## wpiea2023054-print-pdf - Appendix C additionally plots net rewards for classic cards.  Additionally, Panel B of Figure A5 shows the

## Source details

**Canonical URL:** [wpiea2023054-print-pdf - Appendix C additionally plots net rewards for classic cards.  Additionally, Panel B of Figure A5 shows the](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023054-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023054-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023054-print-pdf.pdf.json)

---

### Net rewards by income and FICO
- Net rewards are higher for reward cards than for classic cards in all income groups.
- Average net rewards increase with income but remain positive in the bottom tercile: $1.9.
- Redistribution is primarily from low- to high-FICO cardholders, irrespective of income.
- Sub-prime cardholders (low FICO) net rewards by income tercile:
  - Highest income tercile: -$12.7
  - Middle-income tercile: -$4.9
  - Low-income tercile: -$2.6
- Prime and super-prime cardholders net rewards by income tercile:
  - High-income super-prime: $20.1
  - Middle-income super-prime: $13.6
  - Low-income super-prime: $9.7
- Combined inference: high-income consumers with high FICO scores benefit from reward credit cards largely at the expense of high-income consumers with low FICO scores; findings are not primarily driven by income.

### Financial sophistication: interpretation and measures
- FICO scores used as proxy for financial sophistication (payment history and outstanding debt relative to available credit).
- Higher FICO scores associated with lower interest payments, fee payments, and charge-offs.

Heterogeneity in empirical evidence
- Overindebtedness (bank-initiated credit limit increases, March 2019; DiD ±6 months):
  - Differential increase in spending on reward vs. similar classic cards: $76 (about 9% of average monthly spending).
  - Differential increase in repayments: $32.
  - Resulting differential increase in unpaid balances on reward cards vs. similar classic cards: $19.
  - Spending response larger for high-FICO consumers.
  - Only prime and super-prime cardholders significantly increase repayments; low-FICO consumers show statistically insignificant and near-zero repayment increases.
  - Credit limit increases on reward cards produce significant increases in unpaid balances for:
    - Sub-prime: $33.8
    - Near-prime: $25.3
  - Unpaid balances do not change significantly for high-FICO consumers.
  - Interpretation: reward cards induce sub- and near-prime consumers to overspend and overborrow.

- Sub-optimal repayment behavior (misallocated repayments and heuristics):
  - Misallocated payment share decreases in FICO (higher misallocation for low FICO).
  - Panel-level figures:
    - Super-prime: 6 percent misallocated payments on both reward and classic cards.
    - Sub-prime: up to 14 percent misallocated on reward cards and around 8 percent on classic cards.
  - Regression evidence (most restrictive sample):
    - Share of misallocated payments almost 2 percentage points higher on reward than on classic cards (column 5).
    - For sub-prime cardholders: 4.2 percentage point higher share of misallocated payments on reward cards.
    - No significant difference for prime- and super-prime cardholders.
  - Repayment heuristics:
    - Theoretical rules compared: (i) optimal repayment, (ii) balance-matching heuristic, (iii) equal allocation (1/N heuristic).
    - Actual payments most strongly correlated with balance-matching heuristic (Panel A, Table VIII).
    - Correlation stronger for sub-prime (Panel B) and near-prime (Panel C).
    - Prime (Panel D) and super-prime (Panel E) payments most strongly correlated with optimal allocation rule.
  - Overall interpretation: reward cards exploit over-borrowing and sub-optimal repayment behavior of low-FICO consumers; consistent with redistribution from näive to sophisticated consumers.

### Banks’ perspective: pricing and profits
- Pricing:
  - Average APRs on reward cards systematically lower than APRs on comparable classic cards across the FICO distribution.
  - Regression results (APR outcome):
    - APRs on reward cards on average 1.0 percentage points lower than on comparable classic cards.
    - Interest rate differential by FICO:
      - Sub-prime cardholders: 0.2 percentage points lower APR on reward cards.
      - Super-prime cardholders: 1.7 percentage points lower APR on reward cards.
  - Interpretation: banks incentivize adoption of reward cards by offering better pricing terms.

- Bank profit accounting and components:
  - Profit_i = Interest Paid_i + Total Fees_i + Interchange Income_i − Rewards_i − Realized Charge-Offs_i − WACC×Unpaid Balances_i
  - Interchange Income assessed at 1.5 percent of purchase volume for classic cards and 2.5 percent for reward cards.
  - Realized charge-offs defined as expense on accounts delinquent for 180 days and with outstanding balance no longer an asset.
  - Banks’ cost of financing revolving credit card balances assessed at WACC = 5 percent.
  - Interpretation: realized charge-offs and WACC are important determinants of ex-post profitability.

