A Less Volatile Value-At-Risk Estimation Under A Semi-Parametric Approach

In this paper, we propose a two-step less volatile value-at-risk (LVaR) estimation using the generalized nearly-isotonic regression (GNIR) model. The first step of our LVaR estimation is to produce a VaR sequence under the generalized autoregressive conditional heteroskedasticity (GARCH) framework. The second step is to adjust the VaR sequence by GNIR, and the generated estimate is denoted as LVaR. Our empirical tests based on the daily indices of 20 global stock markets suggest that LVaR outperforms other VaR estimates under the classic equally-weighted and exponentially-weighted moving-average frameworks. Furthermore, we show that LVaR is not only less volatile, but also performs reasonably well in various backtests