A0548
Title: Robust tests for quantile-based directional predictability
Authors: Natalia Bailey - Monash University (Australia) [presenting]
Bonsoo Koo - Monash University (Australia)
Myung Hwan Seo - Seoul National University (Korea, South)
Abstract: Quantile dependence analysis within and between time series offers valuable insights into the dynamics of economic and financial data, particularly in capturing tail dependencies that traditional correlation measures often overlook. The quantilogram and cross-quantilogram are statistical tools designed to capture dependencies across different segments of the distribution of a time series and the joint distribution of two time series, respectively. This makes them particularly useful tools in financial econometrics, especially when analyzing extreme events. Test statistics of no quantile directional predictability are developed, which are also robust to complex volatility structures and non-linearities commonly characterizing financial time series. The asymptotic properties of the self-normalized statistics are established, and it is demonstrated that they yield valid confidence bands which produce correct size and satisfactory power without requiring the estimation of long-run variance or choice of tuning parameters. Monte Carlo simulation results are encouraging. The empirical application evaluates the evolution of build-up in systemic risk in the US financial market over the period 1993-2024 as measured by 5-year rolling estimates of the (cross-)quantilograms of risk-adjusted returns for three selected securities (JPM, MS, AIG). A significant increase in directional predictability at the fifth quantile is uniformly detected during the Global Financial Crisis.
