AN ECONOMETRIC ANALYSIS OF THE SIMEX CURRENCY FUTURES MARKET

Abstract

The thesis investigated the efficiency of the Currency Futures market in the Singapore International Monetary Exchange (SIMEX). The weak sense of market efficiency was tested, with the random walk model being used as the benchmark for comparing the univariate and multivariate models. An outline of the methodology involved in both univariate and multivariate modelling was presented. Aspects of univariate modelling included stationary and non-stationary time series, properties of stationary processes, the autocorrelation function, autoregressive (AR) processes, moving average (MA) processes, autoregressive moving average (ARMA) processes, integrated autoregressive moving average (ARIMA) processes, regression diagnostics, and forecasting. The discussion of multivariate modelling centred on cointegration and vector autoregressive processes. Areas of cointegration discussed include unit root testing, orders of integration, and estimation and testing for cointegration. A multivariate analysis of the data was performed, commencing with augmented Dickey-Fuller (ADF) tests for three individual Futures settlement prices. The ADF tests demonstrated that all three currency Futures were nonstationary variables. Their first differences were, however, found to be stationary, suggesting that all settlement prices of the three currency Futures were I(1) variables. Tests of cointegration were applied to nine cointegrating regressions constructed with one settlement price, normalized as the dependent variable, regressed against one or two other settlement prices. These tests provided no evidence that the three settlement prices were cointegrated. Testing for a VAR relationship involved estimating a VAR model for the first difference of the settlement prices. The results, however, could not establish the existence of a VAR relationship, since a VAR system could not be estimated. Univariate analysis of the data involved fitting several MA, AR and ARMA models to the first differences of the I(1) data. All models fitted to the data were found to be ARMA models. The mean absolute error (MAE) for each of these models was calculated to test their performance against the benchmark of a random walk model. The results provided unanimous support for the time series models being superior to the random walk model. Unlike the results obtained with multivariate modelling, univariate time series models provided some evidence to reject the Efficient Market Hypothesis of the SIMEX Currency Futures market. Two main limits to the study were highlighted: (i) the use of logarithms to transform the data; and (ii) the inadequacies of the Box-Pierce and Ljung-Box statistics in identification tests of univariate models. Several studies of these limits were discussed, namely Granger and Hallman (1991), Franses and McAleer (1995), and Hall and McAleer (1989).