Forecasting the Price of Coffee Beans:A Comparison Between Time Series and Machine Learning Modeling Methods

• Main Authors: Chou, Yu-Yu, 周昱妤
• Other Authors: Shao, Yuehjen E.
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/uma959

碩士 === 輔仁大學 === 統計資訊學系應用統計碩士在職專班 === 107 === Coffee is a "black gold" commodity that has been sprung up in recent years. It is the bulk of the world's trading volume after crude oil, and it is the so-called tide culture of young people. Due to the emphasis on health-preserving and health has created a handful of people, and the popularity that is readily available, the degree of popularity is challenging European wines, American colas, Asian teas, not only brewing coffee culture, but also creating coffee Unlimited business opportunities. The importance of coffee price forecasting is mainly due to the cultivation, trade, transportation and marketing of coffee beans to provide employment opportunities for millions of people around the world, and its price fluctuations are more critical to the political and economic stability of the producing countries and the income of farmers. Therefore, the motivation of this paper is mainly to construct a reliable price forecasting mechanism, which can accurately grasp the market price fluctuations, so as to improve production economic efficiency and market competitiveness. The research method of this paper adopts Autoregressive Integrated Moving Average (ARIMA) and Artificial Neural Network (ANN) and multivariate adaptive regression spline (MARS) to Construct a price forecasting model. Mean Absolute Percentage Error (MAPE) was used to compare the accuracy of each prediction model. The results show that the model established by Brazilian and Colombian coffee using ANN method has better predictive performance; Robusta coffee is better predicted by MARS model, and the research results can be used as reference for international organizations, governments and investors.