RDP 2018-11: Consumer Credit Card Choice: Costs, Benefits and Behavioural Biases 6. Evidence of Behavioural Biases?

I have found that a relatively high share of credit card holders made a net monetary loss on their card. But the fact that consumers make a monetary loss is not in itself evidence that they face behavioural biases or bounded rationality. Despite incurring a monetary cost, cardholders may still receive other net benefits – in terms of their overall consumer surplus or utility – from holding and using their card. For instance, they may value the ability to smooth their spending by borrowing, or they may value – and be willing to pay for – any non-monetary benefits that their card offers.^[29]

The analysis in Section 5.3 suggests, however, that most cardholders would have been better off if they had chosen a different card, while still retaining the ability to borrow money and access to most non-monetary benefits. And most cardholders who made a loss could have avoided or significantly reduced such losses by choosing a card that better suits their card use patterns. This suggests that cardholders face some barriers to selecting an optimal card. To test whether these barriers reflect behavioural biases, I consider three hypotheses corresponding to the principles of behavioural economics outlined in Section 2.2. I assess whether consumers' choices in these three domains appear to depart from a standard economic model of rational behaviour.

For two of the three hypotheses, the data allow me to set up and test each hypothesis within a regression framework, to estimate the potential effect of the bias on consumers' net monetary benefit. For the other, the data do not allow for a formal test of the effect of the bias on the value of the net monetary benefit. Instead, I identify respondents within the dataset who appear to behave consistently with the proposed bias, and use regression analysis to consider demographic and other characteristics that are associated with this behaviour.

6.1 Hypothesis 1: Optimism Bias

The interest rate that a credit card carries is not relevant to cardholders unless they incur interest charges. Consumers who do not intend to use their credit card to borrow may therefore rationally disregard the interest rate when choosing a card. In contrast, those who do intend to borrow on their card may be able to substantially reduce their interest costs by choosing a card with a lower interest rate.

But the survey data show that most people who pay credit card interest did not consider the interest rate when they first chose their card. This is true not just for those who say they paid interest as a one-off or missed their repayment deadline, but also for respondents who regularly pay interest. Just 40 per cent of respondents who reported that they usually or often pay interest, or were paying off a long-term debt, reported that they had considered the interest rate when they chose their card (Figure 8).

As a result, while some consumers who had paid interest in the past year held a card with a relatively low interest rate, most did not (Figure 9). For instance, among those who stated that they usually or often pay interest, the median interest rate was 19 per cent, just slightly lower than the median rate for respondents who did not pay interest at all (20 per cent).

These patterns suggest that a large share of respondents who pay interest may have underestimated their probability of doing so when they first chose their credit card. On average, respondents appear more likely to underestimate this probability than to overestimate it. Twenty-two per cent of respondents had paid interest in the past year but had not considered the interest rate (indicating they may have underestimated their probability of paying interest). In comparison, just 9 per cent of respondents had considered the interest rate but had not paid interest (indicating they may have overestimated their probability of paying interest; Table 2).

These descriptive statistics suggest that at the time when consumers first choose their credit card, a material share may be overly optimistic about whether, and how much interest, they will pay; that is, they may be subject to optimism bias. I test this hypothesis within a regression framework.

My null hypothesis is that, on average, cardholders are forward looking in predicting whether and how much they will borrow. If this hypothesis is true, consumers who considered the interest rate when they chose their card should be those who intended to – and on average, did – incur higher interest costs. My alternative hypothesis is that many consumers who ended up paying interest were overly optimistic about whether and how much they would borrow when they first chose their card. If this is the case, cardholders who did not consider the interest rate would, in fact, incur higher interest charges for the same level of borrowing.

