An empirical analysis about forecasting Tmall air-conditioning sales using time series model
Abstract
Time series model is a hotspot in the research of statistics. On November 11, 2015, Tmall platform’s turnover was more than $91.2 billion which caused the attention of scholars both at home and abroad. So this paper aims to forecast sales of Tmall, which is helpful to the enterprises. Research methods are ARIMA model and VAR model. The first model is single-variable model and the later is multi-variable model. In the study, ARIMA model makes the sequence smooth by using two difference operation. In VAR model, five explanatory variables are transformed into one main component. By contrast, VAR model does not give detailed accurate prediction, but ARIMA model does. Therefore, single-variable time series model is more suitable for sales forecast than multi-variable model.
Keywords
Download Options
Introduction
In recent years, the prosperity of electronic commerce makes the enterprises pay more and more attention to marketing strategy, especially accurate sales forecast. Time series model is that using historical data related to past behavior to infer the future behavior of time sequence. So this paper aims to do an empirical analysis about sales forecast using time series model. Industry data such as sales, price, online active stores, online stores clinching a deal, online products and products clinching a deal need collecting. Use MATLAB and EVIEWS software to process the data. The body of this paper includes four parts. The first part is introduction. The second part is model definition about ARIMA model and VAR model. The third part is modeling and forecasting sales. The last part is conclusion.
Conclusion
As single variable model, ARIMA model establishes equation by calculating ACF and PACF, the residuals included. According to akaike information criterion, the model structure is identified as ARMA (2,3). It is very precise and credible. With regard to multi-variable time series model, it did principal component analysis to reduce dimensions. Five explanatory variables are transformed to one main component. VAR model has two variables. On the contrary, prediction accuracy of VAR model is not high. It is suitable for macro-economic analysis.
This paper studies the issues of sales forecast, which belongs to the category of microeconomic. It comes to a conclusion that single-variable time series model is more suitable for this problem, especially the stochastic time series models.
Although we find an ideal time series model, sales forecast in November may have too big error. This is a problem of structural breaks, which need further research.