CREATION OF SOFTWARE USING AUTOREGRESSION MODELS FOR TIME SERIES ANALYSIS AND FORECASTING

Abstract

The work performed time series modeling using the method of constructing autoregressive models. At the first stage, the given series was checked for stationarity and the autocorrelation and partial autocorrelation function were constructed. Based on the analysis of the autocorrelation and partial autocorrelation function, a number of autoregressive time series models of different orders were constructed, and the best (adequate) model was determined among the set of constructed ones. The determination of the best model was carried out based on the analysis of their statistical characteristics using the consolidated criterion of adequacy of models. 
The modeling was carried out in the econometric package EViews. 
The optimal model was presented, a forecast was constructed in 3 steps, and the forecast accuracy was assessed using the MAPE metric. The results obtained are illustrated in figures and graphs. Software was developed to determine the best model among the constructed models for predicting the dynamics of the time series. The software development process was based on the Python programming language, the choice of which was justified by its concise syntax, wide application capabilities and the presence of a developed ecosystem of libraries, in particular pandas (for data processing), numpy (for numerical calculations), statsmodels (for statistical modeling) and scikit-learn (for machine learning tasks). The development and debugging of the program code was carried out in the integrated environment PyCharm. 
The result obtained as a result of program execution was verified by comparison with the control example calculated in EViews, and its identity was confirmed.