- Bank profits by FICO and card type:
  - Bank profits are hump-shaped in FICO and substantially higher on reward than on classic cards across the FICO distribution.
  - Across all cardholders, bank profits are about $7.4 higher on reward cards than on comparable classic cards.
  - Bank profits per card (reward vs. classic differential) by FICO segment:
    - Near-prime: $15.3
    - Prime: $9.0
    - Sub-prime: $4.1
    - Super-prime: $1.3
  - Interpretation: near-prime and prime cardholders are the largest source of profits for reward cards; sub-prime incur highest charge-offs but smaller reward-classic profit differential; super-prime yield small profit differential due to many rewards and low interest.

### Revenue composition across the FICO distribution
- Average revenue share components vary with FICO:
  - Low-FICO cardholders: revenues largely from interest income.
  - High-FICO cardholders: revenues largely from interchange income.
  - Fee income: smallest revenue source across FICO distribution.

### Geography of net rewards and spatial redistribution
- High spatial correlation between average net reward and average FICO across counties:
  - Regions with high average net rewards (northeast, north, west coast) tend to have high average FICO scores.
  - Regions with low (negative) average net rewards (south) tend to have low average FICO scores.
- Implication: reward-induced redistribution operates at regional level and mirrors individual-level redistribution from naive to sophisticated consumers.

### ZIP code regression analysis: socio-demographic correlates
- Regression: Net Reward_i,z = Σ_k β_k X_kz + γ × CreditScore_z + ε_i,z with X_kz:
  - percentage with high school diploma (no more),
  - median individual income,
  - percentage Black or African American.
- Results (columns 1, 3, 5):
  - Higher net rewards associated with higher educational attainment.
  - Higher net rewards associated with higher median income.
  - Higher net rewards associated with lower share of Black residents.
  - Interpretation: rewards can exacerbate socio-economic disparities (transfers from less to more educated, poorer to richer, high- to low-minority areas).
- When controlling for ZIP average FICO (columns 2, 4, 6):
  - Coefficients become statistically insignificant and close to zero.
  - Interpretation: average FICO (financial sophistication) is the underlying mechanism.

### Key aggregate magnitudes and robustness
- Annual redistribution estimate: $15.1 billion transferred from low-FICO to high-FICO consumers due to reward credit cards.
- Robustness:
  - Redistribution from high-FICO to low-FICO holds within low-, middle- and high-income groups.
  - High-FICO high-income consumers benefit most, at expense of low-FICO high-income consumers.
  - Quasi-experimental evidence: reward cards induce low-FICO consumers to overborrow.
  - FICO scores strongly correlated with share of misallocated payments, especially for sub-prime and near-prime.
  - Banks offer lower interest rates on reward cards than comparable classic cards to incentivize use.

### Key datasets and sample scope (selected figures)
- Baseline sample: 238,000,000 credit cards in March 2019.
- Reward cards in referenced samples: 91,000,000 (Table A1).
- Misallocated payments sample: 34,000,000 credit cards (Figure 4).
- Two-card sample: 13,080,528 observations (Table A7).
- Many tables report observations = 237,573,278 (card-level regressions).

### Representative card-level statistics (Table I; March 2019)
- Rewards (in $): Mean 4.69; Median 0.00; SD 20.42. Reward Cards mean 9.30; Classic Cards mean 0.00.
- Interest Charges (in $): Mean 14.38; Median 0.00; SD 37.91. Reward Cards mean 18.34; Classic Cards mean 10.36.
- Fee Charges (in $): Mean 2.64; Median 0.00; SD 11.01. Reward Cards mean 3.33; Classic Cards mean 1.93.
- Net Rewards (in $): Mean -12.33; Median 0.00; SD 44.41. Reward Cards mean -12.37; Classic Cards mean -12.29.
- APR (in %): Mean 20.63; Median 21.49; SD 7.15. Reward Cards mean 18.64; Classic Cards mean 22.64.
- Bank Profits (in $): Mean 14.53; Median 1.11; SD 232.94. Reward Cards mean 22.54; Classic Cards mean 6.39.
- FICO Score: Mean 729.60; Median 742.00; SD 75.65. Reward Cards mean 743.22; Classic Cards mean 715.77.
- Borrower Income (in $k): Mean 88.44; Median 60.00; SD 1863.36. Reward Cards mean 98.02; Classic Cards mean 78.71.
- Credit Limit (in $k): Mean 7.37; Median 5.00; SD 7.90. Reward Cards mean 10.42; Classic Cards mean 4.28.
- Observations reported: 237,573,278 total; Reward Cards 119,730,353; Classic Cards 117,842,925.

