Cash Usage: Consumer Preferences

The core dataset for this research originates from the 2009 Bank of Canada Method of Payment Survey. This survey is composed of two parts: a questionnaire gathering demographic information (age, income, gender, education, marital and employment status), banking details (linking payment choices to factors like free debit transactions and credit card rewards), and attitudinal data on perceived convenience and safety of different payment methods; and a three-day shopping diary. The diary tracks individual transactions, recording the payment method used and crucially, the 'perceived acceptance' of various payment options at each point of sale. The sample, initially not fully representative, underwent adjustments to correct for sampling bias. Data points with missing information on perceived acceptance, demographics, or transaction values were excluded, leaving 1,452 individuals with 7,908 transactions. The researchers emphasize the two key strengths of this data: multiple observations per individual, allowing for the modeling of individual heterogeneity in payment choice; and the inclusion of 'perceived acceptance,' which helps to disentangle demand-side factors (consumer preferences) from supply-side factors (merchant policies).

A significant advantage of the Bank of Canada dataset is the presence of multiple shopping observations per individual. This enables a more robust analysis of individual-specific effects on payment choices, going beyond the limitations of single-transaction studies. The researchers can identify individual patterns and preferences as these multiple transactions occur in slightly varying shopping contexts (transaction values and shopping types). The data's other notable feature is the concept of 'perceived acceptance.' Participants indicate which payment methods were accepted at each transaction. This is vital in separating consumer preferences from the influence of merchant acceptance policies. By knowing what payment options were available to each consumer, the researchers can model consumer choices more accurately, isolating the true preference from the constraints imposed by merchant acceptance. In essence, the dual aspects of multiple observations and perceived acceptance allow for a nuanced understanding of both demand-side (consumer behavior) and supply-side (merchant policies) drivers of payment method selection.

The initial dataset was refined through several steps to create the final analytical sample. Non-representative samples were weighted to adjust for sampling bias. Incomplete data points, particularly those missing information on perceived acceptance, demographics, and transaction value, were excluded. Further, to effectively model individual heterogeneity, the researchers needed at least three shopping observations per individual, leading to the removal of individuals with fewer transactions. This resulted in a final sample of 1,452 individuals and 7,908 transactions. The consumers within this sample are categorized into four types based on their consistent payment patterns: (1) Cash users, who exclusively use cash; (2) Debit users, primarily using debit cards with cash as a fallback; (3) Credit users, similarly favoring credit cards; and (4) Mixed users, employing all three payment methods seemingly at random. This categorization is crucial to the study's understanding of heterogeneity in consumer payment choices.

To analyze the payment choice data, the researchers employ a generalized multinomial logit (G-MNL) model. This model is chosen because consumer payment choice is inherently discrete, and the G-MNL model is specifically designed to accommodate this discrete nature and the observed heterogeneity. The model is structured to capture individual unobserved heterogeneity (demand-side factors) while simultaneously accounting for the supply-side influence of merchant card acceptance. This method allows for the identification of individual-specific patterns and preferences, distinguishing them from the constraints imposed by merchants' payment policies. Alternative multinomial logit models are also estimated as benchmark comparisons, highlighting the superior fit and statistical significance achieved by the G-MNL model through the incorporation of scale coefficients and random coefficients which better reflect the complex interplay of consumer choice and merchant acceptance in the Canadian payment landscape.

The study's central theme is the significant heterogeneity in consumer payment preferences. Analysis of the 2009 Bank of Canada Method of Payment Survey data reveals that consumers don't exhibit uniform behavior when selecting payment methods. Instead, distinct patterns emerge, leading to a categorization of consumers into four groups: cash users (exclusively using cash), debit users (primarily using debit cards, resorting to cash when debit cards are unavailable), credit users (showing a similar preference for credit cards), and mixed users (randomly utilizing all three payment methods). The existence of these distinct groups underscores the importance of considering individual-specific preferences when building accurate models of payment choice. Ignoring this heterogeneity leads to an incomplete and potentially misleading understanding of consumer behavior and the factors influencing payment decisions. The presence of sizeable proportions of debit and credit users indicates that, for many, switching between card types is not easily influenced by minor changes in transaction values or context. The variation in preferences is crucial to understanding how payment choices are affected by factors beyond the simple transaction amount.

Beyond the basic categorization of users, the study identifies another layer of heterogeneity: heterogeneous thresholds. Even within each user group (cash, debit, credit, mixed), individuals may display different payment choices under identical circumstances. Some people within the same category might select different payment methods when faced with similar transaction values and types. To capture these nuanced preferences, the researchers leverage a generalized multinomial logit (G-MNL) model. This model incorporates scale coefficients (scaling the utility of a particular payment method) and random coefficients (modifying substitution patterns between alternatives). These additions significantly enhance the model's ability to reflect the observed variations in consumer behavior, capturing not only the broad categories of payment users but also the subtle differences in thresholds and substitution patterns that exist even within these categories. This detailed model specification is vital for accurately predicting the influence of external factors such as policy changes.

The incorporation of heterogeneity into the G-MNL model proves statistically and economically significant. Metrics like the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), which measure the goodness of fit, dramatically improve when the model accounts for individual differences in preferences. This improvement confirms that the observed heterogeneity isn't merely random noise; it represents a crucial element of consumer payment behavior. Models that assume homogeneous preferences (i.e., ignoring individual-level variation) perform significantly worse in explaining the observed data. This highlights the limitations of simpler models and emphasizes the need for more complex models that account for the diverse preferences of individuals. The statistical validation of the heterogeneity supports the research's focus on individual variations and suggests that homogeneous models are inadequate for accurate prediction or policy analysis.

