#TITLE_ALTERNATIVE#

• Main Author: DITA AGISTIA ( NIM: 10113043), MAULIA
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/23003
• Institution: Institut Teknologi Bandung

Return shows the dynamic property that allow the behaviour change from one state <br /> <br /> <br /> to another state or called regime-switching phenomenon. Threshold stochastic <br /> <br /> <br /> model is developed to accommodate this phenomenon. Threshold is used as <br /> <br /> <br /> delimiter between regimes. In other words, for different regime, it is allowed to <br /> <br /> <br /> have another different time series model. In this final project, used variant of AR <br /> <br /> <br /> model and GARCH model with a threshold value, TAR(p) and TGARCH(1,1). <br /> <br /> <br /> In constructing both models, parameter estimation is required. The estimation <br /> <br /> <br /> methods are least square method for TAR(p) and maximum likelihood method for <br /> <br /> <br /> TGARCH(1,1). <br /> <br /> <br /> TAR separates the return into several regimes and used for return prediction. <br /> <br /> <br /> Different from TAR, regimes on TGARCH are separated based on the volatility. <br /> <br /> <br /> Volatility is an important aspect of the study of return, so that, besides having a <br /> <br /> <br /> good return prediction accuracy, a model also needs to have an ability to accommodate <br /> <br /> <br /> volatility aspect. In this final project, GARCH(1,1) and TGARCH(1,1) <br /> <br /> <br /> volatility models are used to predict the volatility. Using three stock indices, <br /> <br /> <br /> TGARCH(1,1) model gives more accurate volatility prediction than GARCH(1,1) <br /> <br /> <br /> model. It indicates that a threshold in TGARCH(1,1) model can improve the <br /> <br /> <br /> accuracy of the volatility prediction result.