Application of Multivariate Fuzzy Time Series Models to Consumer Purchasing Decisions

Economists have long acknowledged that the population’s imprecise and subjective perceptions of the economy follow true economic indicators. Researchers have primarily studied this phenomenon through the application of consumer confidence surveys to traditional econometric models. Fuzzy time series models are an alternative modeling paradigm that have been shown to accurately forecast financial and economic movements by leveraging qualitative and pattern-based reasoning inherent to human decision making. Despite this, nobody has assessed if simulating consumers’ qualitative economic perceptions with fuzzy time series is a viable approach to forecasting their purchasing decisions. This paper addresses the gap in the literature by applying multivariate economic fuzzy time series to forecast vehicle purchases in the United States. We evaluated the utility of our approach by comparing the fuzzy time series models to Long-Short-Term-Memory (LSTM) and Vector Autoregressive (VAR) time series models. Results show that the fuzzy time series models perform comparatively or significantly better than LSTM and VAR models in out-of-sample forecasts of vehicle sales. These results suggest that the fuzzy time series approach could have significant future utility for forecasting and interpreting aggregate consumer purchasing trends by imitating their rationality.