Bounded rationality in decision making: a machine learning approach

Abstract

A key task for an effective marketing strategy is to understand how the consumers make choices. The way in which the consumers adopt, maintain or change their preferences is fundamental for the designing of new products, direct marketing campaigns, pricing or demand estimation of new products. This is not an easy task, because people are always influenced by numerous internal factors, such as emotions, or external factors, such as life events, which could affect their choices. In this research, we focus on understanding the consequences of consumer behavior in two different circumstances, both under the bounded rationality framework. First, in a simulated discrete choice experiment context, in which consumers pay selective attention to the attributes of each profile. Second, in an empirical context, where consumers face a life event that makes them adjust their preferences. In our first work, we propose the use of a machine learning approach based on Support Vector Machines (SVM), to identify the non-attendance of attributes at individual level and to predict the consumer choices in a conjoint experiment. We conduct an extensive simulation study to investigate the performance of the proposed approach. We compare the performance of our proposed approach to different benchmarks from the literature. Our results with simulated data show a better performance in terms of the identification of the non-attended attributes, that improve the predictive ability of the choices of consumers. Finally, we test our approach in two empirical applications previously reported in the literature. We demonstrate the superiority of our approach and the alternative insights derived from our method. In our second work, we study how the consumption behavior of first-time parents is affected, both during the pregnancy period and after birth. We combine a unique dataset that identifies precisely the date of childbirth with a supermarket credit card data. We observe detailed supermarket transactions and aggregated purchases made at different external companies using the credit card, to investigate the relationship between pregnancy, childbirth and consumption. To examine the causal effect of pregnancy and childbirth on consumption, we use a causal random forest methodology. Our results show statistically significant impacts in 44\% of the analyzed product categories during the pregnancy period, and in 48\% of the product categories studied during the post-birth period.