ABSTRACT
Business tendency surveys are used to determine the current state or predict economic changes, with various approaches based on econometric models and artificial intelligence algorithms used to analyze respondents’ answers. One such approach is the long-short-term memory (LSTM) model, which allows for the simultaneous inclusion of answers to many questions from the questionary as input data to the algorithm. This concept was used to predict the course of the price indices of sold production of industry in Poland based on a business tendency survey in the Polish manufacturing industry conducted by the Research Institute for Economic Development. Compared to econometric modeling, a disadvantage of this approach is the lack of explainability of the impact of individual questions on the final result. This chapter aims to indicate the SHapley Additive exPlanations (SHAP) approach as a possible extension of the LSTM model to explain the contribution of individual survey questions to the overall result, proving the decreasing advantage of classic econometric methods over machine learning methods in terms of the interpretability of results. It also confirms the universality of the SHAP technique, which has not been generally combined with the LSTM model so far.
