Box-Jenkins vs. artificial neural networks in predicting commodity prices forecasting the export prices of Thai rice

Ace p.

Order the writing of a tailor-made Economy general Essay

Free quote online

Essay Format .pdf

Box-Jenkins vs. artificial neural networks in predicting commodity prices forecasting the export prices of Thai rice

Download

Read an extract

Reader
Abstract
Contents
Extract

Abstract

International commodity prices are determined by supply and demand, and to a large extent, governmental interventions through trade barriers and subsidies. Forecasting rice prices has always been a great challenge to researchers because determinants of supply and demand such as agricultural and environmental factors, meteorological factors, biophysical factors, changing demographics, etc. interact in a complex manner. Among statistical techniques used to predict rice prices, researchers have found the Box-Jenkins method to perform well in predicting agricultural farm prices. To study the underlying forces and structure that produced the observed time series data on Thailand's weekly rice export prices, we first used the Box-Jenkins method to fit the data. Next, we evaluated various aggregate measures of forecast error (the mean absolute deviation, the mean squared error, the mean absolute forecast error, and the root-mean squared error) to assess the performance of the Box-Jenkins models. Then we used the same data to train and cross-validate artificial neural networks. Our findings showed that while both Box-Jenkins and artificial neural networks performed well in forecasting the weekly export prices of Thai rice, the artificial neural networks produce better predictive accuracies in three of the four categories of rice analyzed.

Key words: Forecasting, Neural-Networks, Box-Jenkins, Validation Error

Abstract
Introduction
The Box-Jenkins model
Neural network model
1. Modeling the time series
2. Training patterns
3. The training sessions
4. Results of cross validation
Summary and concluding remarks
References

Get this table of contents for free after login.

Extract

[...] Table 3 Aggregate Measures of Validation Errors (Box-Jenkins) Category Jasmine White Parboiled Glutinous MAD MAPE MSE RMSE In the following section, we used the same data to train and cross-validate the artificial neural networks Neural Network Model 3 Artificial neural networks are nonlinear mapping systems with a structure loosely based on principles observed in biological nervous systems. Artificial neural networks offer many advantages over conventional statistical methods (Shachmurove, 2002). The ANN uses the data to develop an internal representation of the relationship between the variables, and does not make assumptions about the nature of the distribution of the data. [...]

