The evidence

Every headline number on this site, the threshold it was judged against, and how each one came out — including the bars it cannot clear and the ones it cannot score. Nothing on this page is typed by hand — it is read from the backtest artifact and the placebo study in the repository, so it cannot quietly go stale.

Backtest run 2026-10-09 Commit 01b1b83 9,645 companies 69 quarterly rebalances 2009-03-27 to 2026-04-09
These are backtested results, not a track record. They describe what the ranking would have done on historical data that was reconstructed to contain only what was publicly known on each date. They are not money that was made, they are hypothetical, and a backtest is the weakest form of evidence there is. Everything below is arranged to help you discount it correctly rather than to persuade you.
The product

The Valquo Index, measured on its own construction

$10B large-cap tier, score-weighted, 8% position cap, 0.30 no-trade band. This is the book the site actually serves, and these are its own backtested figures — net of modelled trading costs.

Over SPY
+2.2177points a year
18.0169% a year against SPY's 15.8226%. vs SPY total return.
Over its own basket
+4.6177points a year
vs an equal-weighted basket of the same large-cap tier, which earned 13.4226%. This is the like-for-like comparison: same universe, same period, only the weighting differs.
Net Sharpe
0.97
Worst drawdown -28.9973%. Turnover 2.81× a year at a realised 9.4 bps one way.
In a regular taxable account this book has historically landed below SPY. Both treatments are the SAME run with the tax rates as the only knob, so the 5.5133 pp a year is a clean difference on identical lots and trades -- and it is LARGER THAN THE BOOK'S ENTIRE 2.2177 pp margin over SPY, so after tax a regular brokerage account lands BELOW SPY. A quarterly book turning over 2.81x a year realises almost everything inside a year, so almost all of it is taxed at the short-term rate.

These figures are measured on the corrected 9,645-name point-in-time panel, 69 quarterly dates (2009-03-27 to 2026-04-09). The previously published figures came from the 2,531-name point-in-time panel, 69 quarterly dates (2009-01-15 to 2026-01-28), and the two panels share ZERO rebalance dates — the grid is derived from the universe's own trading calendar, so widening the universe 3.8x shifts every date. So each figure moved under a universe change AND a grid change, and the two are not separated.

The panel's forward returns carry no dividends, so dividend income is absent from the return and dividend tax is absent from the after-tax figure. That understates a real taxable investor's drag, most of all for a large-cap book.

Source: INDEX-BOOK, 2026-10-07 — data/free_analysis/INDEX_BOOK_CORRECTED.json.

The forward record

What has actually been recorded so far

The figures above are a backtest. This is the live record, and it is 61 recorded days long.

Recorded days
61
2026-07-31 to 2026-11-19. Current vintage 4, opened 2026-08-13 (no-trade band, width 0.30).
First evidence read
2027-03-23
The operational gate asks whether we actually recorded every day we said we would -- not whether the numbers are good.
Verdict
2031
2031-08-13, five years from the vintage's opening.
A few months of record means nothing yet, and that is arithmetic rather than modesty. A forward record this short cannot distinguish a real edge from luck. The contract's own arithmetic says so: at the tracking error this book runs, the first reading worth taking is the six-month operational gate, and the verdict is five years out. Until then the record is a record of whether we are keeping the record, not of whether the strategy works.
Everything BELOW this point is the RESEARCH DECILE, not the Valquo Index. That is the ranking across all ~9,645 companies, equally weighted, top 10% — the object the research measures, and the most powerful test of the ranking because it covers the whole market. Its numbers are higher than the Index's for three reasons that have nothing to do with skill: it holds every market capitalisation rather than a $10 billion tier, it weights every holding equally rather than by score with an 8% cap, and it is measured gross of the trading costs the Index's figures are net of. They do not earn the same thing.
The test that matters most

We shuffled the signal 100 times and re-ran everything

The obvious objection to any backtest is that the machinery itself manufactures the result. So we destroyed the signal and kept the machinery: within each rebalance date the scores were randomly reassigned between companies — preserving the distribution, the missing data and the correlation between themes, destroying only the link between a company and its own score — and the whole pipeline was run again, 100 times, on seeds 1000..1099.

