The Marginal Propensity to Consume Out of Credit: Deniz Aydın

The Marginal Propensity to Consume Out of Credit: Evidence from Random Assignment of 54,522 Credit Lines Deniz Aydın WUSTL

Marginal Propensity to Consume /Credit Question: By how much does household expenditure increase, if credit lines are increased by 1$?

Marginal Propensity to Consume /Credit Question: By how much does household expenditure increase, if credit lines are increased by 1$? Answer distinguishes competing intertemporal models, ranging from the permanent income hypothesis to rule-of-thumb. Answer matters for understanding macroeconomic fluctuations through financial sector linkages, and formulating monetary, fiscal, macroprudential policy to offset them.

1. Measure the magnitude, heterogeneity and composition of the marginal propensity to consume out of credit i.e. the spending response to a 1$ increase in credit availability.

Event Study Note. Figure plots average interest bearing credit card debt at bank, D t. Levels normalized by pre-intervention values, D 1.

1. Measure the magnitude, heterogeneity and composition of the marginal propensity to consume out of credit i.e. the spending response to a 1$ increase in credit availability. Design a field experiment of unique size (N = 54, 522) and randomized nature, where credit lines are varied. Use in conjunction with comprehensive administrative data on income, spending, balance sheets and untapped credit. Complement literature on consumption response to income shocks Cochrane (1991) Johnson et al. (2006) Blundell et al. (2008) Baker (2013) This paper: Large sample + RCT! Precise/robust estimates Improve literature on borrowing response to credit shocks Gross and Souleles (2002) Agarwal et al. (2015) This paper: Balance sheet correlates, spending patterns

2. Use the empirical findings to test competing predictions of models of intertemporal behavior. Friedman (1957) Bewley (1977) Campbell and Mankiw (1989) Deaton (1991) Carroll (1997) Laibson (1997) Gourinchas and Parker (2002) Kaplan and Violante (2014) & a buffer-stock model with illiquid durables and one-sided adjustment costs is consistent many aspects of the MPC quantitatively. 3. Discuss implications for Consumption booms and household leveraging Hall (2011) Eggertsson and Krugman (2012) Guerrieri and Lorenzoni (2015) Fiscal, monetary and macroprudential policy Jappelli and Pistaferri (2014) Auclert (2015) Korinek and Simsek (2015) Welfare in credit markets Kaboski and Townsend (2011)

Data I European retail bank I I I I I I Credit card variables Categorized expenditures Balance sheet: Liquid assets and debt Credit bureau information Labor income Demographics I Sample size: 10+ million I Unit of analysis is an individual and frequency is monthly. See Credit card market See Installment contract

The Randomized Trial I Experiment participants (N=54,522) Pre-existing cardholders approved for a credit line increase See Sampling

Credit supply function I How the experiment participants are selected. # Type Variable Range Cutoff (1) Profitability Expected value added (-, ) 0 (2) CRM Months since limit increase [0, ) > 6 (3) Months card open [0, ) > 2 (4) Risk Behavior score [100, 800] < 180 (5) Customer score [100, 800] < 180 (6) Credit card score [100, 800] < 180 (7) Non-performing loans [0, ) = 0 (8) Late pay days [0, ) = 0 (9) RCT Control group {0, 1} = 0 (10) Regulatory Total limit to income [0, ) <4

The Randomized Trial I Experiment participants (N=54,522) Pre-existing cardholders approved for a credit line increase I Control group (N C =13,438) Determined by a random number generator Withheld from additional credit lines for T = 9 months. I Treatment group (N T =41,084) Limits extended by a median 120% of monthly income. See Sampling

Summary Statistics Group Count Age Income Limit Debt Credit score Risk score All cardholders >5m 41.3 2,488 8,664 2,261 1,460 428 Participants 54,522 37.3 2,785 5,452 1,236 1,480 355 P( P = All ) < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 Treatment 41,084 37.5 2,870 5,887 1,208 1,505 350 Treatment - US 27,668 37.9 3,047 6,704 1,044 1,555 338 Treatment 13,416 36.7 2,434 4,203 1,548 1,402 373 Control 13,438 36.6 2,494 4,121 1,523 1,401 373 P( T + US = C ) < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 < 0.01 P( T = C ) 0.30 0.29 0.17 0.39 0.78 0.55 Note. Table entries are group means unless otherwise noted. Row (1) based on a 50,000 random sample of all credit card holders in August 2014. Row (2) is based on 54,522 experiment participants in August 2014. Income, credit limit and credit card debt variables expressed in local currency. Labor income information for the subset of customers with direct deposit. Risk score represents the sum of three proprietary credit scores. See Robustness: Sampling

Event Study Note. Figure plots average interest bearing credit card debt at bank, D t. Levels normalized by pre-intervention values, D 1.