### Conclusion: distributional and policy-relevant implications
- Reward credit cards constitute a redistribution from naive to sophisticated consumers, with FICO scores interpreted as a measure of financial sophistication.
- Banks capture higher profits from reward cards concentrated among near-prime and prime consumers.
- Credit card rewards contribute to widening spatial disparities by transferring income from less to more educated areas, from poorer to richer areas, and from high- to low-minority areas.

*wpiea2023054-print-pdf — https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023054-print-pdf.pdf*

### Appendix C additionally plots net rewards for classic cards.  Additionally, Panel B of Figure A5 shows the

### wpiea2023054-print-pdf - Appendix C additionally plots net rewards for classic cards.  Additionally, Panel B of Figure A5 shows the

### Net rewards by income and FICO
- Net rewards are higher for reward cards than for classic cards in all income groups.
- Average net rewards increase with income but remain positive in the bottom tercile: $1.9.
- Redistribution is primarily from low- to high-FICO cardholders, irrespective of income.
- Sub-prime cardholders (low FICO) net rewards by income tercile:
  - Highest income tercile: -$12.7
  - Middle-income tercile: -$4.9
  - Low-income tercile: -$2.6
- Prime and super-prime cardholders net rewards by income tercile:
  - High-income super-prime: $20.1
  - Middle-income super-prime: $13.6
  - Low-income super-prime: $9.7
- Combined inference: high-income consumers with high FICO scores benefit from reward credit cards largely at the expense of high-income consumers with low FICO scores; findings are not primarily driven by income.

### Financial sophistication: interpretation and measures
- FICO scores are used as a proxy for financial sophistication, reflecting payment history and outstanding debt relative to available credit.
- Higher FICO scores are associated with lower interest payments, fee payments, and charge-offs.

#### A. Overindebtedness (responses to bank-initiated credit limit increases in March 2019)
- Empirical strategy: difference-in-differences comparing changes in spending, repayments, and unpaid balances for cards receiving bank-initiated credit limit increases (±6 months).
- Average effects across all cardholders:
  - Differential increase in spending on reward vs. similar classic cards: $76 (corresponds to about 9% of average monthly spending).
  - Differential increase in repayments: $32.
  - Resulting differential increase in unpaid balances on reward cards vs. similar classic cards: $19.
- Heterogeneity by FICO:
  - Spending response larger for high-FICO consumers.
  - Only prime and super-prime cardholders significantly increase repayments; low-FICO consumers show statistically insignificant and near-zero repayment increases.
  - Credit limit increases on reward cards produce significant increases in unpaid balances for:
    - Sub-prime: $33.8
    - Near-prime: $25.3
  - Unpaid balances do not change significantly for high-FICO consumers.
- Interpretation: reward cards induce sub- and near-prime consumers to overspend and overborrow, consistent with FICO as a measure of financial sophistication.

#### B. Sub-optimal repayment behavior (misallocated repayments and heuristics)
- Misallocated payment share (borrower-level, aggregated across card types):
  - Decreases in FICO scores (higher misallocation for low FICO).
  - Panel-level figures:
    - Super-prime: 6 percent misallocated payments on both reward and classic cards.
    - Sub-prime: up to 14 percent misallocated on reward cards and around 8 percent on classic cards.
- Regression evidence (Equation (2) with misallocated payments outcome):
  - In the most restrictive sample (cards with different APRs; individuals with at least two cards with unpaid balances; made minimum payments on all cards; more than the minimum on at least one card):
    - Share of misallocated payments is almost 2 percentage points higher on reward than on classic cards (column 5).
    - For sub-prime cardholders: 4.2 percentage point higher share of misallocated payments on reward cards.
    - No significant difference between reward and classic cards for prime- and super-prime cardholders.
  - Results robust to relaxed sample restrictions and to restricting the sample to individuals with only two cards.
- Repayment heuristics:
  - Calculated theoretical repayment amounts under three rules:
    - (i) optimal repayment rule,
    - (ii) balance-matching heuristic,
    - (iii) equal allocation across all cards (1/N heuristic).
  - Actual payments are most strongly correlated with the balance-matching heuristic (Panel A of Table VIII).
  - Heterogeneity:
    - Correlation between actual payments and balance-matching heuristic stronger for sub-prime (Panel B) and near-prime (Panel C).
    - Prime (Panel D) and super-prime (Panel E) cardholders’ repayment behavior most strongly correlated with the optimal allocation rule.
- Overall interpretation: reward cards exploit over-borrowing and sub-optimal repayment behavior of low-FICO consumers; results are consistent with redistribution from näive to sophisticated consumers. The source of lack of financial sophistication (time-inconsistency, low financial literacy, attentional neglect, etc.) remains undetermined.