To test these hypotheses, I run the following regression:^[30]

I use two outcome variables: the dollar value estimate of interest paid over the past year and the value of the net monetary benefit.

considered interest rate is a binary variable equal to 1 if the respondent reported that the interest rate influenced their card choice, and zero otherwise. reason paid interest is a categorical variable indicating the reason for the respondent's most recent interest payment (options were: usually/often pays interest or paying off a long-term debt; one-off intentional interest payment; missed repayment deadline; or, accidentally overspent relative to budget).^[31] The omitted category is cardholders who had not paid interest in the past year. X is a set of controls including household income, education, employment status, age and number of credit cards that the consumer holds. Extreme values of the dependent variable are censored using a 5 per cent Winsorisation – this reduces the magnitude of coefficients, but does not affect their level of significance.

I focus on whether respondents had considered the interest rate, rather than whether they held a lower-rate card, to abstract from potential supply-side factors. One possible explanation for the incidence of higher-rate credit cards among those who pay interest is that they may have a less favourable credit history, which would make it more difficult for them to obtain a card with a lower interest rate. If this were the main driver of the prevalence of high interest rates among interest payers, we would expect to see these respondents considering the interest rate as part of their decision, but still holding a higher-rate card.

The interaction term between having considered the interest rate and the reasons for paying interest controls for the composition of interest payers. For instance, as may be expected, many respondents who had paid interest but had not considered the interest rate had paid interest as a one-off. They had missed their repayment deadline or accidentally overspent in a given month. The interaction term in the model allows me to abstract from these one-off cases. I focus on the impact of having considered the interest rate for respondents for whom the interest rate is most important: those who state that they usually or often pay interest.

I calculate the marginal effect of having considered the interest rate for respondents who usually or often pay interest as a one-sided test. If the null hypothesis is true, then respondents who considered the interest rate and usually pay interest would pay higher (or equivalent) interest charges $$\left({\beta }_1+{\beta }_3\ge 0\right)$$ and would receive a lower total net monetary benefit. Conversely, under the alternative hypothesis, those who had considered the interest rate would pay lower interest charges $$\left({\beta }_1+{\beta }_3<0\right)$$ or receive a larger net monetary benefit.

I find that, of those who regularly pay interest, the respondents who had considered the interest rate actually paid lower interest charges and received a larger net monetary benefit than interest payers who did not consider it (Table 3). This finding is consistent with the hypothesis that borrowers who did not consider the interest rate tend to pay higher interest charges for a given level of debt, because they are more likely to hold a higher-rate card. It may also reflect differences in borrowing behaviour across these different types of consumers. We can abstract from the effect of at least some of these differences by adding demographic controls to the model; doing so has only a small influence on the coefficients of interest.

Of the respondents who usually or often pay interest, those who had considered the interest rate incurred, on average, $80 less in interest charges than those who had not (Table 4). This difference persists after adding controls, and was significant at the 5 per cent level in the one-sided hypothesis test. Similarly, these respondents also received an average net monetary benefit about $105 larger, significant at the 5 per cent level. After adding controls, this fell to around $75 and was significant at the 10 per cent level.

These findings are consistent with the optimism bias hypothesis. They indicate that many cardholders who pay interest appear to have systematically underestimated their probability of doing so when they first chose their card.^[32] But based on the available data, I am unable to entirely rule out other potential explanations for the observed patterns. For instance, it could be that respondents who incur interest charges do so because they have experienced a shock that was unpredictable at the time that they chose their card. If this type of negative shock is combined with barriers to switching to a more appropriate card (see Section 7), we would observe these same patterns. A definitive test of whether my findings represent optimism bias would require more detailed data, with information on whether respondents had experienced shocks or changes in their financial position since choosing their card.

6.2 Hypothesis 2: Bounded Rationality

Even if consumers are able to accurately predict their own card use behaviour, they may still choose a sub-optimal card due to a constrained ability to discover and compute the value of their options (Simon 1978). That is, consumers may be ‘boundedly rational’ in choosing their card. Factors such as limited time, lack of awareness of alternatives, or information overload may lead cardholders to form an incorrect impression of a card's value. These factors are particularly likely to affect decisions for more complex products.

