The QMJ-TSX paper has two findings: a successful replication of the AQR QMJ-Canada series, and an unsuccessful extension to a price-based proxy on TSX small-caps. The reason I’m comfortable publishing the negative result is that I wrote down the bar for “success” before I ran the test.

This is what people mean by pre-registration. In medicine and psychology it is a formal mechanism; in solo quant work it is usually just a habit, and a rare one. This post is the case for adopting it even when nobody is forcing you to.

The two gates

I committed to two numerical gates in writing before the analysis:

Gate 1 — Replication tolerance. The Sharpe ratio of my recomputed AQR QMJ-Canada series, over the comparable sample, must fall within ±0.30 of the 0.65 figure reported in AFP (2019) Table II for Canada.

Outcome: replicated Sharpe = 0.64. Within tolerance. Pass.

Gate 2 — Calibration of the extension. The Spearman rank correlation between my fundamentals-free paper-Q long-short and the AQR QMJ-Canada series, over the common sample, must be ≥ 0.3 for me to claim paper-Q “captures the same construct.”

Outcome: contemporaneous correlation = −0.03, regression β = −0.08 (t = −0.38), R² ≈ 0. Fail.

These are not p-values. They are pre-committed numerical bands on the actual quantities the paper is making claims about. Setting them in advance is the whole point.

Why this matters more for a solo project, not less

The standard argument for pre-registration is to defend against researcher degrees of freedom — the small choices (sample window, weighting, winsorisation, sector exclusions) that, taken together, let you nudge a borderline result into significance. In a team setting there is at least social friction against this. In a solo setting there is none. You can rerun any cell any number of times, and the only person who would notice is you.

A pre-registered gate creates artificial friction. Once it is written down, moving it requires you to consciously admit you are moving it. That is a low bar, but it turns out to be a meaningful one.

The asymmetry that makes nulls publishable

There is a respectable version of “my strategy didn’t work” and a disrespectable one. The disrespectable version reads:

I tried a bunch of variants. None of them were significant. I’m calling that a negative result.

The respectable version reads:

I committed in advance to the following falsifiable test. The test failed in this specific way. Here is what we learn from the failure.

Only the second version contains information. The first is indistinguishable from a strategy that almost worked, dressed up as humility.

For the paper-Q work, gate 2 is what makes the null informative. The pre-committed claim — “if a price-based proxy captures the same underlying Quality construct, rank correlation with the fundamentals-based version should be at least 0.3” — is the thing being tested. The observed correlation of −0.03 is not “small and inconclusive.” It is “comprehensively below the bar I set.” That is publishable evidence about the limits of fundamentals-free proxies, not a strategy I am still fishing for.

How to actually do it, lightly

Solo pre-registration does not require ritual. A few things that worked for me:

  1. Write the gates into the project plan, not just into your head. I keep them in memories/session/plan.md with timestamps. Anything not in writing didn’t happen.
  2. Make the gates numerical. “Reasonable replication” is not a gate; “Sharpe within ±0.30 of the published number” is.
  3. State the consequence in advance. “If gate 2 fails, the paper’s claim shifts from ‘paper-Q recovers QMJ’ to ‘paper-Q fails to recover QMJ, and here is the per-component decomposition that tells us why.’” The fallback analysis is part of the pre-registration, not a post-hoc rescue.
  4. Don’t tune the gate to the data. The temptation is real. If the observed Spearman is 0.18 and your gate was 0.3, the answer is “fail,” not “0.15 is fine, actually.”
  5. Report the result against the pre-committed gate in the paper. Not just the number — the comparison to the bar.

That is the whole methodology. Five sentences in a markdown file and a discipline about not editing them after the fact.

A second-order benefit

A pleasant side effect of gate 2 failing is that the per-component decomposition and the PCA analysis became the most interesting parts of the paper. If the gate had passed I would have written a competent replication-plus-extension paper that nobody would have cared about. Because it failed, I was forced to ask why it failed — and the answer (“a one-factor post-COVID low-vol unwind that the composite was masking”) is the generalisable finding.

Pre-registration didn’t just protect me from a soft positive result. It pointed me at the real one.

Paper, gates, and the full results table: github.com/faketut/qmj-tsx.