### Banks’ perspective: pricing and profits
#### A. Pricing
- Average APRs on reward cards are systematically lower than APRs on comparable classic cards across the entire FICO distribution.
- Regression results (Equation (2) with APR as outcome):
  - Across all cardholders, APRs on reward cards are on average 1.0 percentage points lower than on comparable classic cards.
  - Interest rate differential by FICO:
    - Sub-prime cardholders: 0.2 percentage points lower APR on reward cards.
    - Super-prime cardholders: 1.7 percentage points lower APR on reward cards.
- Interpretation: banks incentivize adoption of reward cards by offering better pricing terms despite lower APRs.

#### B. Bank profits (profit accounting and components)
- Bank profit on credit card i defined as:
  - Profit_i = Interest Paid_i + Total Fees_i + Interchange Income_i − Rewards_i − Realized Charge-Offs_i − WACC×Unpaid Balances_i
- Definitions:
  - Interest Paid, Total Fees, Rewards as defined in Section III (income from cardholder perspective correspond to bank income/costs accordingly).
  - Interchange Income: assessed at the card level to be 1.5 percent of the purchase volume for classic cards (text ends at this assessment).

*Italic: wpiea2023054-print-pdf - Appendix C additionally plots net rewards for classic cards.  Additionally, Panel B of Figure A5 shows the*

### 2.5 percent for reward cards.  Realized charge-offs are an expense incurred by the bank

### wpiea2023054-print-pdf - 2.5 percent for reward cards.  Realized charge-offs are an expense incurred by the bank

### Bank charge-offs, costs, and profit measurement
- Realized charge-offs are an expense incurred by the bank on accounts that remain delinquent for 180 days and for which the outstanding balance can no longer be considered an asset on the balance sheet (CFPB, 2019).
- From the card-holder’s perspective, charge-offs do not matter for the net cash flow on a credit card; from a bank’s perspective, realized charge-offs are an important determinant of the ex-post profitability of an account and are included in the definition of banks’ profits.
- Banks’ cost of financing revolving credit card balances is assessed at a conservative 5 percent weighted average cost of capital (WACC).
- The analysis references "2.5 percent for reward cards."

### Bank profits by FICO and card type
- Bank profits are hump-shaped in FICO scores and substantially higher on reward than on classic cards across the entire FICO distribution.
- Across all cardholders, bank profits are about $7.4 higher on reward cards than on comparable classic cards.
- Bank profits per card by FICO segment (reward vs. classic differential):
  - Near-prime: $15.3
  - Prime: $9.0
  - Sub-prime: $4.1
  - Super-prime: $1.3
- Interpretation: near-prime and prime cardholders are the largest source of profits in the market for reward credit cards; sub-prime cardholders incur the highest charge-offs but the reward-classic profit differential is smaller; super-prime cardholders earn many rewards and incur low interest payments, yielding a small profit differential.

### Revenue composition across the FICO distribution
- Average revenue share components (interest income, fee income, interchange income) vary with FICO score:
  - Low-FICO cardholders: banks’ revenues largely stem from interest income.
  - High-FICO cardholders: banks’ revenues largely stem from interchange income.
  - Fee income: represents the smallest revenue source of banks across the FICO distribution.

### Geography of net rewards and spatial redistribution
- Aggregate patterns show a high level of spatial correlation between average net reward and average FICO score across counties.
  - Regions with high average net rewards (the northeast, the north, and the west coast) tend to have high average FICO scores.
  - Regions with low (negative) average net rewards (the south) tend to have low average FICO scores.
- Implication: reward-induced redistribution operates at the regional level and mirrors individual-level redistribution from naive to sophisticated consumers.