The core of this section is a policy simulation exploring the impact of universal card acceptance on consumer payment choices. This simulation examines a hypothetical scenario where all merchants accept all payment methods (credit, debit, and cash) regardless of the transaction value. This scenario is constructed to understand the extent to which the observed dominance of cash usage, especially for small transactions, is driven by consumer preferences versus merchant restrictions. The simulation essentially eliminates the supply-side constraint of merchant card acceptance, allowing for a clearer picture of consumer preferences. The results of this simulation are directly compared to the results from models that do not explicitly account for individual heterogeneity in payment choices.

To conduct the simulation, the researchers utilize the estimated parameters from their generalized multinomial logit (G-MNL) model. Given the estimator's asymptotic normality, 2,000 sets of normal random draws are generated to account for uncertainty in the parameter estimates. These draws are used to simulate choice probabilities under the counterfactual scenario of universal card acceptance. In this simulated scenario, consumers' choice sets are no longer limited by merchant acceptance; it is assumed that all consumers have access to all three payment methods (cash, debit, and credit). For consumers lacking credit or debit cards in the original data, hypothetical cards with minimal benefits (zero rewards, low credit limits) are assigned to ensure everyone has the full range of payment options. This carefully controlled simulation setup is necessary for isolating the effect of universal card acceptance on consumer payment behaviors, allowing for a more accurate assessment of the influence of consumer preferences without the confounding effect of merchant card acceptance policies.

The simulation results show that universal card acceptance would decrease overall cash usage, but the magnitude of this decrease is relatively modest. Depending on the specific model used, overall cash usage is predicted to decrease between 6.9% and 8% in terms of transaction frequency and by a similar margin in terms of transaction value. This observation suggests that consumer preferences play a significant role in the persistence of cash usage, particularly for smaller transactions. Even without constraints from merchant acceptance, many consumers would still choose cash. The researchers emphasize the contrast between these results and those from models that do not account for individual heterogeneity. Models that ignore individual-level differences predict even smaller decreases in cash usage, underlining the importance of incorporating this heterogeneity for more accurate predictions. The seemingly small overall effect on cash usage is further explained by the finding that many consumers prefer cash even in situations where cards are readily accepted.

A more granular analysis reveals varying degrees of impact based on merchant acceptance and transaction values. The reduction in cash usage is significantly larger (approximately 34 percentage points) among merchants who currently accept only cash, suggesting that many would switch to cards if given the opportunity. However, for small-value transactions (less than $10), the decrease in cash usage remains relatively small, despite the high initial market share of cash in this segment. For transactions between $10 and $25, the decrease is about 10 percentage points. The findings also highlight that in scenarios where all payment methods are already accepted by merchants, changes in usage are minimal, providing validation for the models. This nuanced examination provides a more comprehensive understanding of the complex dynamics of consumer payment choice and the limitations of assuming homogeneous preferences in policy analysis. The results clearly demonstrate the importance of considering individual-level heterogeneity in assessing the consequences of potential policy changes.

The study's primary finding is the significant influence of consumer preferences on cash usage. Despite the availability of debit and credit cards, cash remains a prevalent payment method, particularly for small-value transactions. The policy simulation, which hypothetically eliminates merchant restrictions on card acceptance, shows only a modest decrease in cash usage (approximately 7.7 percentage points in transaction frequency and 7.5 percentage points in transaction value). This indicates that consumer preference, rather than merchant limitations, is the primary driver of cash usage. This conclusion is further supported by the comparison with models that omit individual-level heterogeneity; these models predict even smaller reductions in cash usage under the universal card acceptance scenario. The finding that consumer choice strongly influences cash usage has important implications for policymakers considering interventions to promote electronic payment adoption.

The research highlights the crucial role of individual heterogeneity in accurately modeling consumer payment behavior. Models that incorporate individual-specific preferences and varying thresholds for payment method selection (i.e., the models which account for unobserved individual heterogeneity) offer considerably better predictions compared to simpler models assuming homogeneous preferences. The inclusion of individual heterogeneity leads to substantially improved model fit, as measured by AIC and BIC. The counterfactual simulation results clearly demonstrate this: Models that account for heterogeneity predict a larger decrease in cash usage under universal card acceptance compared to models without heterogeneity. This difference stems from the fact that the more sophisticated models better capture the behavior of committed debit and credit users, who would utilize their preferred cards more frequently under a scenario of universal merchant acceptance. The findings underline that the complexity of individual preferences must be integrated into policy-relevant models.

The simulation of universal card acceptance reveals a surprisingly limited overall impact on cash usage. While a shift away from cash is observed, the magnitude is smaller than initially anticipated, indicating a strong consumer preference for cash, particularly in low-value transactions. This implies that policies aiming to completely replace cash with electronic payment methods might be less effective than previously thought. The study also acknowledges limitations, including the potential underestimation of welfare gains due to the current practice of consumers selecting shops that accept their preferred payment method. Although the short-term effect of universal card acceptance on cash usage is relatively small, the long-term effect might be significantly larger if consumers decide to switch their credit and debit cards for better reward programs. This suggests a more nuanced approach to policy interventions is necessary, going beyond simply increasing merchant card acceptance.