Penalized quantile regression for ΔCoVaR
Date of Issue2019
School of Physical and Mathematical Sciences
We proposed applying penalized quantile regression for computing ΔCoVaR, which is the change of value at risk (VaR) of the financial system conditional on an institution being under distress compared to median state. Three types of penalized quantile regression: LASSO, adaptive-LASSO and SCAD have been considered. We compared different penalized quantile regression approaches through the a few criteria, which are Granger causality tests, Gonzalo and Granger metric, and Google trend correlation. We find the SCAD the best approach to calculate ΔCoVaR with the United States stock data. Due to the variable selection capability of SCAD algorithm, we derive that TED spread, return of S&P500 index and excess return of real estate industry are three most important variables to predict financial crisis. The advantage of SCAD is further confirmed by market data of Hong Kong and Singapore. Furthermore, to demonstrate the inter-institution correlation in 2009 financial crisis, TENET analysis was applied. Through TENET analysis, we have successfully revealed the main risk transfer from depositories to the insurers, aligning with the current understanding of risk transmission during 2009 financial crisis.
Final Year Project (FYP)