I consider one potential – and important – consequence of bounded rationality; whether consumers have inaccurate estimates of the net monetary value of their own card.^[33] To facilitate this, credit card holders were asked in the RBA Consumer Payments Survey whether they felt that the monetary benefits of their main credit card were greater than, equal to, or less than the monetary costs. Respondents were asked to consider fees, rewards points, interest paid and the interest-free period, so that their responses are comparable with the calculated net monetary benefit used in this analysis.

The distribution of respondents' perceptions of their net benefit lines up reasonably well with my estimates, though the share who believed that they made a net loss was substantially lower than my estimate. In total, around 40 per cent of cardholders believed they were financially better off from their credit card, 40 per cent reported they broke even and 15 per cent reported that the costs of their card outweighed the benefits (Table 5). A further 4 per cent were unsure.

If bounded rationality affects cardholders' valuation of their card, we may expect them to overestimate its value; bounded rationality may lead consumers to place a disproportionate weight on advertised benefits when choosing a card, potentially placing a lower weight on costs which are not as prominently advertised.^[34] I find that while most cardholders accurately estimated their position, around 6 per cent of all cardholders – or one in six of the respondents who believed they were making a net monetary benefit – were actually making a loss by my estimates.

Limiting my sample to respondents who, by my estimates, made a net loss on their credit card, I run a regression to understand which types of respondents were more likely to incorrectly believe they were making a net monetary benefit.^[35] Within the sub-sample of loss-making respondents, I construct a dependent variable equal to 1 if the respondent perceived that they were making a gain, and zero for respondents who perceived that they were making a loss or breaking even. I regress this dependent variable on motivation for holding a credit card (as defined in Section 5.2), and on demographic and other explanatory variables.

While we saw in Section 5.2 that respondents who were motivated by rewards points were more likely overall to make large net monetary benefits from their card, if they made a loss, they were substantially less likely to be aware of it (Table 6, column (1)). After adding controls to hold demographic characteristics, liquid wealth and number of cards constant, this effect falls by nearly a third, but remains large; loss-making respondents motivated by rewards points were almost 27 percentage points more likely to believe they were making a gain.

One potential explanation for this finding is that it reflects differences in the salience of benefits, relative to costs, for cardholders motivated by rewards points. While these cardholders may receive benefits from rewards points each month, they pay annual fees – often as an automatic deduction from their account – just once a year. Therefore, these respondents may underweight irregular costs (or equivalently, overweight regular gains) when estimating the value of their card.

This finding is consistent with an interpretation that consumers' inaccurate estimates of the value of their card reflect bounded rationality. For those who borrow and pay interest, interest costs are frequent and highly visible. But for those who do not pay interest, it can be more difficult to accurately weigh up costs and benefits. To add to this challenge, rewards card holders may have difficulty estimating both the number of rewards points they will accrue, and the value of those points relative to their annual fee. As Friesen and Earl (2015) find with mobile phone plans, wide variation in outcomes across respondents may be because some consumers navigate this complexity by developing expert knowledge of the market, while others do not.

Curiously, my results show that loss-making respondents were more likely to believe they were making a gain if they were more highly educated (Table 6, column (2)). The coefficients on education remain significant but decline substantially in magnitude in a Heckman sample selection probit model (column (3); see full results in Appendix D), suggesting that my model overstates the effect of education on the probability of overestimating a card's value.^[36]

There are plausible reasons that more highly-educated consumers might overestimate the value of their card. For instance, as with respondents motivated by rewards, they may be less likely to pay interest charges, making costs less salient. Or they may devote less effort to developing expertise in the credit card market as they are able to absorb modest losses. But it should be noted that the demographic and preference variables used in the Heckman selection model can only control for observable factors, which together explain only a small part of selection into the group of loss-making cardholders (with an adjusted R-squared at this stage of just 0.05). Unobservable factors appear to be the main drivers of selection into the group of loss-making cardholders, creating the potential for the estimated effect of education to reflect omitted variable bias.