### ZIP code regression analysis: socio-demographic correlates
- Regression specification: Net Reward_i,z = Σ_k β_k X_kz + γ × CreditScore_z + ε_i,z, where X_kz are ZIP code-level characteristics:
  - i) percentage of residents with a high school diploma (but no more) — measure for low educational attainment;
  - ii) median individual income;
  - iii) percentage of residents who report their race as Black or African American.
- Results (columns 1, 3, and 5 of Table X):
  - Higher net rewards are associated with higher educational attainment.
  - Higher net rewards are associated with higher median income.
  - Higher net rewards are associated with a lower share of Black residents.
  - Interpretation: credit card rewards can exacerbate existing socio-economic disparities, implying transfers from less to more educated, from poorer to richer, and from high- to low-minority areas.
- When controlling for a ZIP code’s average FICO score (columns 2, 4, and 6 of Table X):
  - All coefficients become statistically insignificant and close to zero in magnitude.
  - Interpretation: differences in financial sophistication (average FICO score) are the underlying mechanism driving the geographical results.

### Key aggregate magnitudes and robustness
- Annual redistribution estimate: $15.1 billion transferred from low-FICO to high-FICO consumers due to reward credit cards.
- Results are robust within income sub-samples: the redistribution from high-FICO to low-FICO holds within low-, middle- and high-income groups.
- Notable pattern: high-FICO high-income consumers benefit the most from reward credit cards, but high-FICO high-income consumers benefit at the expense of low-FICO high-income consumers.
- Evidence on mechanisms:
  - Quasi-experimental evidence shows reward credit cards induce low-FICO consumers to overborrow.
  - FICO scores are strongly correlated with the share of misallocated credit card payments, especially for sub-prime and near-prime cardholders.
  - Banks offer lower interest rates on reward cards than on comparable classic cards to incentivize use.

### Conclusion: distributional and policy-relevant implications
- Reward credit cards constitute a redistribution from naive to sophisticated consumers, with FICO scores interpreted as a measure of financial sophistication.
- Banks capture higher profits from reward cards concentrated among near-prime and prime consumers.
- Credit card rewards contribute to widening spatial disparities by transferring income from less to more educated areas, from poorer to richer areas, and from high- to low-minority areas.

*Source: wpiea2023054-print-pdf — https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023054-print-pdf.pdf*

### REFERENCES

### wpiea2023054-print-pdf - REFERENCES (Selected empirical findings, figures, and tables)

### Key datasets and sample scope
- Baseline sample: 238,000,000 credit cards in March 2019.
- Analysis subsamples referenced: 91,000,000 reward cards (Table A1); sample of 34,000,000 credit cards for misallocated payments analysis (Figure 4); two-card sample observations reported (Table A7: 13,080,528 observations for some specifications).
- Many tables report observations = 237,573,278 (card-level regressions, Tables I, IV, IX, X).

### Card-level summary statistics (Table I; as of March 2019)
- Panel A (Net Reward Variables, means):
  - Rewards (in $): Mean 4.69; Median 0.00; SD 20.42. Reward Cards mean 9.30; Classic Cards mean 0.00.
  - Interest Charges (in $): Mean 14.38; Median 0.00; SD 37.91. Reward Cards mean 18.34; Classic Cards mean 10.36.
  - Fee Charges (in $): Mean 2.64; Median 0.00; SD 11.01. Reward Cards mean 3.33; Classic Cards mean 1.93.
  - Net Rewards (in $): Mean -12.33; Median 0.00; SD 44.41. Reward Cards mean -12.37; Classic Cards mean -12.29.
- Panel B (Other variables, means):
  - APR (in %): Mean 20.63; Median 21.49; SD 7.15. Reward Cards mean 18.64; Classic Cards mean 22.64.
  - Bank Profits (in $): Mean 14.53; Median 1.11; SD 232.94. Reward Cards mean 22.54; Classic Cards mean 6.39.
  - FICO Score: Mean 729.60; Median 742.00; SD 75.65. Reward Cards mean 743.22; Classic Cards mean 715.77.
  - Borrower Income (in $k): Mean 88.44; Median 60.00; SD 1863.36. Reward Cards mean 98.02; Classic Cards mean 78.71.
  - Credit Limit (in $k): Mean 7.37; Median 5.00; SD 7.90. Reward Cards mean 10.42; Classic Cards mean 4.28.
  - Observations reported: 237,573,278 total; Reward Cards 119,730,353; Classic Cards 117,842,925.

