Overview
The public Tensor repository contains a theoretical manuscript and a redacted Python implementation of the mathematical identities it defines. It is deliberately not a production package, forecast service, trading signal, investment recommendation, or empirical claim about any security or strategy.
The framework separates a fair-value layer from an event-scale move layer, explicit scenario probabilities, a price-convergence coefficient, and expected-value, risk, calibration, and sizing quantities. That separation is the central contribution: an implied absolute move is not the same thing as physical probabilities, and a valuation gap is not automatically an event return.
Methodology
Valuation is represented as a range with an explicit uncertainty summary rather than a false point estimate. The event layer maps an options-implied straddle moment into a tail-move budget, while a three-state bear/neutral/bull model receives probabilities explicitly instead of inferring them from the straddle.
The package supports ordered-logit scenario probabilities, separate surprise and reaction channels, symmetric and asymmetric expected-value identities, convergence estimation with measurement-error correction, Brier/log-loss/ECE calibration, sampling-error floors, and uncertainty-aware sizing identities. Every model-specific quantity is supplied explicitly; the public code provides no proprietary defaults.
What it does
- LaTeX manuscript defining the valuation and event-risk framework
- Explicit scenario probabilities and ordered-logit estimation
- Straddle-to-move identities with neutral-band leakage handling
- Symmetric and asymmetric expected-value, variance, and downside metrics
- Fair-value convergence estimation with intercept and measurement-error terms
- Brier score, log loss, ECE, convergence, and sampling-floor utilities
- Uncertainty-aware Kelly and position-sizing identities with hypothetical fixtures