Application of time-varying approaches to a financial model

Abstract

In this dissertation, we study about application of time-varying approaches to a financial model. In the first study, we investigated the appropriate statistic criteria for selecting the time-varying coefficient model. The Monte Carlo simulation created the time-invariant and time-varying coefficient data set based on the regression equation. We compared a statistical criteria performance between the time-invariant and time-varying coefficient model to find the statistical criteria that choose the time-varying coefficient model. We found that log-Likelihood (InL), Mean Absolute Deviation (MAD), and Root Mean Squared Error (RMSE) are the appropriate statistical criteria in the model selection method for the time-varying coefficient model. In the second study, we applied the Fama - French five-factor model (FF5 model) by using the time-varying coefficient model to show changing relationships between risks and portfolio return over time; the new model called the time-varying coefficient Fama - French five-factor model (TV-FF5 model). We used monthly returns and risk premium data of the U.S. and Japan portfolios from July 1963 to April 2020 and from July 1990 to April 2020. As a result, we showed a changing relationship between risks and portfolio return over time. Next, we used a residual from the TV-FF5 model to show time-varying volatility by using the GARCH(1,1) and GJR-GARCH(1, 1) models.In the tired study, we applied the Google Trend to show the effect of investor panic in an economic crisis on U.S. portfolio volatility by using the FF5 and GARCH(1,1) model. The economic crisis in these studies was the China-US trade war, COVID-19, and oil prices. First data set, we used monthly data of the U.S. portfolios and Google Trends from 1 November 2019 to 30 April 2020. We showed that the percentage change in the trend of COVID-19 would affect the volatility of Small neutral (SM) Big value (BL) and Big growth (BH).