6.3 Hypothesis 3: Present Bias

Sign-up offers – such as bonus rewards points, a discounted annual fee, a temporary low interest rate or a balance transfer deal – are common in the Australian credit card market. Shui and Ausubel (2005) find that these types of short-term offers can appeal to consumers' present bias, attracting them to products that carry substantial long-term costs with continued use.

In the Consumer Payments Survey, around 30 per cent of credit card holders reported that they had originally signed up to their main card with a special offer (Figure 10). At least 10 per cent of cardholders had signed up to their card with an offer that no longer applied at the time of the survey.^[37]

Outcomes for respondents with an expired offer can help to reveal whether consumers are present-biased when choosing a card. If so, we may expect those with an expired sign-up offer to receive a lower net monetary benefit from their card. This is because, at the time of choosing their card, present-biased consumers may overweight the short-term benefits they would receive from a sign-up offer, and may be inattentive to longer-term costs that they will incur after the offer expires.^[38] Alternatively, if consumers respond to sign-up offers in line with a standard rational choice model, respondents with expired sign-up offers would be no worse off, after the offer expires, than similar respondents who did not sign up to their card with a special offer.^[39]

Before controlling for other factors, it appears that average net monetary benefits for cardholders who had responded to a sign-up offer were substantially higher than for those who had not (Figure 11). The median net monetary benefit for respondents who had signed up with a special offer – whether or not it still applied at the time of the survey – was around $50. That is, respondents with an expired sign-up offer received, on average, a larger net benefit than respondents whose card choice was not affected by a sign-up offer.

These descriptive statistics suggest that cardholders do not incur absolute losses due to present bias. But I test for the effect of present bias more formally by estimating the effect of an expired sign-up offer on net monetary benefit, using the following regression:

where X is a set of control variables including household income, age, education, employment status, number of cards held and typical value of bank deposits.

I separate respondents into four categories depending on their self-reported card sign-up behaviour, corresponding to those on the horizontal axis of Figure 11.^[40] The omitted category in the regression is respondents who had not signed up to their card with a special offer. I focus on respondents who received an offer that no longer applied at the time of the survey. I run this regression with the net monetary benefit as the dependent variable (Table 7), as well as with three of its components: the value of rewards points, the annual fee, and the amount of interest paid.

I test the present bias hypothesis by considering the coefficient $${\alpha }_2$$. Under the null hypothesis (a rational choice model), the coefficient would be zero or positive. Under the alternative hypothesis (‘present bias’), the coefficient would be negative, which would indicate that respondents who signed up to an offer that no longer applies are worse off.

The results indicate that, consistent with the unconditional relationship shown in Figure 11, respondents who signed up to their card with a special offer that no longer applied were no worse off (Table 7).^[41] In fact, the results suggest that if anything these respondents were better off than those who did not respond to a sign-up offer – though this difference is not statistically significant for the net benefit or any of its components.

Based on these findings, I fail to reject the null hypothesis. On average, respondents who signed up to their card with a special offer do not appear to be present-biased in choosing their card. These findings contrast with those of Shui and Ausubel (2005) who found that US consumers were present-biased in responding to temporary low interest rate offers. At face value, my results also appear inconsistent with ASIC's (2018) finding that a substantial share of consumers are worse off after responding to balance transfer deals.

There are a number of potential reasons for this contrast. First, my findings may reflect differences between the credit card markets of Australia and the United States. Most notably, there are important regulatory differences; the type of card solicitation tested by Shui and Ausubel (2005) would contravene Australian regulatory standards.^[42]

