Motivation
When fitting GARCH models to financial returns, practitioners routinely need to
assess whether standardized residuals exhibit fat tails. Currently arch provides
no built-in diagnostics for this — users must manually extract std_resid and
compute these statistics themselves.
Discovered this gap while computing GARCH residual diagnostics for research on
fat-tail properties of Indian equity markets:
https://doi.org/10.5281/zenodo.20573035
Proposed additions to ARCHModelResult
excess_kurtosis — sample excess kurtosis of standardized residualshill_estimator(k=None) — Hill (1975) tail index estimator ν̂,
default k = int(sqrt(n)) upper order statisticsvar_ratio(level=0.999) — ratio of Gaussian VaR to model-distribution
VaR, quantifies underestimation from normality assumption
Would you welcome a PR for this?
Motivation
When fitting GARCH models to financial returns, practitioners routinely need to
assess whether standardized residuals exhibit fat tails. Currently arch provides
no built-in diagnostics for this — users must manually extract
std_residandcompute these statistics themselves.
Discovered this gap while computing GARCH residual diagnostics for research on
fat-tail properties of Indian equity markets:
https://doi.org/10.5281/zenodo.20573035
Proposed additions to ARCHModelResult
excess_kurtosis— sample excess kurtosis of standardized residualshill_estimator(k=None)— Hill (1975) tail index estimator ν̂,default k = int(sqrt(n)) upper order statistics
var_ratio(level=0.999)— ratio of Gaussian VaR to model-distributionVaR, quantifies underestimation from normality assumption
Would you welcome a PR for this?