## wpiea2019057

## Source details

**Canonical URL:** [wpiea2019057](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019057.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019057.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019057.pdf.json)

---

### 1. Introduction — overview and key predictions
- Scale and trends:
  - U.S. consumers made 158 billion purchases worth about $8.3 trillion in 2011.
  - Long-run migration from paper to electronic payments: card share rises relative to cash; broader electronic payments (card + ACH + remote payments) rise relative to paper.
- Empirical puzzles:
  - Many electronic payment technologies introduced decades ago (credit cards 1950s, ACH 1970s, debit cards 1980s) yet combined card market share remained around 50-60 percent into the 2000s.
  - Recent innovations (e.g., mobile payments) show slow transitions.
  - Interchange/merchant discount fees rose in the U.S. between 1997 and 2008 for Visa and MasterCard credit cards and major PIN debit networks.
  - Regulatory/legal responses include the Durbin Amendment (July 2010), Regulation II capping debit card interchange at about half of pre-regulation levels, and a $6.2 billion settlement in September 2018 between Visa, MasterCard and major issuers and U.S. retailers (pending court approval).
- Model contributions:
  - A dynamic two-sided-market model with heterogeneous consumers (income) and merchants (size), a monopoly card network that sets usage fees φ_μ (merchant) and φ_χ (consumer) and invests R in R&D to reduce per-dollar card service cost δ over time via δ_{t+1} = Γ(δ_t, R_t).
  - Adoption requires per-period fixed costs K_μ (merchant) and K_χ (consumer); per-dollar card service costs to merchants and consumers are δ_μ and δ_χ, only the sum δ = δ_μ + δ_χ matters.
  - Cash handling per-dollar costs are τ_μ (merchant) and τ_χ (consumer); card is more efficient when τ_μ + τ_χ > δ.
- Main theoretical predictions:
  - Higher-income consumers and larger merchants more likely to adopt cards due to high fixed adoption costs but lower marginal per-dollar costs.
  - R&D that lowers δ causes penetration into lower-income consumers and smaller merchants.
  - Three merchant types emerge: large (accept both cash and cards), medium (specialize; some card-only and some cash-only), small (cash-only).
  - Monopoly card network may set high merchant fees, inflating retail prices and reducing stores serving both payment types; a Ramsey planner would set lower usage fees and invest more in R&D.
  - Over time the model predicts merchant fees rise while consumer fees fall as card service costs decline.

