B1733
Title: Efficient Y-indices for regressions with an application of Covid's impact on stock-market liquidity
Authors: Minh Doan - Deakin University, Melbourne, AUstralia (Australia)
Prabesh Luitel - IESEG School of Management (France) [presenting]
Piet Sercu - Faculty of Economics and Business KU Leuven (Belgium)
Tom Vinaimont - Graduate School of Business Nazarbayev University Nur-Sultan (Kazakhstan)
Abstract: The purpose is to study how stock-by-stock fluctuations of illiquidity are related to Covid news. Facing many proxies $y_j$ for (il)liquidity, an index $Y = \sum_j \alpha_j y_j$ is built for a bilateral regression, $\sum_j \alpha_j E(y_j|X) = \sum_k b_k\,x_k$ with residuals orthogonal on the $x$s. Four normalisations for such a best-fit (`Efficient') Y-index---$Var(\sum_k\alpha_ky_k)=1$ ( $Y_1$), $\sum_k\alpha^2_k=1$ ($Y_2$), $\sum_k\alpha_k=1$ ($Y_3$), or $\sum_k|\alpha_k|=1$ ($Y_4$)--- are evaluated and compared, performancewise, with the simple average and the $y$s' principal component. In the application, the simple mean outperforms only in an uneventful period where noise dominates signals. In two turbulent periods with a better information-versus-noise ratio, the constraint on the sum of the weight ($Y_3$) works best and the one on the sum of the squared weights ($Y_4$) worst, but all indices do better than the individual measures even taking into account uncertainty coming from their weighting schemes.
