Research lab
Tests that ask whether an edge is real, not how large it looks: a published anomaly measured from scratch, the two ways a search inflates a backtest, and the cost level at which an edge dies.
After a company reports, its stock tends to keep drifting in the direction of the surprise. This pools every earnings event across a basket of names, plots the average market-adjusted return over the next 20 days by surprise tercile, and tests whether the beat-minus-miss drift clears the noise. If the free data can't prove it, the verdict says so.
Collecting reports and post-announcement returns across the basket.
How a backtest fools you
Search enough strategies and the best one looks brilliant by luck alone. This generates hundreds of purely random timing strategies, keeps the luckiest, and shows the trap: naive statistics call it significant, while the Deflated Sharpe Ratio, which knows how many were tried, deflates it back to noise.
Probability of backtest overfitting
Sweeps a real trend and mean-reversion grid on one ticker, then runs combinatorially symmetric cross-validation (López de Prado) over hundreds of in-sample / out-of-sample splits to measure how often the in-sample winner lands in the bottom half out of sample. Near 50% means the selection told you nothing. Splits are purged and embargoed so serial correlation can't leak across the seam.
Fitting the champion on every combination of splits and ranking it out of sample.
Where the edge dies
An edge is only real if it survives frictions. This reruns one strategy across per-side costs from 0 to 100 bp, plots the net-return decay, and solves for the break-even cost where its edge over buy-and-hold vanishes. An edge that dies at 5 bp is an artefact; one that survives 50 bp is believable.
Rerunning the backtest at ten different friction levels.