Efficient nonparametric estimation and inference for the volatility function

TL;DR

This paper develops a nonparametric approach for estimating and testing the volatility function in financial data, providing tools that are flexible across models and outperform parametric methods in finite samples.

Contribution

It introduces a nonparametric framework for volatility analysis, including estimation, confidence bands, and symmetry tests, and offers an alternative GARCH(1,1) representation avoiding lagged volatility.

Findings

Nonparametric estimator outperforms MLE in finite samples.

Method provides evidence against or in favor of parametric models.

Consistent estimation and empirical validation on datasets.

Abstract

During the last decades there has been increasing interest in modeling the volatility of financial data. Several parametric models have been proposed to this aim, starting from ARCH, GARCH and their variants, but often it is hard to evaluate which one is the most suitable for the analyzed financial data. In this paper we focus on nonparametric analysis of the volatility function for mixing processes. Our approach encompasses many parametric frameworks and supplies several tools which can be used to give evidence against or in favor of a specific parametric model: nonparametric function estimation, confidence bands and test for symmetry. Another contribution of this paper is to give an alternative representation of the GARCH(1,1) model in terms of a Nonparametric-ARCH(1) model, which avoids the use of the lagged volatility, so that a more precise and more informative News Impact Function can be estimated by our procedure. We prove the consistency of the proposed method and investigate its empirical performance on synthetic and real datasets. Surprisingly, for finite sample size, the simulation results show a better performance of our nonparametric estimator compared with the MLE estimator of a GARCH(1,1) model, even in the case of correct specification of the model.