Tools & Packages

Open tooling

Software, indices, and live dashboards that implement the lab's quantitative methods. Everything here is open-access and citable; install commands and DOIs are listed per package.

01 · Catalogue

What we maintain

Software Packages 2 Open-source libraries implementing the lab's quantitative methods
Live Indices & Dashboards 1 Hosted indices and interactive risk dashboards
Research Code & Formalisations 1 Experiment codebases and machine-checked formalisations behind the papers, released as the research is published
Software Packages · Python package · v0.3.0 Released

GJR-GARCH-X

Student-t GJR-GARCH with exogenous regressors in the variance equation

A small, dependency-light Python implementation of the asymmetric GJR-GARCH(1,1,1) volatility model with Student-t innovations and arbitrary exogenous regressors (the X) entering the conditional variance equation — the specification that appears as TARCH-X in the lab's event-study papers. Built for the cryptocurrency event-study pipeline, where infrastructure and regulatory shocks are encoded as exogenous variance covariates and tested for asymmetric volatility response.

pip install gjr-garch-x
  • Asymmetric leverage effect via the GJR indicator term
  • Student-t innovations for heavy-tailed crypto returns
  • Exogenous regressors in the variance equation (the -X)
  • Maximum-likelihood estimation, NumPy/SciPy only
Versionv0.3.0 DOI10.5281/zenodo.17988193
Software Packages · Python package · v0.1.0 Released

robust-eventstudy

Dependence-robust inference for heavy-tailed cross-asset event studies

Event studies on cross-listed assets violate the independence assumption their standard errors are built on: the same shock hits every asset at once, so abnormal returns are correlated across the cross-section and naive tests reject far too often. This package implements the inference toolkit used in the lab's cryptocurrency event-study work — cluster-robust and copula-based procedures that keep their nominal size when returns are heavy-tailed and cross-sectionally dependent. Extracted from the event-study pipeline so the correction is reusable outside the paper that motivated it.

pip install robust-eventstudy
  • Cluster-robust standard errors for correlated abnormal returns
  • t-copula bootstrap for heavy-tailed cross-sectional dependence
  • Design-effect diagnostics — how much your effective N really is
  • Golden-anchor regression tests pinned to the published results
Versionv0.1.0 DOI10.5281/zenodo.21316120
Live Indices & Dashboards · Live index & dashboard Live

ASRI — Aggregated Systemic Risk Index

A composite early-warning index for systemic risk at the DeFi–TradFi boundary

ASRI aggregates four weighted sub-indices — Stablecoin Concentration, DeFi Liquidity, Contagion, and Regulatory Opacity — into a single 0–100 systemic-risk score for cryptocurrency markets. Validated against the Terra/Luna (2022-05), Celsius/3AC (2022-06), FTX (2022-11), and SVB (2023-03) crises, it correctly flagged the February 2025 Bybit hack ($1.5B, the largest exchange theft in history) as non-systemic — ASRI did not spike, because no contagion channels were active. The live dashboard tracks the index, its sub-components, and the HMM risk regime in real time.

  • Four-channel composite score: Stablecoin · DeFi Liquidity · Contagion · Opacity
  • Alert bands — Low <30 · Moderate 30–50 · Elevated 50–70 · High ≥70
  • Three-regime hidden Markov model: Low Risk · Moderate · Crisis
  • Lead time ≈19d (fixed threshold) / ≈26d (walk-forward) before crisis onset
AUROC (ASRI / D-Y)0.866 / 0.670 AUPRC (ASRI / D-Y)0.298 / 0.121 Precision @ Youden35.2% / 14.9%
arXiv2602.03874
Research Code & Formalisations · Lean 4 proof library Released

FarzullaProofs — Lean 4 Formalisations

Machine-checked proofs for results across the lab's papers

A Lean 4 + Mathlib library formalising core results from the research programme — the Axiom of Consent friction functional, the Replicator-Optimization Mechanism, and supporting results — with formalisation depth tiered (Bronze/Silver/Gold) and tracked per paper. The library builds clean with zero `sorry` placeholders and no added axioms; each paper module ships a map from manuscript theorem IDs to Lean declarations, so a referee can go from an equation in a PDF to the machine-checked statement in one hop.

  • Friction-functional results proved: F = 0 ⇔ σ = 0, positivity, the F ≥ σ/2 lower bound
  • Per-paper theorem maps linking manuscript IDs to Lean declarations
  • Explicit assumptions ledger — no silent axioms
  • CI-checked builds on Lean v4.27.0 + Mathlib
Theorems246 across 15 paper modules Incomplete proofs0 sorry · 0 custom axioms
02 · Contribute

Build with us

These packages are released open-access as the research that produced them is published. If a tool is useful to your work, or broken in an interesting way, we want to hear about it.

research@dissensus.ai