Every alpha below is gross of trading costs — the same basis as the noise runs it is compared against, which is what makes the comparison fair. The net figure is in the costs section.

Grey is where those 100 deliberately worthless runs landed. The dark line is the real result.

Top-decile alpha, a year beat all 84 noise runs
real 6.07%
noise bar
Typical noise run -0.01%  ·  best of 84 noise runs 1.1%  ·  real 6.07%
Top-decile alpha, t-statistic beat all 84 noise runs
real 4.41
noise bar
Typical noise run -0.01  ·  best of 84 noise runs 1.69  ·  real 4.41
Long-short spread, t-statistic beat all 84 noise runs
real 4.59
noise bar
Typical noise run -0.01  ·  best of 84 noise runs 2.44  ·  real 4.59
Decile ladder ordering beat all 84 noise runs
real -0.96
noise bar
Typical noise run -0.04  ·  best of 84 noise runs -0.68  ·  real -0.96  ·  lower is better here: −1.0 is a perfectly ordered ladder

All 4 of these beat every single noise run outright — not one worthless signal matched the real result on any measure. That is a stronger reading than the one this page carried before, and it is worth exactly as much as the three sentences that follow it, so read them.

The noise runs shown are the 84 of 100 that did NOT adopt a tuned weighting, matched to the deployed flat-weight book this page is about. The 16 that adopted are excluded rather than pooled, because adoption is worth about +1.4 of a t-statistic even on a signal known to be worthless, and pooling them would move the bar.

This is not the same measurement that used to be on this page. It is the same model and the same code on a wider universe — 9,645 companies instead of 2,531 — and because the rebalance calendar is derived from the universe's own price history, the old and new panels share 0 of their 69 rebalance dates. So the figures moved under a change of universe and a change of dates, and this project has measured that the dates alone can move the long-short t-statistic by 0.814 (2.703 to 3.517 across seven equally valid calendars on one universe). The result above is the right one for this universe; what is not separable is why it moved.

Counts derived at page load from PLACEBO_CORRECTED.json, which retains all 100 draws (pre-registered as PREREG_ma19_ma13_recalibration.md). The "noise bar" is the 95th percentile of those draws. A tie counts against us: a noise run that merely equals the real result is counted as having matched it.
Against what you could have bought instead

Three benchmarks, including the one that flatters us least

The top-tenth ranked by our score, rebalanced quarterly, held for one quarter, equally weighted. Measured against three different comparisons rather than the single most favourable one.

This is the RESEARCH DECILE (all caps, equal-weighted), not the Valquo Index. The Index holds a $10 billion large-cap tier, score-weighted with an 8% position cap and a no-trade band. Those are different books and they do not earn the same thing: the Index's own measured figures are on the Index tab, and they are lower.

BenchmarkIt returnedTop decileDifferencetQuarters ahead
equal-weight universe (incumbent)
uninvestable — you cannot buy this
every name in the panel, cost-free — uninvestable, the number every historical alpha figure in this project used
14.3% 20.4% +6.07% 4.41 70%
cap-weighted panel average
closest investable analogue buildable from the panel itself
14.9% 20.4% +5.45% 2.03 59%
SPY total return over the same windows
what the user's obvious alternative actually returned
16.4% 20.4% +4.02% 1.33 54%

All figures annualised and gross of trading costs. The equal-weight universe is listed first because every historical figure in this project was measured against it — and it is the hardest of the three, because an equally weighted basket of every company in the panel, charged nothing to trade, beat the S&P over this window. The SPY row is the one that answers "versus just buying the index".

