Commodity Price Forecasting Using Deep Learning Techniques: In the Case of Ethiopian Commodity Exchange (ECX)

Abstract

One of the economic development plans of Ethiopia is to enhance the agricultural sector by creating an effective linkage between producer and commodity markets. ECX has been working on creating this effective linkage by providing market information related to commodity prices for producers and traders. But the fluctuation of commodity prices from month to month as well as year to year is the most challenging problem. These changes in commodity prices affects the gross domestic product (GDP) of Ethiopia. Another impact of commodity price fluctuations is on the farmer who produces the agricultural products. They sell their commodities at a low price due to less information about future market directions. This is due to no scientific way of forecasting the commodity price for long term in the future. People simply give their deductions regarding the future price of a commodity as it increases or decreases without using any systematic ways. So, there is importance of intelligent decision-making techniques that follow scientific methods to solve the forecasting problem. The goal of this study was to use the Deep Learning technique, which is a scientific method, to forecast the future prices of the top three export agricultural commodities: coffee, sesame, and green mung beans based on historical data collected from the Ethiopian commodity exchange (ECX). This raw data, collected from the Ethiopian commodity exchange, passes through many preprocessing steps to be usable by deep learning models. Among deep learning techniques, long short-term memory, gated recurrent unit, and bidirectional long short-term memory algorithms which perform well on historical time series datasets are implemented and the results obtained from each of them were compared after hyperparameter tuning. The outperform model was selected. Accordingly, long short-term memory (LSTM) was selected as the best model after being evaluated with regression model evaluation metrics like mean squared error (MSE), root means squared error (RMSE), and mean absolute error (MAE) with very low loss function and high accuracy. The accuracy obtained using LSTM was 99.93%. Therefore, the long-short term-based model was used to effectively forecast the future price of agricultural commodities like coffee, sesame, and green mung beans. Even if this model plays its role in forecasting the prices of the top three most exported agricultural commodities, we recommend extending this study by including factors that affect the prices of those commodities using new technology.