Research And Application Of Economic Time Series Forecasting Method

Demand forecasting, or sales forecasting, is the activity of estimating the demanding quantity of a product or service based on the analysis of its past sales data. Demand forecasting is the foundation for optimizing the entire supply chain, for according to its results enterprises can specify the sales plan, inventory management plan and production schedule in the future. With the popularization and development of computer science, demand forecasting soft-wares has gradually became an essential part of business management. An excellent demand forecasting software will help enterprises predict future demand quickly and make optimal decisions, so as to ensure the enterprises operate in high efficiency.The accuracy of forecasting result is the vital standard for judging a demand forecasting software. The current demand forecasting methods were introduced in this paper, and based on a deep-going study of the time series forecasting algorithms, this paper put forward two new forecasting algorithms, and proved their validity by data experiments. The author, as one of the major researchers of the methods, took part in the designing and implementation of the DAP (Demand Analysis and planning) system designed by Beijing UNITY Co., in which it adopts these two forecasting algorithms.This paper includes the following aspects:1. Introducing current data analyzing technique and forecasting algorithms. Time series forecasting method is the most widely-used method at present, which generally includes univariate forecasting and mutivariate forecasting. This paper introduced two kinds of common univariate economic time series analyzing methods:X-12 Seasonal Adjustment and Unit Root Test, and introduced commonly-used univariate forecasting models as well, such as SMA modeh ES model and ARMA model. For mutivariate forecasting, we mainly talked about regression analyzing method.2. Proposing a new univariate time series forecasting algorithm based on decomposition. By using X-12 Seasonal Adjustment method, it firstly decomposes the original series into trend, season and irregular factors and tests whether it has seasonal characteristics, then, checks its trend characteristics by Unit Root Testing. Based on the test results of seasonal and trending features, it predicts each factor under decomposion, and after that, combining the forecasting results of each factor and gets the final result. The paper took a simulation experiment to "M-3 competition" in this way, in which the method showed its favorable forecasting accuracy.3. Combining traditional regression analyzing method with cluster analyzing, improving the accuracy of regression analyzing forecasting. First, it classifies the historical data by cluster analysis, then chooses partial data from each category cluster to compose a whole data sample. Afterwards, making a regression analysis of the data sample to build the regression model, and carry on forecasting based on the fixed model in the end. This paper compared the new method with simple regression analyzing forecasting method by data experiment to prove the method its validity and accuracy.4. Illustrating the requirements analysis of DAP system in detail and its overall designing process. First, we analyzed the functional and non-functional requirements of the system. Secondly, we divided it into different function modules according to the requirements, and described the linkage among these modules. Finally, we identify the three-tier C/S architecture of the system.5. Discussing the implementation of univariate data analysis module and forecasting module in detail. Univariate data analysis module achieves three data analysis methods, including X-12 Seasonal Adjusting method, Unit Root Test method and one combined by them. Further more, it developed them into the form of C# class library so that they can be used for other systems. Univariate data analysis module does not only get the traditional univariate forecasting model, but also the decomposition forecasting method proposed in this paper. Similarly, in order to improve the reusability of the codes, all forecasting models was also developed into C# class library.6. Discussing the implementation of mutivariate data analysis and forecasting module in detail. Mutivariate data analysis module achieves the regression analyzing method raised by this paper, which is integrated with cluster analyzing. Clustering analyzing adopted the K-means clustering method, and regression analysis method implemented stepwise regression, and they were seperaterly developed into C# class library form for code reuse. When implementing the mutivariate forecasting module, we emphasize regression analysis by using the class library of multivariate data analysis, and after that, we fix the forecasting model, then carry on forecasting.The implementation of the new forecasting method enables DAP system offer prediction results of high precision, so enlarged the databases categories which the system can process. Last year, a bento manufacturing company of Japan made a simulating calculation for all of its products during a certain period by using of DAP system, in which the prediction precision of A type products reach as high as 80%, and the overall precision get over 70%, leading to high praise of customers.