And net of trading costs this book does not beat SPY. The 4.02% above is gross; the measured cost of running it is 4.29% a year, which leaves -0.27% — and a separate measurement of the same book through the Index machinery puts it at -1.14 pp a year. Two different constructions, two different numbers, both negative. The gross figure also carries a t of 1.33, so it is not separable from zero in the first place. Nothing on this site claims to beat SPY.

Does the score actually sort?

The full ladder, not just the top

A top decile can look good by luck. The question is whether the whole ranking orders returns. Annualised return by score decile, best-ranked first.

Top 10%
20.4%
Decile 2
18.6%
Decile 3
16.4%
Decile 4
15.4%
Decile 5
15.7%
Decile 6
15.6%
Decile 7
14.3%
Decile 8
13.6%
Decile 9
11.0%
Bottom 10%
2.3%

It sorts, and imperfectly — deciles 4 through 7 are effectively tied and one is out of order. The ordering statistic is -0.964, where −1.0 would be a perfect ladder. What carries the result is the top decile and the bottom two; the middle is noise.

What the average hides

It lost to the market in 30% of quarters

An annualised average is one number standing in for 69 quarters, and quoting it alone is how a strategy that is often behind gets described as though it were always ahead.

Quarters behind the benchmark
21 of 69
Nearly three in ten. If you held this you would have spent a lot of time wondering whether it had stopped working.
Worst single quarter
-4.65%
Relative to the benchmark, in 2009-06-26. The best was +9.86% in 2020-12-30.
Typical quarter (median)
+1.51%
Below the mean of +1.52%, because the distribution is skewed: a few strong quarters pull the average above the typical one.
Worst peak-to-trough, after costs
-37.4%
Driven by one quarter — the Covid crash. A ranking of companies cannot protect you from the market falling as a whole.
Trading costs

How expensive would trading have to get before this stops working?

"Net of costs" asks you to believe our cost estimate. The breakeven does not: it is the cost at which the edge reaches exactly zero, so you can compare it to whatever you think you would really pay.

Breakeven trading cost
123 bps
Per trade, each way. Above this the edge is gone.
Estimated actual cost
77 bps
A margin of about 1.6×. The book turns over 247% a year, so costs are not a rounding error.

After costs the top decile's edge over the equal-weight universe falls from 6.83% to 2.54% a year.

What the cost model does not capture

  • keyed on point-in-time market cap ONLY — no spread, no average daily volume, no price level, no participation rate
  • the equal-weight benchmark is charged ZERO cost while the strategy pays, so every 'alpha versus equal-weight' figure compares a book that trades against one that does not
The part most sites leave out

The bars it is judged against — including the ones it cannot clear, and the ones it cannot score

4 standard thresholds. For the book this site actually runs it clears 2 of the 3 that can be scored on it, and the other 1 cannot be scored on it at all — they are shown as unmeasured rather than dropped. All of them are here because a page that showed only the passing ones would be advertising, not evidence.