### 3.1 Within-period analysis — equilibrium characterization and key thresholds
- Game structure:
  - Three-stage discrete-time game solved by backward induction: Stage I network sets fees (φ_μ, φ_χ); Stage II merchants and consumers decide adoption and prices; Stage III consumers choose purchases and payment means.
  - Focus on equilibrium with positive card adoption/usage; multiple equilibria possible.
- Endogenous thresholds determining participation:
  - Merchant-size thresholds: α_1 and α_0 (α_1 > α_0).
  - Consumer-income threshold: I_0.
- Merchant categories and conditions (expressions/thresholds preserved in model text):
  - Large merchants (α ≥ α_1):
    - Accept both cash and cards; charge π_{α;d} ≤ π_{α;h}; zero-profit contestable market condition binds (Eqs. (9)–(12) in source).
    - Threshold α_1 given by an expression involving K_μ, E(α), E_{I≥I_0}(I − K_χ), φ_μ, and τ_μ; if φ_μ > τ_μ, no merchant in this category.
  - Medium-size merchants (α_0 ≤ α < α_1):
    - Specialize: competing merchants for a good split into card-accepting (serve card customers only) and cash-only (serve cash customers only); zero-profit condition yields π_{α;d} (Eq. (13)); threshold α_0 determined by an expression (Eq. (14)).
    - Condition for any merchants to accept cards: 1 − φ_μ 1+φ_χ ≥ 1 − τ_μ 1+τ_χ.
  - Small merchants (α < α_0):
    - Cash-only; all customers pay cash.
- Consumer adoption and I_0:
  - Consumer holds card if ln U^d_{I,d} ≥ ln U^h_{I,d}; adopter income threshold I_0 characterized by a closed-form expression (Eq. (16) in source) involving K_χ, φ_χ, τ_χ, and integrals over π_{α;d}/π_{α;h}.
  - Additional pricing constraint for consumer incentive at the counter: φ_χ ≤ τ_χ.
- Two-sided market interaction notation:
  - Ζ_1 = ( 1 − φ_μ 1+φ_χ − 1 − τ_μ 1+φ_χ )
  - Ζ_0 = ( 1 − φ_μ 1+φ_χ − 1 − τ_μ 1+τ_χ ). (Eq. (18))
- Proposition 1 (existence of thresholds):
  - If fees satisfy τ_χ ≥ φ_χ and 1 − φ_μ 1+φ_χ ≥ 1 − τ_μ 1+τ_χ, then thresholds α_0, α_1, I_0 exist with α_0 and α_1 given by expressions that depend on K_μ, expectations over I − K_χ, and Ζ_0, Ζ_1 (Eqs. (20)–(22)).
- Card network profit maximization within period:
  - π(δ; E(I), K_μ, K_χ) = max_{φ_χ, φ_μ} E_{α≥α_0}(α) E_{I≥I_0}(I − K_χ) / E(α) (1 + φ_χ) × (φ_χ + φ_μ − δ). (Eq. (23))
  - Interpretation: profit = consumers’ total spending on cards × network markup (φ_χ + φ_μ − δ), subject to feasibility and threshold constraints.
  - Within-period equilibrium: network maximizes profit; consumers maximize utility; merchants break even; markets clear.
- Dynamic embedding:
  - Network value function and dynamic program:
    - V(δ_t; E(I_t), K_{μ,t}, K_{χ,t}) = max_{R_t} π(δ_t; ...) − R_t + β V(δ_{t+1}; ...) subject to δ_{t+1} = Γ(δ_t, R_t) and π(δ_t; ...) ≥ R_t. (Eqs. (24)–(26))
  - Inverse R&D cost R_t = Ψ(δ_t, δ_{t+1}); first-order and envelope conditions yield a second-order difference equation characterizing optimal R&D over time (Eqs. (29)–(30) and subsequent).
- Calibration functional forms (F1–F4) used for tractability:
  - F1: α uniform on (0,1): Γ(α) = α and E(α) = 1/2. I_t exponential: F(I_t) = 1 − e^{−λ_t I_t}, E(I_t) = 1/λ_t.
  - F2: mean consumer income growth: λ_{t+1} = λ_t / (1 + γ_I).
  - F3: adoption costs proportional to mean income: K_{μ,t} = κ_μ / λ_t; K_{χ,t} = κ_χ / λ_t.
  - F4: R&D Γ specified so 1/δ_{t+1} − 1/δ_t = ( R_t / λ_t ϖ )^θ δ_t^{−θ−1} with 1 > θ > 0 and ϖ fixed (Eq. (31)); R&D scaled by mean income.
- Within-period equilibrium under F1–F4:
  - Simplified expressions (L1) and (L2) for α_0 and I_0 with uniform α and exponential income:
    - α_0 = κ_μ / (2 e^{−λ I_0}) (1 + λ I_0 − κ_χ) Ζ_0. (L1)
    - I_0 = (1 + τ_χ 1 + φ_χ)^{1 − α_0^2} κ_χ / λ [ (1 + τ_χ 1 + φ_χ)^{1 − α_0^2} − exp(σ α_0^2) ]. (L2)
  - Two positive-adoption equilibria can exist; analysis focuses on the high-adoption stable equilibrium (I_0^*, α_0^*).
  - Simplified within-period profit approx: π(δ; λ) ≈ 1/λ ( α_0 − α_1 δ ), linear in δ for given λ. (Eq. (32))
- Dynamic formulation using F4:
  - R_t = ϖ δ_t / λ_t [ δ_t / δ_{t+1} − 1 ]^{1/θ} with 1 > θ > 0.
  - Dynamic optimization problem under linear profit solves for balanced-growth path (Eq. (33)).