[...] L., “Forecasting The KLSE Index Using Neural Networks”, Omega: International Journal of Management Science, Vol No pp455- US Dollar per Metric Ton US Dollar per Metric Ton W eek W eek Jasmine Rice US Dollar per Metric Ton US Dollar per Metric Ton White Rice Week W eek Parboiled Rice Glutinous Rice Figure Weekly Prices (January to May 2006) - 0.02 - 0.04 - 0.06 0.01 - 0.02 - 0.03 Jasmine Rice - 0.01 - 0.02 - 0.03 - 0.04 White Rice - 0.02 - 0.04 - 0.06 - 0.08 - 0.1 Parboiled Rice Glutinous Rice Figure The Differenced Series of Natural Logarithms 11 Autocorrelation Function for DLJas - 0.2 - 0.4 - 0.6 - 0.8 0.2 - 0.4 - 0.6 - 0.8 - 1.0 Autocorrelation Function for DLwr Autocorrelation Autocorrelation 10 Lag Corr - T - LBQ Lag Corr - 0.01 - 0.13 - 0.09 - 0.03 - 0.02 - 0.02 - T - 0.13 - 1.35 - 0.92 - 0.34 - 0.18 - 0.17 - LBQ Lag Corr - 0.09 - 0.14 - 0.08 - T - 0.89 - 1.44 - 0.81 - LBQ Lag Corr T LBQ Lag Corr 0.02 - 0.03 - 0.05 0.05 T 0.27 - 0.35 - 0.62 0.62 LBQ Lag Corr - 0.04 - 0.01 0.09 - 0.02 - 0.12 - 0.06 - 0.14 T - 0.52 - 0.15 1.04 - 0.24 - 1.46 - 0.75 - 1.59 LBQ Lag Corr 0.10 - 0.11 - 0.02 - 0.04 - 0.06 0.01 T 1.07 - 1.25 - 0.22 - 0.44 - 0.66 0.14 LBQ Lag Corr T LBQ - 0.04 - 0.43 - 0.05 - Jasmine Rice Autocorrelation Function for DLp - 0.2 - 0.4 - 0.6 - 0.8 0.2 - 0.4 - 0.6 - 0.8 - 1.0 White Rice Autocorrelation Function for DLg Autocorrelation Autocorrelation 10 Lag Corr 0.05 - 0.14 - 0.15 0.01 T 0.61 - 1.70 - 1.77 0.14 LBQ Lag Corr - 0.06 - 0.04 - 0.12 - 0.00 0.03 0.02 - 0.00 T - 0.65 - 0.44 - 1.34 - 0.05 0.28 0.18 - 0.01 LBQ Lag Corr - 0.02 - 0.06 - 0.15 0.01 - 0.04 T - 0.26 - 0.64 - 1.60 0.06 - 0.43 LBQ Lag Corr T LBQ Lag Corr - 0.01 0.02 - 0.13 - 0.04 - 0.12 - T - 0.13 0.20 - 1.59 - 0.46 - 1.43 - LBQ Lag Corr 0.01 - 0.02 - 0.01 T 0.15 - 0.21 - 0.07 LBQ Lag Corr - 0.07 - 0.03 0.07 0.09 T - 0.76 - 0.29 0.76 0.96 LBQ Lag Corr T LBQ - 0.01 0.09 - 0.94 - 0.05 - 0.51 Parboiled Rice Glutinous Rice Figure Sample ACF for the Differenced Series of Natural Logarithms 12 Partial Autocorrelation Function for DLJas Partial Autocorrelation - 0.2 - 0.4 - 0.6 - 0.8 - 1.0 Partial Autocorrelation Function for DLwr Partial Autocorrelation - 0.2 - 0.4 - 0.6 - 0.8 - Lag PAC - 0.04 T - 0.58 Lag PAC 0.12 - 0.03 0.02 - 0.12 0.02 - T 1.59 - 0.42 0.24 - 1.57 0.26 - Lag PAC 0.04 0.08 - 0.10 0.06 - 0.01 T 0.48 1.06 - 1.36 0.73 - Lag PAC - 0.04 - 0.03 - T - 0.57 - 0.34 - Lag PAC 0.06 - 0.02 - 0.05 0.07 T 0.73 - 0.29 - 0.65 0.89 Lag PAC 0.12 - 0.04 - 0.11 - 0.02 - 0.13 T 1.58 - 0.47 - 1.38 - 0.22 - Lag PAC 0.09 - 0.11 - 0.01 - 0.01 0.07 T 1.16 - 1.42 - 0.18 - 0.09 Lag PAC - T - Jasmine Rice Partial Autocorrelation Function for DLp Partial Autocorrelation - 0.2 - 0.4 - 0.6 - 0.8 - 1.0 White Rice Partial Autocorrelation Function for DLg Partial Autocorrelation - 0.2 - 0.4 - 0.6 - 0.8 - Lag PAC 0.23 - 0.05 0.05 - 0.12 - 0.09 0.04 T 3.06 - 0.68 0.71 - 1.58 - 1.23 0.50 Lag PAC 0.09 0.12 - 0.02 T 1.12 1.52 - Lag PAC 0.02 - 0.01 0.07 T 0.22 - 0.18 Lag PAC T Lag PAC - 0.09 0.14 - 0.01 - 0.08 - T - 1.22 1.76 - 0.19 - 1.01 - Lag PAC 0.07 0.02 T 0.95 Lag PAC - 0.05 - 0.00 0.11 T - 0.70 - 0.06 1.47 Lag PAC 0.08 - 0.06 T 1.02 - Parboiled Rice Glutinous Rice Figure Sample PACF for the Differenced Series of Natural Logarithms 05/24/ 05/24/02 12/10/02 06/28/03 01/14/04 08/01/04 02/17/05 Actual 09/05/05 03/24/06 10/10/06 12/10/02 06/28/03 01/14/04 08/01/04 02/17/05 Actual 09/05/05 03/24/06 10/10/06 Predicted Predicted Jasmine Rice 05/24/ 05/24/02 White Rice 12/10/02 06/28/03 01/14/04 08/01/04 02/17/05 Actual 09/05/05 03/24/06 10/10/06 12/10/02 06/28/03 01/14/04 08/01/04 02/17/05 Actual 09/05/05 03/24/06 10/10/06 Predicted Predicted Parboiled Rice Glutinous Rice Figure Box-Jenkins Fitting Errors (Through May 2006) 05/03/ 05/03/06 05/23/06 06/12/06 07/02/06 Predicted 07/22/06 Actual 08/11/06 08/31/06 09/20/06 05/23/06 06/12/06 07/02/06 Predicted 07/22/06 Actual 08/11/06 08/31/06 09/20/06 Jasmine Rice 05/03/ White Rice 6 05/23/06 06/12/06 07/02/06 Predicted 07/22/06 Actual 08/11/06 08/31/06 09/20/ 05/03/06 05/23/06 06/12/06 07/02/06 Predicted 07/22/06 Actual 08/11/06 08/31/06 09/20/06 Parboiled Rice Glutinous Rice Figure Box-Jenkins Validation Errors (May 10 through September 2006) 15 Jasmine Rice White Rice Parboiled Rice Glutinous Rice Figure Artificial-Neural Networks Training RMSE Plots 12/10/ 12/10/02 06/28/03 01/14/04 08/01/04 Predicted 02/17/05 Actual 09/05/05 03/24/06 10/10/06 06/28/03 01/14/04 08/01/04 Predicted 02/17/05 Actual 09/05/05 03/24/06 10/10/06 Jasmine Rice 12/10/ 12/10/02 White Rice 06/28/03 01/14/04 08/01/04 Predicted 02/17/05 Actual 09/05/05 03/24/06 10/10/06 06/28/03 01/14/04 08/01/04 Predicted 02/17/05 Actual 09/05/05 03/24/06 10/10/06 Parboiled Rice Glutinous Rice Figure Artificial-Neural Networks Fitting Errors (Through May 2006) 05/03/ 05/03/06 05/23/06 06/12/06 [...]

