A0194
Title: Bayesian value at risk forecast using CARE model with an application of cryptocurrency
Authors: Niya Chen - University of Sydney (Australia) [presenting]
Abstract: In the financial market, risk management helps to minimize potential loss and maximize profit. There are two ways to assess risks; the first way is to calculate the risk directly based on the volatility. The most common risk measurements are value at risk (VaR), sharp ratio, and beta. Alternatively, the quantile of the return is looked at to assess the risk. Popular return models such as GARCH and stochastic volatility (SV) focus on modelling the mean of the return distribution via capturing volatility dynamics; however, the quantile/expectile method will give an idea of the distribution with the extreme return value. It allows forecasting the VaR using return which is direct information. The advantage of using these non-parametric methods is that it is not bounded by the distribution assumptions from the parametric method. But the difference between them is that expectile uses a second-order loss function while quantile regression uses a first-order loss function. Several quantile functions, different volatility measures, and estimates from some volatility models are considered. To estimate the expectile of the model, the realized conditional autoregressive expectile (CARE) model is used with the Bayesian method to achieve this. It is examined whether the proposed models outperform existing models in cryptocurrency, and it is tested by using Bitcoin mainly as well as Ethereum.
