My first project
Nonlinear Beta Tracker
When the market moves 1%, how much does a stock move? And is that the same on a sharp down day as on a sharp up day? Pick a stock or ETF and a benchmark to compare a single straight-line beta with a curve that lets the sensitivity change.
What it measures
Beta is the slope of an asset's returns against a benchmark's. The tracker fits that slope as a straight line and as a curve, and reads the curve's slope at each size of benchmark move. It also estimates a separate beta within slices of the data: the benchmark's worst and best periods, and calmer and more turbulent stretches.
What it cannot tell you
Estimates change with the window, extreme moves are rare, and a few outliers can bend the curve. A better in-sample fit is not a better forecast, and association is not causation. Historical sensitivity is not a prediction or a recommendation. The methodology covers each of these.
About this project
This was the first thing I built, and it is still my favorite. The original was a standalone web app; this version keeps its method (the straight line, the quadratic fit and its slope, the six percentile buckets and the volatility split) and runs it on the same nightly price data as DeanOS. The methodology page lists what was kept and what was corrected, and the source is public.
How it relates to DeanOS
DeanOS's portfolio overview reports a whole portfolio's beta to the S&P 500, with separate slopes on up and down days, from simple daily returns. The tracker looks at one asset at a time against a benchmark you choose, with the curve and buckets described above and log returns by default, as in the original. The two use the same price data, so for the same asset and window their straight-line betas are close but not identical.