基于ARIMA和GM模型的店铺交易额预测

Store Transaction Volume Prediction Based on ARIMA and GM Models

摘要: 店铺、平台、物流、用户作为一个完整的闭环，每一环节都需要各司其职，才能使得电子商务发展的更好。因此对于商家来说能够利用合适准确的模型来对店铺交易额进行分析和预测，观测店铺交易额有着怎样的变化趋势，对调整价格以及营销策略有着非常重要的关系。店铺交易额数据为时间序列数据，可以采用时间序列模型进行预测，ARIMA模型和GM (1, 1)是应用比较广泛的模型，并且有着较好的预测精度。因此本文基于某店铺2017年6月~2022年6月每3个月即21个时间段的数据建立两种预测模型，并预测后面4个时间段的数据。数据显示GM (1, 1)模型具有更高的预测值，ARIMA模型预测值与真实销售额误差更小，具有更好的预测精度。通过对比预测数据和实际数据我们发现虽然存在突发事件会影响销售额但是这种影响会很快恢复。这对于商家在应对短期不可抗力因素影响时，通过观测交易额的发展变化，应该如何调整店铺营销策略具有一定的借鉴意义。

Abstract: As a complete closed loop, each link of the store, platform, logistics, and users needs to fulfill their respective responsibilities in order to promote the better development of e-commerce. Therefore, for merchants, it is crucial to use appropriate and accurate models to analyze and predict store transaction volumes, observe the trends in store transaction volumes, and adjust prices and marketing strategies. The transaction volume data of the store is time series data, which can be predicted using time series models. ARIMA model and GM (1, 1) are widely used models with good prediction accuracy. Therefore, this paper chooses the store June 2017~June 2022 every 3 months, i.e., 21 time periods of data to do the model to establish two prediction models, and predict the data of the next 4 time periods. The data show that the GM (1, 1) model has a higher predictive value, and the ARIMA model predictive value and the real sales error is smaller and has a better prediction accuracy. By comparing the predicted data with the actual data, we find that although there are unexpected events that can affect the sales, this effect will be recovered quickly. By observing the development of transaction volume, it is meaningful for merchants to adjust the marketing strategy of their stores when dealing with short-term force majeure factors.