Replication-first research institute
Every claim of alpha,
dissected.
Nullberg reads every new paper in systematic investing, codes it on real data, runs it, grades the evidence, and tracks which claims hold up over time.
Most don't.
Subscribe
Null results are news. Get them in your inbox.
Free weekly: one published alpha claim, replicated and verdict-graded, with the code that got there. Five verdicts are already in the archive. New ones drop weekly.
Paid tier (coming soon): the full factor decay dashboards, replication notebooks, and the archive of verdicts we are actively monitoring.
The method
01
Replicate
Every paper we cover is coded from its published methodology and run on independently sourced historical data. No selection on which results we report.
02
Grade
Each claim gets a transparent verdict. Replicated, degraded, failed, or inconclusive. The code and data path that led there is published alongside.
03
Track
Verdicts are not one-shot. We monitor every replicated factor's live decay and update the archive the moment the evidence moves.
What makes Nullberg defensible
- Open code. Every replication ships with a public notebook on GitHub. Fork it, run it, argue with the result.
- Zero conflicts. No sponsorships, no affiliate revenue, no paid placement. Subscription is the only revenue line.
- Failures published in full. When a replication fails, we publish the failure with the same rigor as a success. Null results are the whole point.
- Living verdicts. Every graded claim is monitored. Decay is logged. The archive is the living record of what still holds.
Start here
Verdict · Failed · New
Accruals, out of sample, 2016 to 2026
Sloan 1996. Primary VW NYSE TTM accruals: +0.883% per month — exactly the canonical 0.87% — but t = +1.85 just below the rubric bar. Two EW sensitivities REPLICATE at NW t = +3.81 and +2.39. The accruals effect has migrated to smaller stocks.
Read the verdict
Verdict · Replicated
Value (book-to-market), out of sample, 2016 to 2026
Fama-French HML 2×3 at +1.698% per month, i.i.d. t = +2.82. The first non-failed verdict, and the most heavily caveated. Regime-driven by the 2021+ rebound (second half t = +3.20), Newey-West t = +1.94.
Read the verdict
Three more failed verdicts
Failed · Profitability
Novy-Marx 2013
+0.50% per month, simple t = +0.93, 2×3 PMU t = +1.74 with first half t = +2.02. Big-cap earnings-quality effect, fades post-2021.
Failed · Momentum
Jegadeesh Titman 1993
-0.13% per month, t = -0.16. Decayed to a statistical null in both sample halves. NW t = -0.26.
Failed · MAX
Bali Cakici Whitelaw 2011
-1.80% per month, t = -2.57. Sign inverted. NW t = -2.18 robust. Concentrated in 2016-2021.
Primers
Synthesis
Four verdicts, four shapes
Reading the first Nullberg data runs as a set. Evidence versus hypothesis, explicitly labeled. Commits the rigor standards the archive now follows.
Foundations
The replication crisis in systematic investing
Three published meta-studies that anchor the entire reason Nullberg exists. Every number sourced.
Five verdicts across five factor classes. One replicated, four failed, and within the four failures there are four different shapes of failure: inverted (MAX), decayed (momentum), underpowered on large-cap (profitability), and underpowered at the canonical magnitude on large-cap while replicating on EW sensitivities (accruals). Every primary now reports i.i.d. and Newey-West HAC t-statistics plus sub-sample stability.