Univariate Forecasting of Indian Exchange Rates: A Comparison
International Journal of Computational Economics and Econometrics, Special Issue on: Advances in Computational Economics and Econometrics, 5(3), 272-288. DOI: 10.1504/IJCEE.2015.070616.
Posted: 28 Sep 2014 Last revised: 16 Aug 2015
Date Written: September 17, 2014
Objective of this paper is to search for a univariate forecasting model that can provide the most accurate forecasts of monthly exchange rates of Indian rupee over the period 1994m8 to 2014m4. Random walk model is used as the benchmark model. Using Box-Jenkins methodology, ARIMA structures of exchange rates are identified. The presence of heteroscedastic variance of ARIMA residuals has been accomplished through the re-estimation of ARIMA models including ARCH/GARCH parameters. From the estimated models, in-sample (1994m4 to 2010m7) and out-of-sample (2010m8 to 2014m4) forecasts for 1, 3 and 6 months horizon are generated. Forecasting ability of the estimated models is accessed by forecast error statistics. The paper finds that in general the random walk model outperforms ARIMA and ARCH/GARCH models for forecasting exchange rates of Indian rupee under the in-sample and the out-of-sample period.
Keywords: Exchange rate, random walk, ARIMA forecast, ARCH, GARCH
JEL Classification: F37
Suggested Citation: Suggested Citation