A Multi-model Methodology For Forecasting Sales Of Construction Consumables(Anchor)

With the development of information technology,in the commercial fields,all industries have accumulated a large amount of historical sales data.Through data mining,some industries have discovered new application opportunities and created more sales potential;some industries have achieved accurate logistics forecasting and successfully controlled costs;some industries have captured crisis warnings and managed risks stably.Under the conditions of advanced information technology,making full use of data resources and establishing a set of commodity sales models will provide guidance for enterprises to reduce costs,increase profits,and expand markets.The purpose of this paper is to combine the market macro data and the internal sales data of the author’s company(Company H)to establish a scientific sales forecast method for anchor(a kind of consumable building material,belonging to the upstream industry of the construction industry).At the beginning of the thesis,the author first studied and elaborated relevant research materials of domestic and abroad,then specified a suitable target model: time series model and multi-linear regression model.Subsequently,through the exploration and verification of data examples,the models are continuously improved.In data preprocessing,the influence of inflation is fully considered.CPI is used to eliminate the inflation.And then the figures show their real value.In the time series model,the original ARMA model was gradually upgraded to the ARIMA model with seasonal effects by adding seasonal effects and integration.In the multi-linear regression model,the unit root in them macroeconomic data is eliminated by logarithmic transformation,and a principal component analysis step is added and dummy variables were introduced.Based on the above models,the author successfully established two sets of sales forecasting models,which achieved the expected goal of this paper.By comparing the empirical analysis results of training samples and test samples,this paper draws the following conclusions:(1)The prediction model for the sales of company H’s anchor bolt products using time series model can fit most of the trends,and the true value can generally fall within the 80% confidence interval,and the fitting effect is good.(2)The prediction model for the sales of H company’s anchor bolt products using multi-linear regression model can fit the test samples well,with an error of about 0.8%-2%.(3)The multi-linear regression model is better than the time series model in terms of prediction accuracy.But multi-linear regression largely depends on how accurate the explanatory variables are.If there is an accurate forecast of economic development,it is recommended to use the multi-linear regression model.