[...] and Shekhar, S., “Bond Rating: A Non- Conservative Application of Neural Networks”, IEEE International Conference on Neural Networks Fansiri J Forecasting the Rice-Export Price by ARIMA Method. Technical Report, Chiang Mai University. Thailand. Freisleben, B., “Stock Market Prediction with Backpropagation Networks”, Industrial Engineering Applications of Artificial Intelligence and Expert System. 5th International Conference, June 1992, 451-460 Paderborn, Germany. Funahashi, K. (1989). On the approximate realization of continuous mappings by neural networks. Neural Networks 183-192. Geman, S., Bienenstock, E., and Doursat, R. [...]

[...] The weekly rice prices data provided the training and testing patterns for the neural network. For convenience, we transformed the original data by applying the following linear equation: ˆ yt yt ymin ymax ymin In y t is the rice price for week and y min and y max are the minimum and maximum observed values for the entire data set. Note that 0 y t Each training pattern consisted of 9 input and 1 output values. The input and output values are numbers between 0 and 1. [...]

[...] The decomposition method was found to be most suitable in forecasting the prices of Jasmine rice and main-crop rice with broken, while the Box-Jenkins model was best for main-crop rice with 10% broken. In this paper, we analyzed Thailand's weekly rice export prices from January through September To study the underlying forces and structure that produced the observed time series data, we first used the Box-Jenkins method to fit the data. Next, we evaluated various aggregate measures of forecast error (the mean absolute deviation, the mean squared error, the mean absolute forecast error, and the root-mean squared error) to assess the performance of the BoxJenkins models. [...]

pdf