### 4.3 Calibration results — fit to U.S. data and welfare comparisons
- Calibration sample and stylized facts:
  - Sample: U.S. payment card data 1997–2008.
  - Fact: Credit card broad adoption by late 1990s: 78% of U.S. consumers adopted credit cards.
  - Among credit card holders, about 50% were "convenience users."
  - Debit card adoption accelerated since mid-1990s.
  - 2008 volumes/values:
    - Credit cards: 26.5 billion transactions worth about $2.1 trillion.
    - Debit cards: 34 billion transactions worth about $1.3 trillion.
  - Asymmetric fees observed: merchant fees rose while consumer fees declined (consumer rewards).
- Calibrated parameter values (Table 1):
  - Merchant cost of handling cash μ: 4.0%
  - Consumer cost of handling cash χ: 2.5%
  - Merchant cost of adopting card κ_μ: 2.5%
  - Consumer cost of adopting card κ_χ: 0.3%
  - Merchant cost of goods : 1 (normalized)
  - Initial value of card service costs δ_0: 2.25%
  - R&D function curvature θ: 0.5
  - R&D efficiency scaler λ: 10
  - Initial value of mean income 1/_0: 21,215
  - Growth rate of mean income γ_I: 2%
- Interpretation of adoption-cost calibration:
  - Merchant adoption cost calibrated to imply adoption cost about $500 in 1997.
  - Consumer adoption cost calibrated to imply adoption cost about $60 in 1997.
- Model fit outcomes:
  - Rising merchant card fee over time; level falls into range between average merchant fees for credit and debit cards in data.
  - Declining consumer card fee over time, consistent with increasing consumer card rewards.
  - Increasing consumer card adoption and card transaction share (credit + debit); model slightly overstates card transaction share vs. data.
  - Merchant card adoption increases over time; shares of large merchants and small merchants decline.
  - Card network invests in R&D; R&D-to-mean-income ratio declines over time.
  - Declining card costs and rising card spending share lead to increasing network markup and rising profit.
  - Consumer welfare (scaled by mean income) rises over time in the card economy; in a cash economy it would be constant (normalized to 100 in figures).
- Welfare proposition:
  - Proposition 2: "No consumer is worse off given any positive measure of card adoption."
    - Equations (34) and (35) characterize welfare gains for adopters and nonadopters; cash consumers’ welfare cannot be lower in a card economy than in a cash economy under model assumptions.
- Ramsey social planner vs. monopoly network (calibrated comparative findings):
  - Ramsey sets lower merchant and consumer fees than monopoly; under Ramsey both fees decrease over time while monopoly raises merchant fees over time and lowers consumer fees.
  - Ramsey induces higher card adoption and usage; larger and increasing share of large merchants under Ramsey benefits cash users more than under monopoly.
  - Ramsey conducts more R&D, yielding a faster decline in δ; Ramsey charges a decreasing markup over time, monopoly an increasing markup.
  - Consumer welfare increases faster under Ramsey than under monopoly.
  - Social-welfare accounting:
    - Under Ramsey (zero profit) consumer welfare = social welfare.
    - Under monopoly, assessing social welfare by rebating profits to consumers proportionally to income: mean consumer income increases but adoption thresholds unchanged; in early years social welfare under Ramsey may be lower due to higher R&D but Ramsey social welfare grows faster and surpasses monopoly in a few years, with the gap widening.

### 5.3 Policy experiments — regulatory scenarios and dynamic trade-offs
- Regulatory approaches studied:
  - Marginal-cost pricing regulation: requires φ_μ,t + φ_χ,t = δ_t; network maximizes consumer welfare subject to zero markup.
  - Merchant fee cap regulation: caps merchant (interchange) fee at specified levels (simulated Caps: 0.5%, 1%, 2%).
- Implementation assumptions in simulations:
  - Regulation assumed implemented at beginning year of sample.
  - Under marginal-cost pricing the network lacks R&D incentive/resources so δ_t remains at δ_0.
  - Footnote: the monopoly profit π(δ;λ)=1/λ(α0−α1 δ); merchant fee cap reduces α0 but not α1 (may even increase it slightly); α1 influences rate of cost decline along balanced growth path.
- Key outcomes — marginal-cost pricing:
  - Compared with unregulated monopoly:
    - Fees to both merchants and consumers are lowered.
    - Card adoption by consumers and merchants increases.
    - Fraction of large merchants serving both card and cash users increases substantially.
    - Consumer welfare in each period rises relative to monopoly.
  - Dynamic cost and welfare trade-off:
    - Eliminates profit for R&D → δ_t stays at δ_0 (no cost decline).
    - Social welfare under marginal-cost pricing is lower than unregulated monopoly in present value despite higher static consumer welfare; marginal-cost pricing yields lower present value of social welfare except initial years.
- Key outcomes — merchant fee cap (Cap 1% example and general patterns):
  - With binding Cap 1%:
    - Card network raises consumer fee to recoup revenue.
    - Consumer card adoption is lower vs. unregulated monopoly.
    - Merchant adoption (including large merchants serving both payment types) is higher vs. unregulated monopoly.
    - Card transaction share changes little.
    - Markup (f_m + f_c − d) is lower; network profit reduced.
    - Network R&D spending constrained initially → slower decline in δ until late sample period.
  - Welfare dynamics:
    - Consumer welfare in each period is higher under the merchant fee cap than under monopoly.
    - Social welfare may be lower in early years and higher in the longer run.
  - Present-value comparisons:
    - Merchant fee cap redistributes surplus from network profit to consumers without greatly impairing R&D; lower caps raise consumer welfare but reduce network profit; social welfare changes little and can slightly increase with cap value.
- Scenarios explicitly compared in simulations:
  - Monopoly
  - Monopoly with Cap 0.5%
  - Monopoly with Cap 1%
  - Monopoly with Cap 2%
  - Marginal-Cost Pricing
  - Social Planner (Ramsey)
- Policy implications and trade-offs:
  - Marginal-cost pricing:
    - Static strength: maximizes consumer welfare.
    - Dynamic weakness: eliminates network profits and R&D, potentially reducing dynamic social welfare.
  - Merchant fee cap:
    - Easier to implement; improves consumer welfare and generally preserves much of R&D dynamics; redistributes surplus from network to consumers.
    - Trade-off: stricter caps raise consumer welfare but lower network profits; social welfare effects small and can be slightly positive for certain cap values.
  - Ramsey social planner:
    - Sets lower usage fees and conducts more R&D than monopoly, achieving higher adoption, usage, consumer and social welfare over time.
- Reporting and normalization:
  - Present values of consumer welfare and network profits computed with cash-economy present value normalized to 100.
  - Figures referenced in source (time-series and present-value comparisons): Figure 10, Figure 11, Figure 12, Figure 13 (policy experiment and welfare comparisons).