Relatedly, my findings are likely to reflect differences in the nature of the offers being examined, and the types of consumers they attract. Shui and Ausubel (2005) consider offers of discounted interest rates only, while ASIC (2018) focuses on balance transfer deals. But questions asked in the Consumer Payments Survey cover all types of offers, so I do not have information on the nature of the offer that applies to each respondent. Overall, it appears that the consumers who respond to sign-up offers generally have higher incomes and higher levels of liquid wealth, suggesting that, on average, these offers are targeted at, or are more attractive to, higher-spending customers. It may be that card-issuing banks target offers to these groups, as they may have multiple financial service needs, making them attractive customers (Figure 12).^[43] It therefore seems likely that for most cardholders who responded to a sign-up offer, these were offers of bonus rewards points or discounted annual fees, rather than discounted interest rates or balance transfers. It may also be that lower-income consumers or those with a poor credit history are not eligible for the types of cards that these introductory offers usually apply to, or that respondents who would make a loss after the introductory offer expires tend to switch cards in response. It is possible that, in aggregate, the positive or neutral outcomes for higher-income cardholders may outweigh potential negative outcomes for cardholders who respond to balance transfer offers or temporarily low interest rates; more detailed data would be needed to test this hypothesis.

Footnotes

Though, as is evident from the analysis in Section 4.1 and Appendix A, most survey respondents appear to view monetary features as more important than non-monetary features. [29]

Note that the sample for this regression includes only respondents who hold a credit card. [30]

Because this variable is reported based on activity in the past year, it is not perfectly correlated with the value of interest paid (which is based on the month of the most recent credit card statement). [31]

An alternative explanation, with the same effect, is that respondents are ‘myopic’; that is, it is not that they underestimated their probability of paying interest, but that they failed to consider the probability at all. [32]

An ideal test for whether bounded rationality is important in this context would be to consider whether respondents with more complex cards (e.g. those that have more features, incidental fees, or complex rewards programs) receive lower net monetary benefits than those with similar, but less complex card products. Unfortunately, I do not have a measure of product complexity that is independent of other card features that affect the net monetary benefit. [33]

Note that the perceived monetary benefit is measured as at the time of the survey, not at the time of choosing the card; if, for instance, respondents revise their estimate of the value of their rewards program after using their card for some time, the survey results would incorporate this learning. [34]

Although respondents who broke even or made a net benefit may also overestimate the value of their cards, I am unable to observe this based on the categorical measure of perceived gain or loss. [35]

The Heckman sample selection model re-estimates the model in column (2), but accounts for selection into the loss-making cardholder group based on observable factors. [36]

The actual share of respondents with expired sign-up offers is likely to be larger than 10 per cent, and include some of the 17 per cent of respondents who could not remember whether a sign-up offer had applied to their card. [37]

These long-term costs may include the time cost of switching cards, if consumers intend to choose a different card after their offer expires, but fail to follow through. Relatedly, but outside of the scope of this paper, the sign-up offer may lead beneficiaries to form habits that persist after the offer expires. For instance, those with a temporary low interest rate may develop spending patterns based on that temporary low cost of debt. [38]

More precisely, a forward-looking consumer may compare the time-discounted costs and benefits of a card with a sign-up offer to a card without an offer, and choose the card with the highest net present value. If costs increased substantially after the offer expired, we would expect forward-looking consumers to switch cards when the offer expires such that we would not observe them with expired offers. [39]

The value of the offer itself would not be captured in the net benefit calculation unless it was a discounted annual fee. [40]

If instead of respondents' self-reported sign-up behaviour, I use an indicator of whether respondents listed ‘bonus points, flights or a cash back offer’ as an important reason for choosing their card (from Figure 2), I also find no significant effect. [41]

Australian regulations prohibit card issuers from sending consumers unsolicited pre-approved offers of credit or credit limit increases. [42]

Income and liquid wealth are controlled for in the regressions in Table 7. Findings in Section 6.2 suggest that some highly-educated consumers believe they are making a net gain from their card when in fact they are making a loss. But the value of these losses is modest as these consumers are less likely to incur interest charges. As discussed in Section 5.1, most highly-educated respondents received a net benefit from their card. These findings are therefore consistent with banks potentially using credit card sign-up offers as a ‘loss leader’ to attract customers to other financial products. [43]