Lectures
Short, self-contained explainers behind the missions. Each one builds a single idea from scratch in a notebook you can run — the concepts the Lab’s out-of-sample grader is built around. Read one before (or alongside) the mission it supports.
Before you ever run a backtest, you can measure whether a signal has any predictive power with the Information Coefficient (IC) — the rank correlation between today’s signal and tomorrow’s return across the cross-section. We build a real signal and a fake one and learn to tell them apart.
Supports missions 1, 3 · Open in Colab
Test enough signals and some will look brilliant by pure chance. This is the single most important reason backtests lie — and exactly what the ConvexPi Lab’s out-of-sample grading is built to catch. We manufacture the illusion, measure it, and then fix it.
Supports missions 1, 3 · Open in Colab
Real returns aren’t Gaussian white noise — they have heavy tails, near-zero return autocorrelation but persistent volatility clustering, and gain/loss asymmetry. We generate returns with and without these stylized facts (via finmlsim) and measure the difference — the same GARCH engine behind the Lab’s realistic market mode.
Supports missions 1, 8 · Open in Colab
Unwind a big position too fast and you pay huge market impact; too slow and you carry price risk the whole time. Almgren-Chriss makes the trade-off precise, and we trace its efficient frontier (via finmlsim) — the per-order companion to Mission 8’s portfolio-level cost of trading.
Supports mission 8 · Open in Colab