*Source: IMF working paper — "1. Introduction"; "3.1 Within-Period Analysis"; "4.3 Calibration Results"; "5.3 Policy Experiments" (wpiea2019057).*

### 1. Introduction ___________________________________________________________ 4

### 1. Introduction

### Overview and scale of payments
- U.S. consumers made 158 billion purchases worth about $8.3 trillion in 2011 (data source: Nilson Report #1008, December 2012).
- Over recent decades there has been steady migration from paper to electronic payments: the share of card relative to cash payments increases steadily over time, and the share of broadly measured electronic payments relative to paper payments also rises (card payments include credit and debit; broader electronic payments include card payments plus ACH and remote payments).

### Puzzles and empirical patterns
- Despite long-standing forecasts of a paperless payments system, adoption of electronic payments has been slow: many electronic payments familiar today were introduced decades ago (credit cards in the 1950s, ACH in the 1970s, debit cards in the 1980s), yet combined card market share remained around 50-60 percent into the 2000s.
- Recent innovations (e.g., mobile payments) have also seen slow transitions.
- Merchants have complained about rising fees to accept cards (merchant discount rates), with interchange fees increasing in the U.S. between 1997 and 2008 for Visa and MasterCard credit cards and major PIN debit networks (figure plotted 1997–2008; data source: American Banker).
- Regulatory and legal responses include the Durbin Amendment (a provision of the Dodd-Frank Act in July 2010 directing the Federal Reserve Board to regulate U.S. debit card interchange fees), subsequent Regulation II that caps debit card interchange fees at about half of the pre-regulation level, and a $6.2 billion settlement agreement in September 2018 between Visa, MasterCard and major issuers and U.S. retailers (pending court approval).

### Literature and limitations
- Money-theoretic approaches (e.g., Kiyotaki and Wright 1989; Lagos and Wright 2005) explain how payments arrangements overcome exchange frictions but do not directly explain slow adoption of electronic payments and often assume benevolent planners or clubs, leaving competitive efficiency issues unaddressed.
- Industrial-organization approaches (e.g., Rochet and Tirole 2002, 2011) highlight two-sided market failures and usage externalities but typically:
  - Ignore endogenous adoption decisions and technological progress (R&D).
  - Impose ad hoc distributions of “convenience benefits” and assume consumer demand for final goods is invariant to payment choices.
  - As a result, they do not fully explain adoption and usage patterns or provide comprehensive welfare analysis.

### Contributions and model overview
- The paper develops a unified, dynamic two-sided-market framework with heterogeneous consumers (income) and merchants (size) where a monopoly electronic payment network sets usage fees and conducts R&D to lower per-dollar service cost.
- Key modeling features:
  - Adoption requires merchants and consumers each to incur fixed costs each period (K_m and K_c).
  - Per-dollar card service costs to merchants and consumers are δ_μ and δ_χ; only the sum δ = δ_μ + δ_χ matters.
  - Cash handling imposes per-dollar transaction costs τ_μ (merchant) and τ_χ (consumer); card is more efficient when τ_μ + τ_χ > δ.
  - The card network charges percentage fees φ_μ (merchant) and φ_χ (consumer).
  - Time is discrete; the card network can invest R in R&D to reduce δ next period via δ_{t+1} = Γ(δ_t, R_t) with ∂Γ/∂δ_t > 0 and ∂Γ/∂R_t < 0.
  - Within each period the game is: Stage I network sets φ_μ and φ_χ; Stage II merchants and consumers decide on card adoption and merchant prices; Stage III consumers choose purchases and payment means. Merchants that accept cards still accept cash and post a single price (price coherence).

### Main theoretical predictions and equilibrium characterization
- Adoption and usage implications:
  - Payment cards have high fixed cost but low marginal cost relative to cash; thus higher-income consumers (higher consumption/purchases) are more willing to adopt cards than lower-income consumers.
  - Larger merchants or those selling higher-valued goods are more likely to accept cards.
  - As the card network conducts R&D and lowers card service costs over time, cards penetrate lower-income consumers and smaller merchants.
- Merchant heterogeneity yields three equilibrium merchant types:
  - Large merchants: accept both cash and cards, set a price lower than cash-only merchants, and attract both cash and card users.
  - Medium-size merchants: some are cash-only; others accept both cash and cards but set a higher price than competing cash-only merchants and therefore attract only card users (cards are overall cost-saving for them).
  - Small merchants: all cash-only.
- Competitive efficiency and network market power:
  - The card network’s market power helps explain slow adoption of electronic payments: it may set merchant fees high to extract rents, inflating retail prices and reducing the number of stores that serve both card and cash users.
  - A Ramsey social planner would set lower usage fees and invest more in R&D, achieving higher adoption and usage of electronic payments.
  - Over time the model predicts the card network raises merchant fees but reduces consumer fees as card service costs decline—consistent with observed data patterns.
- Cash users are disadvantaged by high merchant fees not because they subsidize card users at multi-payment merchants (in the model, cash users at such merchants are subsidized by card users due to lower overall payment costs), but because high merchant fees reduce the number of stores serving both card and cash users, limiting cross-subsidization.