### Net rewards and distributional patterns (Figures 1–3; Tables II, IV, V)
- Figure 1: Net rewards across FICO percentiles plotted for 100 equal-sized FICO buckets between 480 and 830; dashed vertical cut-offs at FICO scores 660, 720, and 780 (near-prime, prime, super-prime). Based on 238 million cards.
- Table II (Baseline net rewards regressions; Equation (2) outcomes are dollar net rewards):
  - Coefficient Reward Card (pooled) reported as 4.66*** (column 1), 3.88*** (column 2), 3.48*** (column 3).
  - Reward Card × Sub-Prime: -5.37***.
  - Reward Card × Near-Prime: -6.80***.
  - Reward Card × Prime: 7.28***.
  - Reward Card × Super-Prime: 16.05***.
  - Observations: 237,573,278.
- Table IV (Net reward components; dollar outcomes):
  - Reward Card → Rewards: 6.38*** (col 1) and 2.20*** (col 2 in alternative specification).
  - Reward Card → Interest Charges: 0.70*** (col 3) and 0.70*** (col 4?) [table presents interest results in columns 3 and 4].
  - Reward Card × Sub-Prime → Interest Charges: 6.38***.
  - Reward Card × Near-Prime → Interest Charges: 10.86***.
  - Reward Card × Super-Prime → Interest Charges: -7.09***.
  - Reward Card → Total Fee Charges: 0.70*** (presented as positive and significant in fee columns).
  - Observations: 237,573,278.
- Table V (Net rewards by income terciles):
  - Reward Card (Lower tercile): 1.86***.
  - Reward Card (Middle tercile): 2.73***.
  - Reward Card (Upper tercile): 5.36***.
  - Reward Card × Sub-Prime (Upper tercile): -12.75***.
  - Reward Card × Super-Prime (Upper tercile): 20.10***.
  - Observations by income tercile: 75,159,536; 79,540,729; 82,873,013.

### Misallocated payments and repayment heuristics (Figure 4; Tables VII, VIII, A5, B description)
- Figure 4: Share of misallocated payments plotted across 100 FICO buckets between 480 and 830; cut-offs at 660, 720, 780. Based on sample of 34,000,000 credit cards of borrowers with multiple credit cards at same bank in March 2019.
- Appendix B defines misallocated payment (MP) share:
  - MP Share = (Actual Payment Amount_i,b − Optimal Payment Amount_i,b) / Total Payment Amount_i,b if APA_i,b > OPA_i,b; 0 if APA_i,b ≤ OPA_i,b. (Equation A5)
  - Interpretation: share of payments incorrectly made on cheaper cards that should have been made on more expensive cards.
  - Statement: Low-FICO consumers misallocate more than 6 percent of all credit card repayments; high-FICO consumers misallocated share is less than 2 percent.
- Table VII (Share of Misallocated Payments regressions):
  - Reward Card main effect: 1.24*** (col 1), 1.71*** (col 2), 1.74*** (col 3).
  - Reward Card × Sub-Prime: 2.65***; 3.74***; 4.18*** across specifications.
  - Reward Card × Near-Prime: 0.76***; 1.15***; 1.08***.
  - Restrictions and sample sizes indicated (e.g., Observations up to 21,288,917 and down to 12,858,916 across specifications).
- Table VIII (Heuristics comparison; mean shares and correlations):
  - Panel A (All cardholders, n = 21,288,917):
    - Actual Share of Payment: Payment on Reward Card(s) Mean 48.7%; Payment on Classic Card(s) Mean 35.9%.
    - Optimal Share of Payment: Reward Mean 47.0%; Classic Mean 37.5%; correlation (ρ) with actual: 0.50 (Reward), 0.49 (Classic).
    - Balance Matching Heuristic Share: Reward Mean 47.5% (ρ 0.52); Classic Mean 37.0% (ρ 0.54).
    - 1/N Heuristic Share: Reward Mean 42.8% (ρ 0.38); Classic Mean 41.4% (ρ 0.35).
  - Panel E (Super-prime, n = 1,924,235):
    - Actual Share of Payment: Reward 53.8%; Classic 32.9%.
    - Optimal Share: Reward 52.1% (ρ 0.63); Classic 34.5% (ρ 0.58).
    - 1/N Heuristic Share: Reward 43.4% (ρ 0.37); Classic 43.2% (ρ 0.26).

