The Comparative Analysis of Univariate and Multivariate Time Series Models on Forecasting of Maize Prices

• Main Authors: Shih-Hui Wang, 王詩惠
• Other Authors: Esher Hsu
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/36546207166302294947

碩士 === 國立臺北大學 === 統計學系 === 101 === Maize is one of the major inputs of livestock industry, which directly or indirectly affect livestock market. Recently, energy crisis and climate change has increased the instability of international maize prices. In Taiwan, the maize mainly rely on import, therefore, in order to catch the trend of the domestic animal husbandry development, it is important to grasp the fluctuations of international maize prices. This article aims to apply univariate and multivariate time series models to forecast maize prices, including futures prices and cash prices. According to the characteristics of the trend of corn prices and previous study findings, this study intends to expand the application of the GARCH model and Markov switching model, and find a better forecasting model for maize prices. Univariate GARCH family models, multivariate GARCH model, univariate Markov switching model and multivariate Markov switching autoregressive model are selected for comparative analysis. RMSE is used to compare the forecast performance of the forecast models. The empirical results show that (1) for univariate models: ARMA (1,1)-TGARCH (1,1,1) model has the best performance on predicting futures prices of maize, while ARMA (1,1)-EGARCH (1,1) is the best on predicting cash prices; (2) multivariate models: MSIA (2)-VAR (4) has the best performance on predicting futures prices of maize and MSIA (2)-VAR (4) is the best on cash price. Using GARCH family models for maize prices forecast, the univariate models perform better than multivariate models; whereas, for using Markov state transition autoregressive model, the multivariate models perform better than univariate models. Study results also found that the futures price and the cash price of maize are highly related. Accordingly, this study suggests using multivariate Markov state transition model as the model for forecasting the maize prices.