. It gets scored in mid-September, and the post about it will be written the same way whichever direction the staircases slope.

The full staircases, medians, hit rates, every combination, and every country slice are on the Decile Tracker, with the methodology written up on the methodology page. The rankings being tracked are the same ones live on Ranked Stocks every day.