A comparison of range value at risk forecasting models

Abstract

Risk forecasting is an important and helpful process for investors, fund managers, traders, and market makers. Choosing an inappropriate risk forecasting model can trigger irreversible losses. In this context, this study aims to evaluate the quality of different models to forecast the Range Value at Risk (RVaR), in both univariate and multivariate analysis, and compare the forecasts to other important risk measures like Value at Risk (VaR) and Expected Shortfall (ES). To assess the performance of both the univariate and multivariate models to RVaR forecasting, we consider an empirical exercise with different asset classes, rolling window estimations, and significance levels. We also evaluate prediction accuracy using Monte Carlo simulations in the univariate analysis, considering different scenarios. We evaluate the empirical forecasts with the score functions of each risk measure. We identified that different models could forecast better different assets, and the GARCH model with Johnson’s SU distribution overcoming the other distributions. We observed the RVine and CVine copulas as better models in the multivariate study. Besides that, we noted that the models with Student’s t marginal distribution have better performance according to realized loss (score function). We identified that even if a model can forecast RVaR well, that does not imply that the same model will forecast other risk measures well.