Do Liquidity Constraints and Interest Rates Matter for Consumer Behavior? Evidence from Credit Card Data

The paper tests whether liquidity constraints and interest rates “really matter in practice” by using a panel of individual credit card accounts that separately record credit limits and balances, which “allows us to distinguish credit supply and demand, as well as account-specific interest rates” (Sec. I, pp. 1-2). Under the canonical PIH the marginal propensity to consume (MPC) out of a predictable increase in liquidity “should be zero,” and consumption/saving elasticities to interest rates “depend on model parameters” but “most studies have found small effects” (Sec. I, pp. 1-2); the paper’s design – estimating dDebt/dLimit and dDebt/dr separately – lets it test both predictions against the same high-quality account-level data, which the authors argue has “little measurement error” relative to household survey data like the SCF or PSID (Sec. IV, p. 13).

The paper uses two complementary strategies: a rich set of control variables (monthly time dummies, account fixed effects, indicators for self-requested “manual” line changes, and the issuers’ own internal/external credit-risk scores) to absorb the “fundamental” part of credit-supply variation, and an instrumental-variables strategy that exploits “exogenous timing rules built into the credit-supply functions” – for instance, issuers’ reluctance to reconsider an account for a line or rate change within six or twelve months of its last change (Sec. II.2, pp. 7-8). The authors report that these timing-rule instruments are statistically significant (p-value < .0001) in the first stage, with an R² of about 0.03 for credit limits and pass the overidentification test, and show that the estimated MPC “does not change very much” across the full battery of controls and instruments – from about 0.13 in the baseline, to 0.14 with account fixed effects, to 0.08-0.11 under the various IV specifications (Sec. V, Table II, pp. 17-21).

In the full sample, the estimated long-run response of debt to a credit-line increase, bTot, is “relatively large at just under 0.13 and quite significant (t-ratio of 6),” meaning each extra thousand dollars of credit line generates on average a $130 increase in debt over the following year (Sec. V, p. 17). In absolute terms (using an indicator for any line increase rather than its dollar size) this corresponds to “a $350 average increase in debt per line increase” (p. 17); across the paper’s various specifications addressing endogeneity, the MPC estimate “ranges between 10-14 percent” (Sec. V, p. 21). The authors emphasize that debt “rises sharply and significantly over the first two months after a line increase, and then smoothly asymptotes” to this long-run value, a pattern they attribute to the large sample and low measurement error rather than to any unusual smoothing assumption (p. 17).

Accounts with an initial utilization rate above 90 percent have an MPC out of liquidity of “almost 50 percent,” significantly larger than the MPC for accounts with utilization between 50-90 percent (about 0.16) or below 50 percent (about 0.07) (Sec. VII, Table IV, pp. 25-26). The authors read the sharp response of the highest-utilization group as “concrete evidence that liquidity constraints are often binding,” but note that the significant, non-zero MPC even “for people starting well below their limit… is inconsistent with the view that only currently binding liquidity constraints matter” (p. 26) – and argue this pattern is “potentially consistent with the interaction of precautionary motives and the possibility that liquidity constraints bind in the future,” as in buffer-stock models (Deaton 1991; Carroll 1992; Ludvigson 1999) that treat unused credit as a precautionary buffer (p. 26).

Using a homothetic buffer-stock specification in which the optimal buffer of unused credit is a constant fraction of the credit limit (following Ludvigson 1999), the authors re-estimate the model with the utilization rate itself (rather than debt) as the dependent variable, and find the long-run response “quite small, within a percentage point or so of zero,” implying that utilization returns close to its pre-increase level for each group (about 90 percent, 50 percent, etc.) (Sec. VII, pp. 26-27). They describe this as “consistent with people having constant target utilization rates, at least on average,” while allowing that such targeting “could be due to behavioral rules-of-thumb” as well as to conventional precautionary motives, and stress that under either reading “additional liquidity increases debt even at low utilization rates” (p. 27).

