Crypto Quant Techniques Lab

A single lab, eight standalone techniques applied to crypto market data — signal detection, market microstructure, portfolio construction, and NLP. Each technique lives in its own numbered folder with its own README, requirements, and tests, and can be run independently. This repo replaces eight separate single-technique repos that used to live on this profile; consolidating them here makes the actual point clearer: these are variations on a shared toolkit (statistical arbitrage, anomaly detection, supervised/unsupervised ML), not eight unrelated projects.

Code and method Full README

This page shows the project's results. The methodology, the design decisions and the limitations are documented in the repository's README.

Five strategies, five negative Sharpes, and the simplest one wins

Net Sharpe and equity curves, five strategies — **How to read it.** Left: annualized Sharpe **net of fees and slippage**, one bar per strategy — note that the axis runs from −0.8 to 0, so every bar is a loss and a *shorter* bar is better. Right: the equity curves behind those bars, all starting at $1.

An improvement that is still a loss

Kalman vs. static OLS equity curve, net of costs — **How to read it.** Both curves are net of transaction costs and start at 1.0. Red is the static OLS hedge ratio fitted once over the whole sample; green is the Kalman filter's online beta, re-estimated day by day from data available up to that day only.

Diversification collapses when it is needed

Correlation by regime — **How to read it.** One correlation matrix per discovered regime, over the same eight assets. The regimes are **unsupervised** — no "this day was a crash" label was used; k=2 was chosen by silhouette score over k=2..6.