Assessing Pollution Risk Using (Asymmetric) GARCH Models and Dynamic Correlation

Exposure to high levels of air pollution represents a major risk factor, as it may produce adverse health effects documented by numerous studies (e.g., Fuller et al., The Lancet Planetary Health, 6(6), e535–e547, 2022). In order to work towards reaching SDG target 3.9.1, that aims to reduction in illnesses and deaths attributed to ambient air pollution, we need to define a model for measuring and predicting risk associated to exposure. Given that ambient air pollution is the outcome of complex mixtures of air pollutants emitted from various activities, an approximation of their combined effects and of impacts on health can be assessed if we could assume some form of independence and little correlation between the pollutants. But these assumptions are quite unrealistic: in fact, in order to be able to control pollution and to measure its risk on health, it’s important to have reliable estimates of the relations among pollutants and of their correlations. In the present paper, for the purpose of estimating time-varying correlations, we consider the extension of univariate volatility models to a new class of dynamic multivariate models called Dynamic Conditional Correlation (DCC) models, introduced by Engle (Journal of Business & Economic Statistics, 20(3), 339–350, 2002): this class of multivariate models uses a nonlinear combination of univariate Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models, with time-varying cross equation weights, to model the conditional covariance matrix of the errors. In other words, for measuring and predicting pollution risk we use GARCH-DCC models developed for measuring and hedging financial risk. The dataset consists of daily standardized concentrations, over two years, on three pollutants, PM10, NO2 and O3, which are interrelated and represent the so-called photochemical pollution factor. The three variables are observed at a single urban monitoring site. As we see from the empirical analysis of our dataset, pollutants concentrations show the presence of significant and different GARCH effects and of interesting trends in their dynamic conditional correlations. The purpose of this paper is to explore whether the use of a multivariate (asymmetric) GARCH-DCC model can lead to a more accurate risk prediction for air pollution. In particular, we aim to determine how positive shocks to the observed pollutants can increase health risk. Interesting results emerge for particulate matter and ozone, both of which have great effects on human health.