Long Run Exchange Rate Determination: An Empirical Evidence of a Simultaneous Equation Model
Abstract
This
paper is connected with the empirical problems involved in the
theoretical exchange rate models. Since the adoption of flexible
exchange rate systems, a number of theoretical models of exchange
rate determination have come into existence, which tried to explain
the exchange rate behavior under different exchange rate systems. A
review of empirical evidence on these models suggest that there are
some empirical problems like poor fit, statistically insignificant
parameters, sign reversals, serial correlation in the residuals,
simultaneity and multicollinearity involved in these models. There
is a need for a model, which is free from all these empirical
problems. In the present paper, a simultaneous equation model has
been developed, which has been tested for all these empirical
problems. The variables used in the model are exchange rate of
Indian rupee against US Dollar, Trade Balance, External Debt, Current
Account Balance, Gross Domestic Product at current market prices and
Money Supply. The causal chain of the proposed model runs from ERt+1
Ù TRt Ù
ERt-1 and ERt Ù
EDt Ù Mt
, where ERt+1
is the exchange rate at period t+1 ERt-1
is the exchange rate at period t-1 TRt
is the trade balance of period t EDt
is the external debt at period t
Mt
is the money supply at period t
From
this causal chain it is clear that the model is a recursive model.
So recursive method in simultaneous equation models has been used to
develop the model. Hence, each equation in the model has been
estimated by using ordinary least squares estimates. To remove the
serial correlation in the residuals, Praise-Winsten generalized least
squares method is used. All the equations are tested for the
empirical problems discussed above and it is found that the proposed
model is free from all these empirical problems. The model is tested
for the out-of-sample forecasting performance by using the data for
the years 1999 and 2000. The results are compared with that of the
random walk model. It is found that the proposed model outperforms
the random walk model. The model can be used by business concerns