Time-series data analysis for commodity demand forecasting : construction output modeling and prediction

• Main Author: Peng, Shibo
• Other Authors: Xiao Gaoxi
• Format: Theses and Dissertations
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/69504
• Institution: Nanyang Technological University

Time series is an important type of data in a large number of empirical researches in macroeconomic and financial fields. This dissertation aims to model and forecast a kind of commodity demand say construction output by its historical records and other related economic time series through autoregressive integrated moving average (ARIMA) and generalized autoregressive distributed lag (GARDL) statistical models. Before economic time series are selected in the research, the first thing is to check their stationarity by unit root tests such as augmented Dickey-Fuller (ADF) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) tests. Furthermore, autocorrelation function (ACF) and partial autocorrelation function (PACF) figures are plotted to give better understanding of the stationarity of the time series. Then, the best time lag is explored for the relationship between independent variable and dependent variable so that they have the maximum correlation coefficient in view of simple regression. We can select suitable time series from the related economic indicators including gross domestic product (GDP), its direct components say interest rate, export, and its indirect influenced factor say population. Next, we consider both ARIMA model which is constructed by Box-Jenkins (JK) methodology and GARDL models which are established on a basis of multivariable linear regression and co-integration theory. During this process, model parameters are estimated, model assumptions are tested and corrections are made if necessary. Finally, mean absolute error (MAE), mean relative error (MRE) and root mean squared error (RMSE) are adopted to evaluate the fitting and predictive performance of the employed models. The results show that GARDL model outperforms ARIMA model in terms of fitting and predictive performance.