### Overindebtedness and credit-limit increase difference-in-differences (Tables VI, A6)
- Table VI (Difference-in-differences; comparing bank-initiated credit limit increases on reward vs classic cards; 6-month pre/post window):
  - ∆Spending: Reward Card coefficient 75.77*** (col 1), 31.96*** (col 2 alternative).
  - ∆Payments: Reward Card coefficient 31.96*** (col 2), 19.17** (col 3? columns indicate multiple outcomes).
  - ∆Unpaid Balances: Reward Card coefficient 19.17** (col 5) and other reported coefficient patterns by FICO group.
  - Reward Card × Super-Prime on ∆Spending: 164.85***; on ∆Payments: 153.22***; on ∆Unpaid Balances: -28.20 (not significant in reported specification).
  - Observations: 1,236,604 (reported for DiD analysis).
- Table A6 (DiD only cards with a bank-initiated credit line increase):
  - Reward Card ∆Spending 75.21***; ∆Payments 43.76***; ∆Unpaid Balances 48.28***.
  - Reward Card × Super-Prime ∆Spending 169.17***; ∆Payments 156.26***; ∆Unpaid Balances -12.77 (not significant).
  - Mean Y (for spending outcomes) reported as 860.315 and 851.559 for sample means.

### APRs, bank profits, and revenue shares (Figure 5, Figure 6, Table IX)
- Figure 5: APRs and Bank Profits plotted across 100 FICO buckets between 480 and 830; cut-offs at 660, 720, 780. Based on 238 million cards.
- Table IX (APR and Profit regressions):
  - Reward Card → APR: -0.96*** (column 1); Reward Card × Prime: -1.34***; Reward Card × Super-Prime: -1.65***.
  - Reward Card → Profit: 7.48*** (column 3).
  - Reward Card × Near-Prime → Profit: 13.10***.
  - Reward Card × Prime → Profit: 9.80***.
  - Observations: 237,573,278.
- Figure 6: Bank revenue share components across FICO percentiles for 100 equal-sized FICO buckets between 300 and 850, separately for reward and classic cards; components plotted as interchange income (black), fee income (dark gray), and interest income (light gray). Based on 238 million cards.

### Geography of net rewards and FICO scores (Figure 7; Table X)
- Figure 7 maps county-level averages:
  - Panel A: Average Net Rewards Across Counties with ranges shown (e.g., -46.83 − -23.01 up to -14.74 − 0.76; also categories No data).
  - Panel B: Average FICO Scores Across Counties with ranges shown (e.g., 687.2 − 726.0 up to 750.9 − 774.3; also No data).
  - Based on 238 million credit cards in March 2019.
- Table X (ZIP code-level net rewards regressions; Equation (6)):
  - Education (ZIP share with bachelor’s) coefficient 0.29*** in one specification; alternative reported -0.01 (not significant).
  - Income coefficient 0.21*** in one specification; alternative 0.00 (not significant).
  - Black Population Share coefficient -0.14*** in one specification; alternative 0.00 (not significant).
  - Credit Score coefficient 0.19*** (reported in multiple columns).
  - Observations: 237,573,278.

### Fee composition (Appendix Table A3)
- Reward Card associated with higher fee component means (annual fee, late payment fee, other fee) with coefficients:
  - Reward Card → Annual Fee Charges 0.51***; → Late Payment Fee Charges 0.14***; → Other Fee Charges 0.06***.
  - Reward Card × Super-Prime → Other Fee Charges -0.12***.

### Estimating monthly net rewards (Online Appendix A; methodology and validation)
- Monthly net rewards defined: Net Rewards_i,t = Rewards_i,t − Interest Paid_i,t − Total Fees_i,t (Equation A1).
- Rewards estimated via product-cluster median reward rates (Equation A3–A4) across clusters defined by bank, credit card type, product type, card network, reward type, fee type, and fee level; yields 380 product clusters.
- Resulting average monthly reward in raw sample: $13.34 per reward card, implying extrapolated average annual reward $160.08; compared to CFPB (2019) reported annual rewards per account $167 (not created by this document but cited for validation).
- Total Fees calculated as sum of late, over limit, non-sufficient funds (NSF), cash advance, debt suspension, balance transfer, other, and monthly fees.

*This overlay summarizes the empirical figures, tables, and methodological definitions contained in the REFERENCES and accompanying appendix figures/tables of the supplied content unit.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023054-print-pdf.pdf_