|
Abstract:
|
This study aims to examine the relationship between six independent variables obtained from three groups of background theories and the dependent variable, namely Thai Baht Real Effective Exchange Rate (REER), by comparing the predictive performance for this Thai Baht REER between traditional econometric approach and recently innovated ANN technique.
Monthly data obtained from outside sources is divided into two groups. The first group is used to make the predicting models “learn” or construct the relationship between variables. The second group is employed as a touchstone to compare the forecasting accuracy between the two contesting models.
The result indicates that Linear Regression Model (econometrics) transcends ANN in forecasting performance. This is contrary to the results of many comparative studies concerning these two predictive methods. One reason could be that ANN was a “learning” model. The higher number of training samples, the better forecasting results. In many cases of other researches, the amount of training samples is thousands, or even ten thousands, while the number of training samples in this study is only 198 due to the limitation in historical financial data in Thailand. Thus, the small size of input data restricts ANN from performing at its highest potential.
Key words: Real Effective Exchange Rate (REER), Linear Regression Model, Artificial Neural Networks (ANN), training samples, testing samples.
|