### Policy insights (from calibrated model and experiments)
- Two regulatory approaches are compared:
  - Mandating marginal-cost pricing by the card network can maximize consumer welfare in a static setting but may shut off the network’s R&D incentives and reduce social welfare dynamically.
  - Capping merchant fees improves consumer welfare and causes less dynamic inefficiency relative to marginal-cost pricing.
- The calibrated model is used to match U.S. payment card pricing, adoption and usage data and to conduct welfare and policy analysis.

*Source: IMF working paper — “1. Introduction” (from wpiea2019057).*

### 3.1  Within-Period Analysis

### wpiea2019057 - 3.1  Within-Period Analysis

### Overview
- The model is a three-stage game solved backward: Stage II and III (merchants and consumers decide acceptance/usage given card fees φ_μ and φ_χ), Stage I (card network sets fees anticipating responses).
- Multiple equilibria may exist; focus is on an equilibrium with positive card adoption and usage.
- Key endogenous thresholds determine participation:
  - Merchant-size thresholds: α_1 and α_0 (α_1 > α_0).
  - Consumer-income threshold: I_0.

### Merchants’ Choices
- Merchants take card fees as given and expect consumers I ≥ I_0 to hold cards.
- Merchants are indexed by size α. Three merchant categories emerge:

  - Large merchants (α ≥ α_1):
    - Accept both cash and cards and charge price π_{α;d} ≤ π_{α;h}.
    - Revenues from card customers (I ≥ I_0) and cash customers (I < I_0):
      - π_{α;d} x_{card}^{α;d} = α[ E_{I≥I_0}(I − K_χ) ] / E(α) (1 + φ_χ)
      - π_{α;d} x_{cash}^{α;d} = α[ E_{I< I_0}(I) ] / E(α) (1 + τ_χ)   (Eq. (9))
    - Zero-profit contestable market condition:
      - (1 − φ_μ) π_{α;d} x_{card}^{α;d} + (1 − τ_μ) π_{α;d} x_{cash}^{α;d}
        = υ_α x_{card}^{α;d} + υ_α x_{cash}^{α;d} + K_μ   (Eq. (10))
    - Price π_{α;d} solved by Eqs. (9) and (10) (Eq. (11)).
    - Threshold α_1 given by
      - α_1 = K_μ / E(α) [ E_{I≥I_0}(I − K_χ) ] ( 1 − φ_μ 1+φ_χ − 1 − τ_μ 1+φ_χ )
        if φ_μ ≤ τ_μ.   (Eq. (12))
    - Note: if φ_μ > τ_μ, no merchant in this category.

  - Medium-size merchants (α_0 ≤ α < α_1):
    - Specialize: for each good, one accepts cards and serves card customers only; one accepts cash only and serves cash customers only.
    - Card-accepting merchant revenue only from card customers and zero profit implies:
      - π_{α;d} = υ_α / α [ E_{I≥I_0}(I − K_χ) ] / (1 + φ_χ) (1 − φ_μ) [ α E_{I≥I_0}(I − K_χ) / (1 + φ_χ) − K_μ E(α) ].  (Eq. (13))
    - Threshold α_0 determined by
      - α_0 = K_μ / E(α) [ E_{I≥I_0}(I − K_χ) ] ( 1 − φ_μ 1+φ_χ − 1 − τ_μ 1+τ_χ ).   (Eq. (14))
    - Condition for any merchants to accept cards:
      - 1 − φ_μ 1+φ_χ ≥ 1 − τ_μ 1+τ_χ.

  - Small merchants (α < α_0):
    - Only accept cash; all customers pay cash regardless of card holdings.

- Additional pricing constraint for consumer incentive at the counter: φ_χ ≤ τ_χ.

