Forecast of key indicators of retail network in time

The purpose of the study is to forecast the commodity demand, within which key indicators related to the operation of the retail network are applied. The relevance of the study is determined by the need to forecast the indicators significant for retail trade development to increase the planning efficiency of trade organization activity. The importance to forecast the future values of the indicators under consideration - the temperature regime and the number of checks - has been indicated. These indicators are significant for adequate forecasting of demand and for solving other management tasks as well. The range of the analysis methods of time series has been determined. These series have been used to solve the tasks of forecasting demand for the retail network. The approach of the state-space model is considered in almost every method. The theoretical basis of each method has been described to illuminate a sufficient variety of the applied mathematical tools. The fact that conventional (ARIMA, exponential smoothing) and modern methods used by large IT companies (Facebook and Google) are among the applied methods of forecasting has been emphasized. The choice of the prediction quality metric for the problem has been justified. The metrics are a square root of the root-mean-square error and the absolute error in percentage. A set of daily data about the amount of checks in the retail network of Izhevsk, as well as the average daily temperature conditions in the geographical area of the city have been used as the input data to make the forecast. To make short-term forecasts, the initial sample is suggested to be divided into a training sample and a test one in the proportion 9 to 1, due to the short-term of the forecast. The importance of the temperature series indicator for the activity of a retail store and for the dynamics of consumer demand has been characterized. The issue to forecast temperature accurately only on the basis of temperature time series has been discussed in the study. The models for temperature series have been calculated and the quality indicators for each model have been estimated. The value of the indicator of the number of checks to demonstrate the activity of retail trade has been described. A number of external factors affecting the dynamics of the number of checks have been listed in the present research. Among them are days of the week, whether it is a pre-holiday or a holiday. We have made conclusions about the high efficiency of the composite forecast with several methods on the basis of the predictive modeling of the check amount. Even with the help of the arithmetic mean of the forecasts for several methods, it is possible to create a more accurate prediction than for each method separately. Our further research will concern the improvement of the tool and the development of a computer-based system to forecast commodity demand.