Three Essays on Modelling Business Cycles and Stock Market Volatility in Taiwan with Markov-Switching Models

Autor:	Shyh-Wei Chen, 陳仕偉
Rok vydání:	2000
Druh dokumentu:	學位論文 ; thesis
Popis:	88 This dissertation investigates the business cycles and stock market volatility in Taiwan. The nonlinear and asymmetric behavior, i.e., expansion, peak, contraction and trough, during different phases of the business cycle have long been recognized; see Burns and Mitchell (1946). The traditional linear models cannot adequately explain these nonlinear and asymmetric phenomena because they assume that the growth rate of GDP is linear and stationary, and we need to turn to nonlinear models. As advocated by Hamilton (1989), the Markov-switching model maintains the assumption that time series data may display frequent changes and accounts for such changes through switches in states, where the data-generating process and average duration of each state are allowed to differ. More importantly, the statistical features and identification of the states are not imposed exogenously on the data, but are rather determined endogenously by the estimation procedure. In Taiwan, the leading and coincident indexes have been regularly compiled and published by the Council for Economic Planning and Development (CEPD). These are two particularly important statistical indexes because the Taiwan government actually uses them to monitor the economy and changes her economic policy accordingly. An important question then arises: how useful are these two statistics in dating business cycles and forecasting the future growth rate of GDP? It would be unwise to devise a discretionary policy rule based on statistics that have no predictive power. The primary purpose of the first chapter of my dissertation is to investigate this alluded question. I employ Hamilton''s (1989) original Markov-switching model and the time-varying Markov-switching model developed by Filardo (1994), respectively, to investigate the business cycle and evaluate the usefulness of the coincident and leading indexes in dating the business cycle and in predicting future GDP in Taiwan. By allowing the transition probability to depend upon economic predictors, the time-varying Markov-switching model provides a convenient framework for investigating the usefulness of these two indicators in dating business cycles and forecasting future GDP. More specifically, we can examine the impacts of these two indicators on filtered and smoothed probability estimates and then compare the model-defined chronologies with the CEPD-defined chronologies. The empirical results suggest that these two indexes help date the business cycle in Taiwan and improve precision in predicting turning points. As for forecasting future GDP, the coincident index is useful whereas the leading index is not. From this empirical study we have learned that the traditional univariate Markov-switching model, with or without time-varying transition probabilities, fails to identify Taiwan''s turning points for the post-1990s. The failure of these models can be explained by the fact that the annual growth rate of GDP from 1962 to 1999 has changing trend. The average economic growth rate over this period is 8.20 percent, and there are 38 quarters where the economic growth rate is over 11 percent. However, it is well-known that Taiwan''s economic growth rates are relatively high in the periods of the 1970s and the 1980s, but slowed down for the post-1990s. The average economic growth rates for the pre-1990s and the post-1990s are 8.91 and 6.20 percent, respectively. If we estimate Hamilton''s (1989) Markov-switching model with mean-switching on the GDP growth rate, the estimates of high-growth and low-growth states are 11.48 percent and 6.77 percent in Taiwan, respectively. It is reasonable that the post-1990s will be identified as a contraction phase, although there are three more CEPD-defined contraction chronologies for the post-1990s. The second chapter aims to solving the alluded problem. As noted by Burns and Mitchell (1946, p. 3), a business cycle ''consist of expansions occurring at about the same time in many economic activities, followed by similar general recessions, contractions and revivals...'''' That is, they established two defining characteristics of the business cycle -- the co-movement among economic variables through the cycle and the nonlinearity in the evolution of the business cycle. My strategy is to generalize Hamilton''s (1989) univariate Markov-switching model to the multivariate case. In particular, I pick up the idea of Diebold and Rudebusch (1996) and estimate a multivariate dynamic Markov-switching factor model for a vector of macroeconomic variables. The approach captures both the idea of the business cycle as co-movement in several macroeconomic variables and the asymmetric nature of the business cycle phases. The empirical model is first transformed to the state-space representation, and Kim''s (1994) algorithm is adopted to implement the estimation. The empirical results suggest that the business chronologies identified by the multivariate Markov-switching factor model with GDP, consumption and investment are more consistent with the CEPD-defined chronologies than those of defined by the univariate Markov-switching models, especially for the post-1990. The third chapter turns to analyze Taiwan''s stock market volatility. As the stock market is closely intertwined with the economy in Taiwan, it is important to understanding the stock market volatility behavior. The high persistence in the GARCH model is difficult to reconcile with the poor forecasting performance is well-known in the literature. Diebold (1986) and Lamoureux and Lastrapes (1990) argued that the high persistence may reflect structural change in the variance process. Does Taiwan''s stock market also has above problems? I examine the volatility of Taiwan''s stock market by means of the generalized autoregressive conditional heteroscedasticity (GARCH) and the switching autoregressive conditional heteroscedasticity (SWARCH) models. The daily return of Taiex is used as a summative measure of stock volatility. The empirical results conclude that the SWARCH models do a better job in forecasting than the GARCH models. In addition, for Taiwan stock market there exists a positive and significant leverage effect that a stock price decrease has a greater effect on subsequent volatility than would a stock price increase of the same magnitude. We have identified every episode causing the high volatility state in Taiwan stock market. The estimates attribute most of the persistence in stock price volatility to the persistence of low, medium and high volatility regimes and the high volatility regime is associated with the business recession at the beginning of the 1990s.
Databáze:	Networked Digital Library of Theses & Dissertations
Externí odkaz:	http://ndltd.ncl.edu.tw/handle/56583530859454637652 Zobrazit plný text záznamu