Forecasting Value At Risk With Long Short Term Memory (Lstm)

Abstract

In consideration of the current financial situation, managing risk and forecasting losses play a vital role in financial investment. This thesis aims to apply Long Short Term Memory Model (LSTM) to forecast and estimate Value at Risk (VaR) of a stock portfolio. There are many researches about applying LSTM to forecast in various fields like stock price[1][2], sales[3], weather[4][5], water level of river[6], etc. So LSTM model is expected and trusted in bringing positive benefits and good results when applied to forecast VaR. Traffic Light Backtesting[7] is used to evaluate forecasting accuracy between LSTM and several methods such as Moving Average, and GARCH models[8], besides, the combinations of LSTM with Moving Average, and LSTM with GARCH model are carried out to find out positive or negative effects of LSTM in improving forecast VaR. To make the forecasts, this thesis will focus on predicting volatility and return of portfolio by using LSTM model and others. All in all, this research will point out what LSTM applied in forecasting Value at Risk and how better it affect to the predicted results.