Three Essays on Time Series Econometrics Analysis and Financial Market Applications

Abstract

This dissertation rst revisits the mean reversion properties of relative stock prices when the US and the UK sever as reference countries. We present strong evidence of nonlinear mean reversion of the stock price indices of OECD countries relative to that of the UK. However, the panel nonlinear unit root test failed to reject the null of nonstationarity even when the UK served as the reference country. The results appear inconsistent. Via principal component analysis and extensive Monte Carlo simulations, we demonstrate a potential pitfall in using panel unit root tests with cross-section dependence when a stationary common factor dominates nonstationary idiosyncratic components in small samples. We believe the empirical ndings in this chapter provide useful implications for international asset market participants.

In the second chapter, we studied the impact of exchange rate shock on the commodity prices using VAR (Vector Autoregressive) Model, a forecasting technique in time series analysis. We report that rst, the long-run adjustment of prices is very slow. Prices typically take 8 to 12 months to stabilize except for the oil prices which stabilize in about 4 months. Second, the responses of commodities are heterogeneous. Some commodities, like wheat, cocoa beans, beef, pork, chicken, bananas, oranges, and soft wood, under-correct, i.e. the price elasticities of these commodities are less than one. Others, like corn, lamb, sugar, hide, and crude oil adjust on par with the exchange rate movement. Finally, the prices of the commodities like barley, peanuts, rice, sun ower oil, olive oil, rubber, aluminum, nickel, and coal, over-correct. This might call for price stabilization policy implications especially for the developing countries.

The third chapter deals with the ordering of recursively identi ed VAR models and reports potentially useful facts that show under what circumstances these impulse response functions are robust to this so-called Wold ordering. This adds important technical contributions to the existing multivariate time series model literature.