A machine learning technique embedded reference-dependent choice model for explanatory power improvement: Shifting of reference point as a key factor in vehicle purchase decision-making

Machine learning is a powerful tool with the potential to improve a choice model's ability to explain consumer behavior. Although the reference-dependent choice model, developed with an emphasis on real decision-making processes, has an advantage over general discrete choice models in terms of explanatory power and interpretability, there is still a lack of consensus on how the reference point should be set. Currently, the common practice is to design a reference point-based utility equation to make an arbitrary decision between past experience, the status quo, and future expectations as the reference point. However, as individual consumers may differ from researchers in how they set their reference points, the current method is rather limited for understanding consumer choice behavior. Therefore, this study proposes a new approach to more accurately understand consumer choice behavior by shifting individual reference points using ANNs (Artificial Neural Networks). The analysis results show that the model proposed in this study has better explanatory power than both the discrete choice model and the existing reference-dependent choice model. This implies that the reference point typically set by researchers approximates each individual's actual reference point through artificial neural networks. This study is significant in that it confirms the possibility that the current status, which showed the highest model fit among several reference price proxy candidates in vehicle choice, may not function as the actual reference price, while also proposing a foundational framework for identifying each consumer's true reference price. Furthermore, it can contribute to corporate strategies and government policy recommendations based on consumer preference analysis, where high explanatory power is required.