### Consumers’ Choices
- A consumer takes prices, card fees, and merchants’ acceptance as given and decides to hold a payment card if ln U^d_{I,d} ≥ ln U^h_{I,d}.
- Utility if not adopting (ln U^h_{I,d}) integrates over merchant categories and prices (expression given in text).
- Utility if adopting (ln U^d_{I,d}) accounts for card use at categories (1) and (2) and cash at category (3) (expression given in text).
- Consumer adoption condition leads to inequality (Eq. (15)):
  - E_{α≥α_0}(α) ln(1+τ_χ 1+φ_χ) ≥ E(α) ln( I I − K_χ ) + ∫_{min(α_1,α)}^{α} (α_0 α ln( π_{α;d} / π_{α;h} )) dΓ(α).
- Adopter income threshold I_0 characterized by (Eq. (16)):
  - I_0 = ( 1+τ_χ 1+φ_χ )^{E_{α≥α_0}(α)/E(α)} K_χ [ (1+τ_χ 1+φ_χ )^{E_{α≥α_0}(α)/E(α)} − exp( ∫_{min(α_1,α)}^{α_0} α E(α) ln( π_{α;d} / π_{α;h} ) dΓ(α) ) ]^{−1}
  - with π_{α;d} / π_{α;h} for α_0 ≤ α < min(α_1,α) given by Eq. (17).

### Two-sided Market Interactions and Proposition 1
- Define:
  - Ζ_1 = ( 1 − φ_μ 1+φ_χ − 1 − τ_μ 1+φ_χ )
  - Ζ_0 = ( 1 − φ_μ 1+φ_χ − 1 − τ_μ 1+τ_χ ).   (Eq. (18))
- Proposition 1 (conditions and thresholds):
  - If card fees satisfy τ_χ ≥ φ_χ and 1 − φ_μ 1+φ_χ ≥ 1 − τ_μ 1+τ_χ, then there exist thresholds α_0, α_1 and I_0 above which merchants and consumers accept and hold cards, where:
    - α_0 = K_μ / E(α) [ E_{I≥I_0}(I − K_χ) ] Ζ_0.   (Eq. (20))
    - α_1 = Ζ_0 / Ζ_1 α_0 if φ_μ ≤ τ_μ.   (Eq. (21))
    - I_0 expression given by Eq. (22) (replicates Eq. (16) with notation Ζ_0).

### Within-Period Equilibrium (Card Network Profit Maximization)
- Card network maximizes profit by choosing (φ_χ, φ_μ) anticipating thresholds (α_0, α_1, I_0):
  - π(δ; E(I), K_μ, K_χ) = max_{φ_χ, φ_μ} E_{α≥α_0}(α) E_{I≥I_0}(I − K_χ) / E(α) (1 + φ_χ) × (φ_χ + φ_μ − δ).   (Eq. (23))
  - Subject to constraints: (19), (20), (21), (22).
- Interpretation: network profit = consumers’ total spending on cards × network markup (φ_χ + φ_μ − δ).
- Optimal fees (φ*_χ, φ*_μ) internalize two-sided externalities and determine adoption thresholds; at the within-period equilibrium:
  - Card network maximizes profit.
  - Consumers maximize utility.
  - Merchants break even.
  - Goods and payments markets clear.

### Dynamic Analysis (Embedding within-period equilibrium)
- Time subscripts added. Network’s value function:
  - V(δ_t; E(I_t), K_{μ,t}, K_{χ,t}) = max_{R_t} π(δ_t; E(I_t), K_{μ,t}, K_{χ,t}) − R_t + β V(δ_{t+1}; E(I_{t+1}), K_{μ,t+1}, K_{χ,t+1}) subject to δ_{t+1} = Γ(δ_t, R_t) and π(δ_t; ...) ≥ R_t.   (Eqs. (24)-(26))
- Inverse of Γ: R_t = Ψ(δ_t, δ_{t+1}). If budget constraint never binding, dynamic problem reduces to:
  - V(δ_t; E(I_t), K_{μ,t}, K_{χ,t}) = max_{δ_{t+1}} π(δ_t; ...) − Ψ(δ_t, δ_{t+1}) + β V(δ_{t+1}; ...).   (Eq. (28))
- First-order and envelope conditions:
  - Ψ_2(δ_t, δ_{t+1}) = β V'_0(δ_{t+1}; ...).   (Eq. (29))
  - V'_0(δ_t; ...) = π'_0(δ_t; ...) − Ψ_1(δ_t, δ_{t+1}).   (Eq. (30))
- Combined second-order difference equation:
  - Ψ_2(δ_t, δ_{t+1}) = β [ π'_0(δ_{t+1}; ...) − Ψ_1(δ_{t+1}, δ_{t+2}) ].
- Boundary conditions: δ_0 given and δ_∞ = 0; laws of motion for E(I_t), K_{μ,t}, K_{χ,t} pin down the dynamic equilibrium path.

