Special Relativity in Financial Modeling

Price paths as worldlines.

Every bar is tested against a light cone. Ordered, timelike bars build mass; when the mass crosses a threshold a black-hole well forms. This is the lab where that idea gets pushed through backtests, Monte Carlo and a paper trader.

Run it in a minute View on GitHub

Research code, not financial advice. The diagram below is the real output of MinkowskiClassifier and BlackHoleDetector from lib/srfm_core.py on the SPY, QQQ and DIA daily closes bundled in the repo.

window

Time runs left to right; up is x = ln(P / P₀) / c, so the light cone sits at 45°. Hover or use the arrow keys on the diagram to read a bar.

Quick start

Five commands, no API keys.

Python 3.12+ and the pinned core requirements. The example runs the physics on hourly bars cached in the repo, prints a table and saves a chart. The output below is what it printed today.

git clone --filter=blob:none https://github.com/Mattbusel/srfm-lab
cd srfm-lab
python -m venv .venv && source .venv/bin/activate
pip install -r requirements-core.txt
python examples/bh_quickstart.py

srfm-lab

$ python examples/bh_quickstart.py
       proxy  bars        from          to  timelike %  BH active %  onsets  all days |ret 5d| %  onset signed ret 5d %  onset hit rate %
symbol
ES       SPY  1542  2020-02-10  2026-04-02       43.45         2.85      23                 1.87                  -0.73             21.74
NQ       QQQ  1541  2020-02-10  2026-04-02       38.94         2.79      16                 2.42                  -0.23             50.00
YM       DIA  1542  2020-02-10  2026-04-02       39.23         1.23       8                 1.76                  -0.38             12.50

Descriptive statistics on one historical sample; not a backtest, no costs, not financial advice.
chart saved to examples/output/bh_quickstart.png

On Windows, activate with .venv\Scripts\activate. Run the tests with pip install -r requirements-dev.txt && pytest tests -q, which is what CI runs on Python 3.12 and 3.13.

How it works

A light cone per bar, a well when order persists.

Two small classes carry the idea. Everything else in the lab is built on top of what they emit.

Timelike or spacelike

MinkowskiClassifier divides each bar's return by a per-instrument speed of light c. Below 1 the bar is timelike, an ordered move; at or above 1 it is spacelike, an anomalous jump. The interval is ds² = c²dt² − dx².

Mass, then a well

BlackHoleDetector adds mass on consecutive timelike bars and bleeds it on spacelike ones. A well forms when mass crosses bh_form after five ordered bars in a row, and points the way the price has been moving.

The rest of the lab

GARCH vol scaling, an OU sleeve, geodesic and Hawking monitors, backtests with Monte Carlo, an Alpaca paper trader, and an idea engine that mines backtest trades and proposes parameter changes. Most of it needs keys or services.

SPY, QQQ and DIA daily closes over the last 750 days with black-hole wells shaded and BH mass under each panel — The quick start's own chart, restyled: every well the detector found on the three instruments over the last 750 days of bundled data.

The SRFM family

Four repositories, one idea.

The paper sits at the origin. The C++ core and the Python SDK implement its formulas; this lab builds trading research on them.

Map of the SRFM repositories drawn as a light cone from the paper

The paper · Python, Rust

srfm-paper-impl

The preprint PDF, scripts and a notebook that regenerate its figures, and a dependency-free Rust reference of the core formulas.

github.com/Mattbusel/srfm-paper-impl → C++20 core

Special-Relativity-in-Financial-Modeling

Price velocity, Lorentz factor, spacetime-interval labels, Christoffel symbols and geodesic deviation on OHLCV bars, with Python validation scripts.

github.com/Mattbusel/Special-Relativity-in-Financial-Modeling → Python SDK

srfm-python

A pandas df.srfm accessor and a Polars wrapper for the Lorentz-factor pipeline. NumPy only.

github.com/Mattbusel/srfm-python → Research lab · you are here

srfm-lab

The black-hole signal, backtests with Monte Carlo, the idea engine and paper-trading research, across nine languages.

github.com/Mattbusel/srfm-lab →

Related: fin-stream, a Rust crate for streaming market data with a lorentz module built on the same transform.

Read this before anything else

Research code, not financial advice.

What the quick start shows

On the bundled 2020 to 2026 sample, the 5-day move after a well forms, signed by the well's direction, averages below zero on all three ETFs. With these parameters the bare signal does not predict the next week. The lab exists to test ideas like this one, including when they fail.

What the rest of the repo is

A research monorepo, not a product. Backtest and paper-trading results are experiments on historical or simulated data and swing widely by period and parameters. Nothing here should trade real money without your own independent validation.