JinSeo.Cho - Modal Estimation and Inference of Autoregressive Distributed Lag Model Applied to COVID-19 Pandemic Data

<Abstract> This study applies modal regression to the autoregressive distributed lag (ARDL) model and estimates the long- and short-run relationships between integrated series with outliers. Using the proposed modal ARDL (MARDL) methodology, we provide asymptotic distributions of the parameter estimators in the model. We corroborate the theoretical results by simulation of the short- and long-run parameter estimators and Wald tests defined by these estimators. We then apply the MARDL methodology to analyze the short- and long-run relationships between COVID-19 confirmed cases and deaths that are notably characterized by outliers. To assess the dynamic changes in confirmed cases and deaths during the pre- and post-vaccination periods and to examine how the case fatality rate moves together with government response, we examine global data and data from Belgium, Germany, Denmark, and Sweden. Our empirical findings demonstrate that the in-sample forecasts obtained from MARDL broadly outperform those obtained from ARDL and QARDL in terms of the coverage quantile function. Therefore, the MARDL methodology is useful in investigating how the pandemic evolved in response to vaccination and government mitigation efforts.