A new model for forecasting VaR and ES using intraday returns aggregation

Abstract

This paper proposes a new risk measurement model that directly incorporate information from high‐frequency data to predict daily Value‐at‐Risk and expected shortfall. In this model, Regular‐Vine copula and Monte Carlo simulation are applied to produce the predicted intraday returns that have nonlinear dependences. And the time‐varying marginal distribution of intraday returns is estimated under the framework of generalized autoregressive score. The model is so named R‐Vine‐Copula‐GAS (abbreviated as RVCGAS). The predicted intraday returns in the same day are added up to obtain the daily return and form its empirical distribution. The risk measurements are then calculated based on this simulated distribution. An empirical analysis is conducted using data from the index in China's stock market and some of its constituents, and the effects of the proposed model and several Benchmark models are compared through some backtestings. The results show that RVCGAS has an advantage over others in predicting tail risk when the risk level is high, since it could cover more risky returns and reduce more costs.