### Model Calibration — Functional Forms (F1–F4)
- F1. Distributions:
  - Merchant size α ∈ (0,1) uniformly distributed: Γ(α) = α and E(α) = 1/2.
  - Consumer income I_t ∈ (0,∞) exponentially distributed: F(I_t) = 1 − e^{−λ_t I_t} and E(I_t) = 1/λ_t.  (Footnote: E_{α≥α_0}(α) = (1 − α_0^2)/2; E_{I_t ≥ I_0}(I_t − K_{χ,t}) = e^{−λ_t I_0}( 1/λ_t + I_0 − K_{χ,t} ).)
- F2. Mean consumer income growth:
  - λ_{t+1} = λ_t / (1 + γ_I).
- F3. Card adoption costs proportional to mean income:
  - K_{μ,t} = κ_μ E(I_t) = κ_μ / λ_t.
  - K_{χ,t} = κ_χ E(I_t) = κ_χ / λ_t.
- F4. R&D function Γ:
  - 1/δ_{t+1} − 1/δ_t = ( R_t / λ_t ϖ )^θ δ_t^{−θ−1} with 1 > θ > 0 and ϖ fixed.   (Eq. (31))
  - R&D scaled by income: R_t / E(I_t) = R_t / λ_t enters function.

- Notes:
  - Uniform α distribution is inessential; results robust to other positively skewed distributions.
  - Exponential income matches U.S. data and yields constant Gini = 0.5.
  - F2–F3 simplify dynamics: linking K_{μ,t}, K_{χ,t} to mean income means exogenous mean income growth shifts profit function proportionally; adoption thresholds depend on δ only.
  - F4 models decline in card service cost δ_t due to R&D; scaling by mean income captures opportunity cost of R&D.

### Model Characterization — Within-Period and Dynamic Equilibrium
- Within-period equilibrium under F1–F4:
  - With α_1 < ̄α = 1, Eqs. (20) and (22) become (L1) and (L2):
    - α_0 = κ_μ / (2 e^{−λ I_0}) (1 + λ I_0 − κ_χ) Ζ_0.   (L1)
    - I_0 = (1 + τ_χ 1 + φ_χ)^{1 − α_0^2} κ_χ / λ [ (1 + τ_χ 1 + φ_χ)^{1 − α_0^2} − exp(σ α_0^2) ].   (L2)
    - σ is a scalar determined by model parameters (details in Appendix A2).
  - For plausible parameter values, two positive-adoption equilibria can exist:
    - High-adoption equilibrium (I_0^*, α_0^*): stable — selected for analysis.
    - Low-adoption equilibrium (I_0^{0}, α_0^{0}): unstable.
- Simplified profit function under F3:
  - π(δ; λ) ≈ 1/λ ( α_0 − α_1 δ ), linear in δ for given λ.   (Eq. (32))
- Dynamic equilibrium:
  - From F4, R_t = ϖ δ_t / λ_t [ δ_t / δ_{t+1} − 1 ]^{1/θ} with 1 > θ > 0.
  - Dynamic optimization problem becomes:
    - V(δ_t; λ_t) = max_{δ_{t+1}} π(δ_t; λ_t) − ϖ δ_t / λ_t [ δ_t / δ_{t+1} − 1 ]^{1/θ} + β V(δ_{t+1}; λ_{t+1}).   (Eq. (33))
  - Using the linear profit representation (Eq. (32)), the dynamic problem can be solved for a balanced-growth path (details in Appendix A3).

*Source: wpiea2019057 - 3.1  Within-Period Analysis (IMF working paper section).*

### 4.3  Calibration Results

### 4.3  Calibration Results

### Industry background
- Calibration sample: U.S. payment card data from 1997-2008.
- Historical and market facts reported in the source:
  - Credit card broad adoption by late 1990s: 78% of U.S. consumers adopted credit cards.
  - Among credit card holders, about 50% were "convenience users" who only used the payment function (paid off purchases each month).
  - Debit cards adoption accelerated since mid-1990s.
  - 2008 transaction volumes and values:
    - Credit cards: 26.5 billion transactions worth about $2.1 trillion.
    - Debit cards: 34 billion transactions worth about $1.3 trillion.
- Empirical controversy motivating analysis: asymmetric fees where merchants are typically charged a much higher fee than consumers; interchange fee is the major component. Over the calibration period, merchant fees rose while consumer fees declined (consumers in some cases received rewards).

