A running collection of small, self-contained statistics and quant experiments. Each one lives in its own file or folder, answers one specific question, and is meant to be read top-to-bottom without needing the rest of the repo.
No fixed layout — each experiment is added as either a single notebook/script or its own subfolder, named for what it tests. Plots, data, and write-ups for an experiment stay next to its code.
- Each experiment states the question it's answering up front, not just the method.
- Where there's a "which is better" question, answer it with a measurable test (e.g. coverage, error, backtest result) — not just a visual.
-
bollinger_normal_vs_t_1.ipynb— fixed-multiplier vs. Student-t-multiplier Bollinger Bands, compared by empirical coverage on candlestick price data. -
'Turbulence-Partitioned Gaussian Mixtures: NIFTY 50 / GOLDBEES' - flagged the market's craziest 20% of days (both assets moving abnormally, together or apart), then described the data as a blend of two bells: a "calm days" one and a "wild days" one. That simple blend matches real returns where the usual single bell curve badly fails because crashes and spikes happen way more often than one bell allows.