Hybrid Quantile Regression (QR) and Support Vector Regression (SVR) for Conditional Value at Risk Modelling of Banking Stock Returns in Indonesia Ahmad Hilal. A (a*), Dedy Dwi Prastyo (b), Kartika Fithriasari (c)
Department of Statistics, Faculty of Science and Data Analytics, Institut Teknologi Sepuluh Nopember (ITS)
*hilalilal27[at]gmail.com
Abstract
This research departs from the urgency of understanding stock investment risks in order to minimize the potential for significant losses. Banking stocks have experienced performance fluctuations from 2018 to early 2025, influenced by various factors, one of which is the impact of the COVID-19 pandemic that occurred in early 2020. This study aims to measure the potential losses that banks can experience in the face of market instability, so that it can be a consideration for investors before making a decision to invest in the banking sector. The methods used include Peaks Over Threshold (POT) to calculate Value-at-Risk (VaR), which measures risk univariately. Furthermore, Quantile Regression and Support Vector Regression (QR-SVR) are applied to calculate Conditional Value-at-Risk (CoVaR-QR-SVR), which considers dependencies between banking stocks as well as macroeconomic variables that are thought to affect risk estimation. The results show that the estimated value-at-risk using VaR-POT and CoVaR-QR-SVR increases as the level used increases. ARTO, BBHI, BRIS, BTPS, MEGA, and PNB stocks are identified as having a high risk of loss, indicating that they are more suitable for speculation. ARTO, in particular, has a high variance value, indicating a high level of risk fluctuation and potentially large losses. At the 1%, 5%, and 10% quantile levels, the CoVaR-QR-SVR model is better at estimating risk than VaR-POT based on backtesting results. Validation results using Expected Shortfall and statistical tests such as the Kupiec Test show that the CoVaR-QR-SVR model is more accurate in measuring risk than VaR-POT at all quantile levels tested.
Keywords: CoVaR, Quantile Regression, Support Vector Regression, Banking
