Presented at the DLSU Research Congress 2018 De La Salle University, Manila, Philippines June 20 to 22, 2018 A Comparison of ARMA-GARCH and Bayesian SV Models in Forecasting Philippine Stock Market Volatility Jonathan S. Agahan 1,* , Carmelito B. Miral 1,* and Shirlee R. Ocampo 1 1 Mathematics and Statistics Department, De La Salle University *Corresponding Author: shirlee.ocampo@dlsu.edu.ph *Presenters: jonathan_agahan@dlsu.edu.ph, carmelito_miral@dlsu.edu.ph Abstract: Forecasting volatility is vital in the financial field since it represents one of the risk indicators available. Several models are incorporated by several analysts, investors, and traders but no single superior model was found. In this paper, we present Bayesian and Non-Bayesian methods in forecasting the Philippine Stock Market volatility. The Autoregressive Moving Average – Generalized Autoregressive Conditional Heteroskedasticity (ARMA-GARCH) and the Stochastic Volatility (SV) models were introduced in this paper. Since the best representation for the Philippine market is the Philippine Stock Exchange Index (PSEI), it is then used as the primary data for the paper. The assumptions for the ARMA-GARCH models are tested first to ensure convergence, then the selection for the autoregressive and moving average terms are chosen based on the automatic selection in the forecast package in R studio. After which, the parameters for the GARCH model is estimated using the R package rugarch. For the Bayesian SV model, the Markov Chain Monte Carlo (MCMC) sampler, embedded in the R package stochvol, was used to generate forecasts for the volatility. Results show that ARMA(5,0,4)-GARCH(1,1) model performed significantly better than the SV model in forecasting the stock market volatility with forecasts yielding the lowest mean absolute percentage error (MAPE) and root mean squared error (RMSE). Key Words: stock market volatility; Bayesian SV; ARMA; GARCH; MAPE 1. INTRODUCTION In the financial and economic field, volatility refers to the spread of all likely outcomes of an uncertain variable. In this study, the standard deviation of the daily returns of closing stock prices was used. Volatility has been the subject of trading, financial regulation, monetary policy, and macroeconomy. Thus, it is established that volatility is vital to the financial world (Poon, 2015). Although asset returns are of equal importance, risks should be monitored as well. To create sound investment decisions, risks measure potential losses and volatility plays its role as the purest form of risk in financial markets. Volatility also affects the public confidence which in return, provide an impact on the global economy (Marra, 2015). A vast collection of literature regarding volatility can be found in journals. Consequently, several models are also used in forecasting volatility. However, no single superior model was found. In this paper, the researchers compared Bayesian and Non- Bayesian approaches in forecasting volatility, namely the Bayesian Stochastic Volatility (SV) and the Autoregressive Moving Average – Generalized Autoregressive Conditional Heteroskedasticity (ARMA-GARCH) models.