multiple-testing

Statistical inference for Sharpe ratios: probabilistic Sharpe ratio, minimum track record length, and FDR/FWER corrections for screening many strategies

Repository for R and Python packages and reproduction codes in Weighted Conformalized Selection paper

This repository contains a collection of functions to evaluate investment strategies regarding multiple testing concerns.

Fast exact grid search and backtesting of moving-average crossover strategies, with vectorized evaluation and multiple-testing-aware Sharpe diagnostics.

Sequential Hypothesis Testing with e-Values and p-Values

NumPy-only Independent Hypothesis Weighting for covariate-weighted multiple testing, with transparent validation and performance benchmarks.

Monte Carlo audit of multiple-testing procedures under block dependence

Backtest overfitting audit for factor research: probability of backtest overfitting (PBO), deflated Sharpe ratio, point-in-time data, purged walk-forward. Searches published factor libraries and reports what actually survived costs.

A quantitative research platform built for falsification: a validation toolkit written from the papers (purged CPCV, PBO, deflated Sharpe, SPA, MCPT) and 23 pre-registered experiments, failures included.

Statistical toolkit for A/B experiments: inference, SRM diagnostics, power analysis, and multiple testing.

Pre-registered falsification of the commodity carry premium (XS + TS) on 18 CME futures, 2010–2026 — design frozen before data; 0/2 arms promoted; four exchange-data identity traps documented along the way.

Open and reproducible empirical research on cryptocurrency cross-sectional factors, asset-pricing evidence, and robust validation.

What can a small cross-section actually detect? A power-aware evaluation protocol for option-implied return prediction, with a measured (not assumed) standard-error scaling law, signal injection, and a volatility detectability control.

How good does the best of 1000 worthless strategies look? A pure-noise study of multiple testing in backtests.

Five statistical tests every candidate trading edge has to survive before anyone writes code. Extracted from a live systematic-trading stack.

A backtester built so the obvious ways to fool yourself raise an exception: fill-time causality enforced in code, execution realism, and multiple-testing accounting that counts every hypothesis you ever ran.

Reproducible statistical-power simulation suite for adaptive studies with an embedded active-inference agent: multiple-testing corrections, sequential e-processes, and action-loop operating characteristics, using real pymdp inference.

Time-series forecast comparison and uncertainty: chronological backtests, validation selection, dependence-aware inference and online intervals.

A research protocol designed to say no: hashed evaluator, append-only trial log, adversarial audit, sealed one-shot validation. Applied to three 2026 semiconductor-drawdown questions.

Genetic programming over formulaic alpha expressions on public Binance data (34 coins, 6.7 years): delay-1 signals, trading costs, pre-registered train/validation/test splits and an equal-budget random-search control.