Multivariate cointegration and error -correction analysis of narrow money demand in Trinidad and Tobago
This dissertation empirically investigates whether a theoretically consistent and temporally stable long-run and short-run equilibrium money demand function for narrow money exists in Trinidad and Tobago over the period 1961:1 to 1991:4. To achieve this objective the maximum likelihood multivariate (deterministic) cointegration procedure developed by Soren Johansen (1991) is used to statistically determine if a stable long-run equilibrium relationship exists between narrow money demand and its determinants, while single equation least-squares error-correction analytical techniques are used to model the underlying dynamic structure of the corresponding short-run function. Our empirical research finds a stationary cointegrating money demand relationship to exist between narrow money balances, the price level, real (permanent) income, and a short-term nominal interest rate. This long-run money demand function is observed to be homogenous of degree zero in prices, and characterized by substantial diseconomies of scale in real narrow money holdings. Further, within-sample temporal stability of the implied elasticities of this function is found to be dependent upon the maintenance of this long-run price homogeneity restriction, as well as the conditioning of the underlying short-run dynamics of the cointegrating system when estimating the cointegrating vectors themselves. In our error-correction analysis we find a statistically well-specified, relatively stable, and theoretically coherent short-run demand equation for real narrow money. The primary determinants of this function include lagged money balances, changes in the volume of real aggregate transactions (proxied by either real gross domestic product or real private consumption expenditure), lagged "excess" real money demand, and either current inflation or movements in nominal interest rates. No clear evidence of a direct contemporaneous effect of current inflation independent of the Fisher effect operating via corresponding changes in nominal interest rates is found. Finally, the ex post forecasting performance of our error-correction models is found to be inferior vis-a-vis previous conventional real and/or nominal partial, adjustment specifications of short-run narrow money demand. Both the root mean-squared error and the Theil Inequality Coefficient are used to compare the relative predictive ability of our competing short-run money demand equations.