MPC out of Credit I Distributed lag DD it = T Â f j DL it j=0 j + # it I MPC is the cumulative response after t months. MPC(t) = t Â f j j=0 I Because the magnitude of the limit change is not random, instrument using I {DL it > 0}

Magnitude I Markets are incomplete and borrowing limit tighter than the natural limit. See Balance shifting See Robustness: Magnitude

Heterogeneity I Response not driven by a small fraction of credit constrained or rule-of-thumb consumers. See Heterogeneity by Age, Education, Gender, Marital Status See Heterogeneity by Income, Assets, Cash-on-hand, URE

Composition I Installments used to invest in durables and services. I Revolving debt used for self-insurance. DDebt = DInstallments + DRevolving

Spending patterns Note. Figure displays the share of each category in the additional credit card spending done by the treatment group. Some spending above is crowded out from cash transactions, or transactions outside the bank. See Spending patterns: Installments

Utilization Dynamics Note. Figure plots average debt/limit of consumers binned at t=0. Consumers in the top bin have debt/limit in excess of 0.9 at t=0, and so on. Figure uses data from the universe of all cardholders. See Robustness: Universe of cardholders

Utilization Dynamics: Percentiles Note. Figure plots percentiles of debt/limit of consumers constrained at t=0. Figure uses data from the universe of all cardholders. See Dynamics: Percentiles, unconstrained

Credit in a Simple Intertemporal Model Discrete time, infinite horizon consumption/savings problem (1) Budget constraint: No credit card puzzle. (2) Uninsurable idiosyncratic income shocks: Realistic persistence r and unemployment risk. (3) Credit constraint: Ad-hoc limit L, tighter than the natural limit. Ṽ(A 0, Y 0 ; L) = max C t, A t+1 E 0 Â t=0 b t C1 t g 1 g s.t. C t + DA t apple Y t (1) Y t+1 P(Y t ) (2) D t+1 apple L (3) See Calibration

Conceptual framework

Nested Intertemporal Models: MPCL and Testable Implications 0 MPC out of Credit 1! Buffer-stock CEQ-PIH Buffer-stock w/durables Myopia Rule-of-thumb See Quantitative model

Nested Intertemporal Models: MPCL and Testable Implications 0 MPC out of Credit 1! Buffer-stock CEQ-PIH Buffer-stock w/durables Myopia Rule-of-thumb Credit affects behavior? N X X X X See Quantitative model

Nested Intertemporal Models: MPCL and Testable Implications 0 MPC out of Credit 1! Buffer-stock CEQ-PIH Buffer-stock w/durables Myopia Rule-of-thumb Unconstrained respond? Deliver magnitude quantitatively? mean- Utilization reverting? Credit affects behavior? N X X X X N X X X N N N X X N N X X N N See Quantitative model

Credit in a Simple Intertemporal Model Discrete time, infinite horizon consumption/savings problem (1) Budget constraint: No credit card puzzle. (2) Uninsurable idiosyncratic income shocks: Realistic persistence r and unemployment risk. (3) Credit constraint: Ad-hoc limit L, tighter than the natural limit. (+) Illiquid durables: Depreciation d, liquidation cost z, no adjustment cost upwards. (+) Borrowing/lending rate spread. Ṽ(A 0, Y 0 ; L) = max C t, A t+1,k t E 0 Â t=0 b t (C a t K1 a t ) 1 g 1 g s.t. C t + DA t apple Y t + DD t L(K t+1, K t ) (1) Y t+1 P(Y t ) (2) D t+1 apple L (3) See Calibration

Thank you! daydin@wustl.edu