ThresholdResultBar
Long-short t vs. this project's own placebo floor
The 95th percentile of 100 shuffled-signal runs. Beating it means the result is bigger than 95 out of 100 runs on a signal known to be worthless. This bar is re-measured on the SAME 9,645-name panel as the result, from the 84 shuffled runs above that did not adopt a tuned weighting. The artifact still carries the previous panel's figure (2.2837), which is not the bar for this one.
4.594 1.364 clears
Long-short t vs. the Harvey-Liu-Zhu multiple-testing hurdle
A bar that rises with the number of tests you have run. We have run a lot. On the previous 2,531-name universe this was the headline's clearest FAILURE (2.62 against 3.29); on the corrected 9,645-name universe the same model clears it, so both bars now pass at once. That is a restatement rather than a new result -- the same composite on a wider universe and a rebalance calendar that shares no dates with the old one -- and the artifact records both sides of the argument either way.
CLEARS the bar derived from counting its own trials, and CLEARS the bar measured against a shuffled-signal placebo. The trial-counting bar prices the best of N attempts, and the model this site runs is flat-weighted and was never tuned, so the tests counted are overwhelmingly alternatives that were rejected. One qualification: on this universe the selection step did prefer a tuned weighting. The flat weights are kept anyway, because the tuned book earns less and because nothing is adopted without the owner approving it. The run's own stored wording for this is out of date and is not shown: it says the headline fails the trial-counting bar, and this run clears it; it says no weighting was ever adopted, and this run adopted one. The figures above are the run's; only that sentence was hard-coded, and the code that writes it now derives it.
4.594 3.362 clears
Probability of backtest overfitting
UNMEASURED FOR THE BOOK THIS PAGE IS ABOUT, and the number beside it is not a pass to quote. This statistic scores the step that CHOOSES between the eight candidate weightings -- it never reads which one was adopted, so it returns the same figure either way. The book this site runs chooses nothing: its weights are flat and were never tuned, so there is no selection for this measure to score and it is not evidence about the deployed book in either direction. Re-measured for the deployed book on 2026-10-13 it comes back as a pure ranking of the eight schemes rather than a probability, so it is reported as unmeasurable rather than filled in. The bar is close to useless here as well: on this panel's own shuffled runs 41 of 100 worthless signals 'pass' it. Shown rather than dropped, because a measure we cannot score for the deployed book should be visible as unscored.
This figure describes the SELECTION STEP over the eight candidate weightings -- not any one book, and not the deployed flat-weight book.
0.0 0.5 max unmeasured
Deflated Sharpe ratio
THE NUMBER BESIDE THIS IS NOT THE BOOK THIS PAGE IS ABOUT. It is computed on whichever weighting the selection step chose, and on this universe that is a tuned one rather than the flat weights the site runs -- the two differ by a factor of about 462, so the figure beside it is not evidence about the deployed book in either direction. The deployed book's own figure WAS measured on 2026-10-13: 0.72 against the conventional 0.95 bar, which it FAILS, and against this panel's own shuffled floor, which it CLEARS on either reading of that floor -- 0.59 across all the shuffled runs and 0.57 across the ones whose weights were left flat, as the deployed book's are -- exactly the relationship the previous universe's 0.79 had to the same two bars, so the finding is unchanged and only the figure moved. Most of that move is this project's own trial counter rather than the data: about two thirds of it is the count of tests rising from 224 to 297, which raises the bar the statistic deflates against.
This figure describes the CPCV-adopted book; the deployed book's own reading is stated beside it and was measured separately.
0.002 0.95 fails
What we tested and threw away

631 recorded tests. The overwhelming majority failed.

The reason a backtest is weak evidence is that you can keep trying until something works. The defence is to write down every attempt, including the ones that failed, so the successes can be discounted by how many shots were taken. Every test below was registered in writing — hypothesis, threshold and direction — and committed to version control before it was run.

Stock-ranking tests
307
Each one raises the bar the headline has to clear. The hurdle they imply is cleared above — and cleared on a count that includes every one of them, which is the only thing that makes clearing it worth anything. The bar only ever rises as the count grows: it was 3.3623 at the 285 stock-ranking tests the published run counted, and it is 3.3843 at the 307 counted today. The headline clears both.
Options tests
324
Which produced no usable signal at all. See below.

Ideas that sounded good and were rejected on measurement

  • Sector-neutral ranking. Ranking companies against their own sector instead of the whole market. Tested three times, on two different panels, rejected every time. Now permanently closed.
  • A machine-learning model over the same inputs. A gradient-boosted tree trained on one half of the history. Out of sample its deciles ran backwards — its top decile underperformed its bottom one. The plain equally-weighted sum beat it in both directions of the split.
  • Tuning the weights. Five separate schemes for weighting the themes — shrinkage, inverse-volatility, recency-weighting and two others. All five rejected. The shipped model weights every theme equally and has never been tuned, because tuning it on this data made it worse.
  • The options entry signal. Our own options alert was tested against simply picking random days, and lost. It is not sold, not recommended, and the finding is published rather than buried.
  • Using the stock score to pick options. Rejected in both the versions we tried. The score describes the underlying company, not the trade.
  • A dozen published academic anomalies. Short-term reversal, idiosyncratic volatility, maximum daily return, low volatility, analyst estimate revisions, return seasonality and others. None replicated here at a threshold set in advance.

