Automobile Sales Forecasting Based On Time Series Analysis

China has become the world's largest automobile production and consumer market.The development of the automobile industry has led to the development of many neighboring industries,which has great positive significance for stimulating domestic demand and promote export growth.So forecast the automobile sales accurately can not only help to capture the development tendency of the whole market for policy makers but also specific marketing strategies for manufacturers.Existing research on automobile sales prediction mainly focus on the whole automobile market instead of one particular brand.Another drawback is that current research fail to effectively use the online reviews for the automobile,resulting the low accuracy of sales prediction due to the discard of the influence of word-of-mouth on purchase intention.To address these issues,an improved BOAR model is introduced in this paper to predict each specific automobile.The proposed model considers historical sales within various time windows,and incorporates users' opinion for the certain automobile mining from online reviews to predict the sales of that brand.Empirical studies show that this model can accurately predict the sales for an individual automobile brand with better stability.At the same time,we realize that the BOAR model and even the traditional time series analysis model have the following problems:(1)many factors affecting the sales of cars,the influencing factors of different brand of car sales may also be inconsistent,for different car brands it is unreasonable to use the consistent influencing factor as the explanatory variable of the forecast.(2)It is necessary to assume a linear relationship between the influencing factors and the sales volume,and it is difficult to accurately fit the mathematical model for the highly nonlinear relation.Therefore,this paper proposes a general sales forecasting MISF model,which is based on the combination of MARS variable selection process and BP neural network to predict the monthly sales of single automobile brand.The experimental results show that the prediction error of MISF model is 4.04% on average,which is further reduced by 1.49 percentage points compared with BOAR model.This also confirms that variable selection and neural networks play an important role in automobile sales forecast research.The BOAR model and MISF model proposed in this paper accurately predict the monthly sales of single automobile brand.The research results are not only useful for the research of sales forecast in the automobile domain,but also can provide more effective decision making support for the automobile manufacturer's production planning.