Published as found, negatives included
Most quantitative research dies quietly. We publish ours, positive or not, because the discipline that kills a weak signal in backtest is the same discipline that protects capital in production. Each note below is a complete study: the question, the test design, the statistics, and a fixed verdict.
August 2026 series
Risk is forecastable. Direction is not.
We moved part of a single-stock book into Treasury bills when the stock became turbulent. The largest loss became smaller on all 31 names tested. Nothing we tried told us which way the stock would go next, and the book cannot beat the stock it holds.
31 names · 2008–2024
August 2026 series
Start-date luck, engineered out
A backtest starts on the date you choose for it. Live capital starts on the date it starts. Across 29 quarterly start dates, the rebuilt engine ends behind the market three times, where the old engine ended behind nine times.
29 start dates · 6-month hold
June 2026 series
Alt-data signals at institutional scale
We scaled three free alternative-data signals from 10-name panels to a 500-name universe: 18 test cells, tens of thousands of position decisions each. Nothing clears the significance bar, the small-panel anomalies dissolve, and the research line closes with a measured answer rather than an open question.
18 backtests · verdict: closed
June 2026 series
Which way do alternative-data signals point?
Insider buying is good, bad news is bad, an expensive borrow is a warning, or so the textbook says. A symmetric sign test across 58 backtests finds two of the three assumptions are backwards, and the one that survives fails a pre-registered confirmation.
58 backtests · pre-registered
June 2026 series
Does alternative data improve a simple trend model?
Three free alt-data overlays (insider filings, news sentiment, borrow fees) applied to a single-name trend rule, scored with honest conditional statistics. No overlay adds significant return, and the trend rule itself loses to holding the index.
36 backtests · sweep