And corrections we found in our own work

  • An early version of the panel had an inverted universe in its first third. Fixing it cut the headline alpha by roughly a third. The corrected figure is the one on this page.
  • Five factors were silently empty for the project's entire history — they computed to nothing and raised no error. A coverage guard now fails the run instead.
  • Option strike prices were being compared against split-adjusted share prices, which manufactured a fake +31,000% trade. Repaired at source; no published verdict changed.

None of these were discovered by an outside auditor. They were found, written down and published by the project itself, which is the only reason you are reading about them.

Don't take our word for it

How to check this yourself

Four things you can verify without any special access, in rough order of how much effort they take.

1. Check that the numbers on this page match the artifact

Every figure above is read at request time from BACKTEST_RESULTS.json and MA19_RECALIBRATION.json. Nothing is typed into this page. The run is stamped 01b1b83, dated 2026-10-09 — if that date is old, these numbers are old, and this page will say so rather than quietly refresh the label.

2. Check the arithmetic against a benchmark you can price yourself

The SPY row above says the S&P returned 16.4% a year over 2009-03-27 to 2026-04-09. That is a public number you can look up in thirty seconds. If our benchmark is wrong, everything measured against it is wrong, and it is the easiest thing on this page to falsify.

3. Ask the question that breaks most backtests

The usual killer is look-ahead: scoring a company in 2012 using a number that was not published until 2013. This panel is rebuilt as of each date from filings with their real filing dates, prices with no future adjustment, and institutional holdings lagged by their actual disclosure deadline — and companies that went bankrupt or were delisted stay in rather than being dropped. The How it works page explains the construction and the specific places it is still imperfect.

And one correction to what this page used to say here. It claimed the record was "not a survey of survivors". Dead companies' returns were always kept — that part was true — but the list of companies was itself a 2,531-name subset of a 9,645-name export, and nobody had checked what selected it. It has now been re-run on all 9,645. The headline got worse (top-decile alpha 7.17% → 6.07% a year, both gross of costs so the two are like for like), which is the direction that tells you the check was worth doing. Every figure on this page is from the wider universe.

4. Read the tests that failed

Each one was registered before it ran, with the threshold written down first, and the verdict recorded whichever way it came out. That record is what makes the 307 figure above meaningful rather than decorative — a count of attempts is only honest if the failures were logged at the time rather than reconstructed afterwards.

The strongest evidence is not on this page

The most convincing test this project has run is an out-of-sample replication on international data: the same untuned model, mapped onto a different vendor's dataset, different countries, a different construction and a different period. It works there, and the United States is the weakest region tested. Those figures are derived from a dataset licensed for non-commercial research only, so they cannot be published on a product page. That is a licensing constraint, not a hedge — and a page claiming to show proof should say what it is leaving out and why.

Read this before you use any of it

What this is not

  • It is not a track record. No money was managed. Backtested results are hypothetical and benefit from knowing which companies still existed to be studied.
  • It is not evidence that it will keep working. The result comes from one panel over one period. Published factor premia routinely shrink after they are published, and there is no reason to assume this one is exempt.
  • It is not advice, and it is not tailored to you. A score is a model output about a company, produced without any knowledge of your circumstances, your tax position or your time horizon.
  • It is not a reason to concentrate. The figures above describe a basket of about 411 companies — the top tenth of a typical cross-section — rebalanced quarterly and equally weighted. Nothing here supports buying a handful of names, and a single name can go to zero regardless of its score.
  • The middle of the ranking means very little. Deciles four through seven are indistinguishable from each other. Treating a score of 61 as better than a 57 reads precision into the number that is not there.