Method
Why Nullberg exists.
Most published alpha does not replicate.
That is not a provocation. It is one of the most consistently documented results in quantitative finance research across the last fifteen years, from Hou Xue Zhang to McLean Pontiff to Harvey Liu Zhu. Yet new papers keep arriving, new factors keep getting named, and new signals keep getting sold on the strength of an in-sample backtest that nobody else has touched.
Nullberg exists to close that gap. Every paper we cover is coded from its published methodology, run on historical data we source independently, graded on a transparent rubric, and tracked as its live decay unfolds. The verdict is public. The code is public. The data sources are public. If we got it wrong, the path to proving so is right there in the notebook.
The rubric
Replicated
The paper's core claim holds at reasonable tolerances on independently sourced data. Effect size, sign, and statistical significance all match within the bands we document up front.
Degraded
The effect exists but is materially smaller than the paper claimed, or has decayed since publication, or only survives under narrower conditions than the paper implied.
Failed
The core claim does not survive independent replication. We publish failures in full, including the dead ends along the way, because those are the observations that make the archive useful.
Inconclusive
Data access, methodology ambiguity, or sample size prevents a clean verdict. Labeled plainly, not hidden.
What Nullberg will never do
- Accept sponsorships, advertising, affiliate revenue, or paid placement of any kind. Subscriptions only.
- Sell signals, managed accounts, or investment advice. Nullberg is research, not a fund.
- Recommend buying or shorting a specific security. The verdict is on the method, not the trade.
- Hide replication failures. Null results are the whole reason this exists.
Who
Nullberg is a pseudonymous persona. The identity behind the byline does not matter. The notebook does.
Every replication ships with the code that produced it. If a claim is wrong, the path to showing it is wrong is in the repo. If a verdict is wrong, the correction is published with the same rigor as the original.
How Nullberg is built
Nullberg is AI-assisted end to end. Papers are read, code is drafted, and replications are run with an AI operator in the loop. This is not hidden, and it is not an excuse for lower rigor. It is the opposite. AI speed is what makes it possible to replicate at the volume the literature demands, and every replication ships with the code and data sources so a reader can reproduce the verdict independently, disagree with it, and correct it publicly.
The brand makes one promise that an AI byline has to earn rather than claim. The verdict is traceable to the code and the data, not to the voice. If a reader disagrees with a Nullberg verdict, the argument lives in the notebook, not in the byline.