In the main specification, a one-percentage-point increase in the account’s interest rate reduces debt by about $110 over nine months (“t-ratio about 13”), corresponding to short-run and long-run elasticities of about -0.8 and -1.3 respectively, which the authors describe as “larger and more significant than in most other micro studies of interest rates” (Sec. VI, p. 22-23), attributing the difference to the large sample and to reduced measurement error in both debt and account-specific interest rates. They find this elasticity “is smaller for people starting near their credit limits” (Sec. VII, p. 27) and larger in magnitude for rate decreases than for rate increases – -$320 versus -$90 in the paper’s decomposition (Sec. VI, Table III row 10, p. 24) – which the authors say “can explain the widespread use of promotional offers like teaser rates” (p. 24).

For credit-limit increases, the estimated effect on the account-holder’s other card balances is “insignificant and positive,” which the authors say shows “no evidence of offset from the other cards,” so the main-sample debt increase “represent[s] a net increase in debt” (Sec. V, p. 22). For interest-rate changes, by contrast, the authors do find some balance-shifting: a one-point rate increase raises other-card balances by about $40 while lowering main-card debt by about $110, “for a net decline of about $70,” so that “the balance-shifting is only partial; there is a net decline in total borrowing,” with a net long-run elasticity “still large at about -0.85” (Sec. VI, p. 24-25).

The paper finds the MPC out of liquidity and high utilization rates are concentrated among low-credit-score, low-income, and especially young accounts – for instance, about 18 percent of young account-holders have utilization above 90 percent versus 14 percent overall, and people with low credit scores have an MPC “substantially larger” than those with high scores, “significantly so at the 7 percent level” (Sec. VII, Table V, pp. 28-29). The authors flag the credit-score result explicitly: “reverse causation is likely also to be at play, because high debt levels can lead to lower scores,” while noting the MPC result (as opposed to the raw correlation between scores and utilization) is “unless the scores are themselves functions of the MPC, which seems unlikely… not due to reverse causation” (p. 28-29).

Using 1995 Survey of Consumer Finances data, the paper finds that “over 90 percent of people with credit card debt have some very liquid assets in checking and savings accounts,” and that even after allowing a generous one month of income for transactions purposes, “1/3 of credit card borrowers still have more than this much income in liquid assets” that could have been used to pay down expensive card debt instead (Sec. VII, p. 32). The authors call this behavior “puzzling, apparently inconsistent with no-arbitrage and thus inconsistent with any conventional model,” and discuss hyperbolic discounting / self-control models (citing Laibson et al. 2000) as a candidate explanation, noting that such models “cannot justify simultaneously borrowing and holding liquid assets, which violates no-arbitrage” on their own, and also discuss strategic bankruptcy-exemption planning (Lehnert and Maki 2000) as a partial, narrower explanation (p. 32-33).

Extrapolating from the automatic-line-increase multiplier, the authors estimate that an average $2,000 credit-line increase per account would raise aggregate debt by “over $40B,” about “10 percent of the size of the annual increase in aggregate consumption,” while a simultaneous rise in interest rates would partly offset such an expansion (Sec. VIII, pp. 34-35). They conclude that “the consumer plays a potentially important role in the transmission of monetary policy and other credit shocks,” but are explicit that the paper studies only “the second half of the consumer-credit channel… how people respond to changes in the supply of credit from lenders,” while “little is known about the first half of the channel, how credit card issuers and other consumer lenders change credit supply in response to monetary policy or other business cycle shocks” (Sec. VIII, p. 35-36) – a question the authors flag as a direction for future research rather than one this paper answers.

The working-paper text repeatedly points readers to “the working-paper version of this paper” for material omitted from what became the published article – for example, the intermediate cumulative impulse-response coefficients underlying Table II, and the full results on debt’s differential sensitivity to large versus small interest-rate changes (Sec. V, fn. 13 area / p. 18; Sec. VI, fn. 22, p. 24). This indicates the text obtained here is itself an iteration that anticipates further condensation before final publication; Crossref deposits no abstract for the published QJE version, so the two versions’ headline numbers could not be cross-checked against each other for this record, and the figures quoted above should be understood as the working-paper’s own reporting of its results pending that check.