Inferring psychological traits from spending categories and dynamic consumption patterns

Abstract

In recent years there has been a growing interest in analyzing human behavioral data generated by new technologies. One type of digital footprint that is universal across the world, but that has received relatively little attention to date, is spending behavior.In this paper, using the transaction records of 1306 bank customers, we investigated the extent to which individual-level psychological characteristics can be inferred from bank transaction data. Specifically, we developed a more comprehensive feature space using: (1) overall spending behavior (i.e. total number and total amount of transaction), (2) temporal spending behavior (i.e. variability, persistence, and burstiness), (3) category-related spending behavior (i.e. diversity, persistence, and turnover), (4) customer category profile, and (5) socio-demographic information. Using these features, we first explore their association with individual psychological characteristics, we then analyze the performances of the different feature families and finally, we try to understand to what extent psychological characteristics from spending records can be inferred.Our results show that inferring the psychological traits of an individual is a challenging task, even when using a comprehensive set of features that take temporal aspects of spending into account. We found that Materialism and Self-Control could be inferred with relatively high levels of accuracy, while the accuracy obtained for the Big Five traits was lower, with only Extraversion and Neuroticism reaching reasonable classification performances.Hence, for traits like Materialism, Self-control, Extraversion, and Neuroticism our findings could be used to improve psychologically-informed advertising strategies for specific products as well as personality-based spending management apps and credit scoring approaches.

Explainable AI for Psychological Profiling from Behavioral Data: An Application to Big Five Personality Predictions from Financial Transaction Records

“…(The users of the money management application have relatively large financial constraints. For this reason, we set more flexible criteria compared to related studies (see for example Tovanich et al [13])). This exclusion procedure leaves us with N = 6408 data subjects of whom we have transactions data (linked to their self-reported personality profiles) that can be fed into the prediction models.…”

“…(3) Conscientiousness, the tendency to show self-discipline, aim for achievement, and be organized; (4) Neuroticism, the tendency to experience unpleasant emotions easily; and (5) Openness to Experience (or simply Openness), the tendency to be intellectually curious, creative and open to feelings [13,44]. Personality theory specifies that traits are hierarchically organized [44,45]: each domain subsumes more specific facets that have a unique variance not entirely explained by the higher order Big Five.…”

“…The spending data reflect a wide range of behavioral patterns, which we organize into two broad categories of features based on related work [12,13]: (1) overall spending that comprises summary statistics of spending aggregated over time and features that enrich the aggregated measures with finer-grained, time-dependent information (e.g., how much does the daily amount someone spends vary over time?) and (2) category spending that reflects a person's spending category profile and relative spending per category.…”

“…As Matz et al [6] define it, psychological profiling is "the automated assessment of psychological traits from digital footprints". Over the past decade, researchers have been tapping into a broad variety of data sources for psychological profiling, including social media data (e.g., Facebook likes and status updates [2,7]), mobile sensing data [8], music listening preferences [9,10], mobility behaviors [11] as well as financial transaction records [12,13].…”

“…Although prior work on psychological profiling has predominantly focused on predictive performance [22,35], there have been attempts at explaining the underlying mechanisms. Some studies have highlighted the face validity of predictive models by showing the most related predictors to the target (based on univariate correlations prior to modeling that do not necessarily reflect what the model learned from the data [2]), or by providing a list of important features [8,13,36]. While such approaches offer initial insights, they do not reflect how the (combination of) feature values impact(s) the predicted classes, nor the extent to which the classifications are explained.…”