Comparing the forecasting performance of neural network and purchasing power parity: The case of Turkey

a r t i c l e i n f o JEL classification: F31 C13 C45 Keywords: Exchange rate forecasting Purchasing power parity Artificial neural network Investors consider foreign exchange as being among the most significant financial markets. Many discussions regarding economic development, growth strategies and stabilization policies place real exchange rate to play the most important role in the macroeconomic adjustment mechanism. This study compares a structural model and a statistical model, namely, purchasing power parity and artificial neural network models respectively , for the long term forecasting of exchange rates. Monthly data sets for the US dollar during the period of 1986–2010 and euro during the period of 1999–2010 are used. ANN has been confirmed as an effective tool in forecasting exchange rates through the evaluation of the empirical results. A possibility of extracting hidden information from the exchange rates and using this information to predict the future has been investigated by this technique. The average behavior of the above stated loss functions are estimated to form the basis for evaluating the proposed model. Investors consider foreign exchange as being among the most significant financial markets. Nevertheless, the rapid changes in exchange rate over short periods of time in addition to possessing a highly volatile structure make this market an option for which the investors or the hedgers yearn for the determination of effective methods that evaluate the dynamic tendency of the changes and thereby reduce the risk (Chen et al., 2008). The efforts for improving currency forecast accuracy have been catalyzed by this property of the foreign exchange market which is being extremely volatile (Shady and Shazly, 1997). A still important issue that requires improvement is the exchange rate forecasting accuracy even though many financial models, which attempted to explain and analyze the exchange rate behavior have been presented (Thaski, 2004; Yu, 2004). The non-linear structure of the exchange rate behavior has been reported in many research studies. Therefore, the conventional methods utilizing statistical and economical approaches would not be used properly to present forecasts of the exchange rate Several techniques including the purchasing power parity (PPP), interest rate parity, flexibility approach, income–expenditure approach, Mundell–Fleming approach, monetary approach, portfolio balance approach and the balance of payments approach have been proposed for predicting exchange rates. The use of the economic theory is also encountered in the provision of a benchmark for the evaluation of the exchange rate level …