Research Library › Model validation

Working paper

When Backtests Lie: A Governance-Grade Validation Framework for Machine-Learning Value-at-Risk Models under Regime Shifts

Affiliation
Calderyn Institute
Status
Working paper

Summary

A governance-oriented validation framework for machine-learning Value-at-Risk models. It extends conventional backtesting with regime-conditional coverage analysis, the Probability of Backtest Overfitting and conformal recalibration, to test whether aggregate evaluation can hide weaknesses in stressed or changing markets.

Methods and topics

Machine-learning VaRRegime-conditional coverageProbability of Backtest OverfittingConformal recalibrationModel governance

Read and reproduce

The full paper, including its data, results and references, is available on SSRN. The code and data used in the paper are published on GitHub.

How to cite

Neupane, N. When Backtests Lie: A Governance-Grade Validation Framework for Machine-Learning Value-at-Risk Models under Regime Shifts. SSRN Working Paper. Available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7652318

Related research

About these papers. This is a working paper by Niraj Neupane (Calderyn Institute). Unless a paper is identified as peer-reviewed or formally published, it has not completed journal peer review, and its findings may change as methods, data and validation evolve. The research reflects the author’s independent views and is not investment advice. See our Editorial Policy and Corrections Policy.