Forecast of Changes in Exchange Rate between Thai Baht and US Dollar Using Data Mining Technique

Abstract

Floating exchange rate system and several factors made it hard to forecast changes in exchange rate on a daily basis. However, taking several factors into account can predict changes in exchange rate. Therefore, this study aims to forecast daily changes in exchange rate between Thai Baht and US Dollar by using data mining technique. 9 algorithms were used to forecast: 1) Naive Bayes 2) Generalized Linear Model 3) Logistic Regression 4) Fast Large Margin 5) Deep Learning 6) Decision Tree 7) Random Forest 8) Gradient Boosted Trees, and 9) Support Vector Machine. Each algorithm was tested accuracy by using 10-fold cross validation with train/test ratio following: 90:10, 80:20 to 10:90 respectively.17 factors were used to analyze data for example, exchange rate between Thai Baht to US Dollar, gold prices, US Dollar price index, crude oil price, price stock exchange index in Thailand, USA, Europe, Britain, Japan, and China. Dataset from January 3, 2002 to April 18, 2019 were used to categorize data. The data were collected from the Bank of Thailand, the Federal Reserve Bank of Saint Louis, th.investing.com, and finance.yahoo. The results reported that Logistic Regression was reached the highest accuracy at 64.86% in train/test portion 80:20, Fast Large Margin was reported at 64.66% in train/test portion 80:20 and 90:10, whereas Logistic Regression was exhibited at 64.61% in train/test portion of 70:30. Decision Tree was shown the lowest accuracy at 57.74% in train/test portion of 20:80. Three factors: US Dollar price index, gold price, and Nasdaq price index were respectively reported as the three most significant correlation of changes in exchange rate. The least factor was Nikkei price index. The result shows that the proposed techniques can be used to support exchange risk management and to forecast other foreign exchange rates.