### Model fit and parameter calibration
- Data sources used for calibration:
  - Interchange fees: American Banker (various issues); value-weighted averages for Visa, MasterCard, and top PIN debit networks; a merchant acquiring fee added so total merchant discount rate for credit cards is around 3% in late 2000s.
  - Adoption of credit and debit cards: Survey of Consumer Finance (various years).
  - Annual transaction values for cash, credit card and debit card: Nilson Report (various issues).
- Key calibrated parameter values (as reported in Table 1):
  - Merchant cost of handling cash μ: 4.0%
  - Consumer cost of handling cash χ: 2.5%
  - Merchant cost of adopting card κ_μ: 2.5%
  - Consumer cost of adopting card κ_χ: 0.3%
  - Merchant cost of goods : 1 (normalized)
  - Initial value of card service costs δ_0: 2.25%
  - R&D function curvature θ: 0.5
  - R&D efficiency scaler λ: 10
  - Initial value of mean income 1/_0: 21,215
  - Growth rate of mean income γ_I: 2%
- Interpretation of adoption-cost calibration:
  - Merchant adoption cost calibrated to imply adoption cost about $500 in 1997.
  - Consumer adoption cost calibrated to imply adoption cost about $60 in 1997.
- Model fit outcomes (consistent with data and generated by the calibrated model):
  - Rising merchant card fee over time; level falls into range between average merchant fees for credit and debit cards in data.
  - Declining consumer card fee over time (consistent with increasing consumer card rewards).
  - Increasing consumer card adoption and increasing card transaction share (sum of credit and debit), with the model producing a slightly higher level of card transaction share than the data.
  - Additional model patterns:
    - Overall merchant card adoption increases over time; shares of large merchants (accept both cash and card but charge lower prices than cash-only stores) and small merchants (accept cash only) decline.
    - Card network invests in R&D to reduce card service cost δ; R&D expenditure to mean income ratio declines over time.
    - Declining card costs and rising card spending share lead the card network to charge an increasing markup and earn increasing profit.
    - Consumer welfare (scaled by mean income) rises over time in the card economy; in a cash economy it would be constant (normalized to 100 in figures).

### Welfare implications and policy analysis
- Proposition reported:
  - Proposition 2: "No consumer is worse off given any positive measure of card adoption."
    - Equations (34) and (35) characterize welfare gains for card adopters and nonadopters respectively and show cash consumers’ welfare cannot be lower in a card economy than in a cash economy given model assumptions.
- Payment card and welfare improvement:
  - Introducing payment cards increases welfare for both card adopters and nonadopters in the model.
  - Welfare gains differ by consumer income: adopters receive gains from shopping at card-accepting stores across merchant categories and from lower payment costs net of adoption and usage costs; nonadopters benefit from lower retail prices charged by some merchants.
- Ramsey social planner vs. profit-maximizing monopoly network (comparative findings from solved calibration):
  - Within-period Ramsey problem: choose card fees (f_c,t, f_m,t) and R&D R_t to maximize present value of consumer welfare subject to adoption incentive constraints and network balanced-budget constraint; constraint (37) (revenue covering operation and R&D) is kept binding to minimize distortion.
  - Dynamic Ramsey problem: value function V(d_t; E(I_t), K_mu,t, K_chi,t) maximizes current-period consumer welfare U(d_t, R_t; ...) plus discounted continuation value with d_{t+1} = Γ(d_t, R_t).
  - Comparative outcomes (Ramsey social planner vs. monopoly network):
    - Both merchant and consumer fees are lower under the Ramsey social planner; under Ramsey both merchant and consumer fees decrease over time (monopoly leads to an increasing merchant fee over time).
    - Ramsey induces higher card adoption and higher card usage than monopoly.
    - Share of large merchants is higher and increasing under Ramsey, implying greater benefits to cash users under Ramsey than under monopoly.
    - Ramsey conducts more R&D, yielding a faster decline in card service cost δ.
    - Ramsey charges a decreasing markup over time; monopoly charges an increasing markup over time.
    - Consumer welfare in each period increases faster under Ramsey than under monopoly.
  - On social welfare accounting:
    - Under Ramsey (zero profit), consumer welfare equals social welfare.
    - Under monopoly, social welfare can be assessed by assuming monopoly rebates profits to consumers proportionally to income; with such profit rebates the mean consumer income increases (distribution remains exponential) and adoption thresholds remain unchanged because adoption costs are proportional to mean income.
    - Calibrated comparison: in early years social welfare (with profit rebates) may be lower under Ramsey because Ramsey invests more in R&D; however, social welfare under Ramsey grows faster and surpasses monopoly social welfare in a few years, with the gap widening over time.

*Source: IMF Working Paper — "4.3  Calibration Results" (content unit: wpiea2019057).*

### 5.3  Policy Experiments

### 5.3  Policy Experiments

### Regulatory approaches analyzed
- Two regulatory approaches studied: marginal-cost pricing regulation and merchant fee cap regulation.
- Marginal-cost pricing regulation: requires the card network to set card fees (φμ,t, φχ,t) to maximize consumer welfare subject to the zero markup constraint φμ,t + φχ,t = δt.
- Merchant fee cap regulation: caps the merchant (interchange) fee at a specified level (examples simulated include Cap 0.5%, Cap 1%, Cap 2%). Easier to implement because it regulates only the fee on the merchant side.

### Calibration and simulation assumptions
- Each regulation is simulated by assuming the regulation is implemented at the beginning year of the sample period.
- For the marginal-cost pricing regulation the network maximizes consumer welfare given the constraint φμ,t + φχ,t = δt.
- Under marginal-cost pricing, the regulation deprives the card network of R&D incentive and resources, so the card service cost δt stays at the initial level δ0.
- Footnote explanation: the monopoly network’s profit function is π(δ;λ)=1/λ(α0−α1 δ). The merchant fee cap reduces the value of α0 but not the value of α1 (and even increases it slightly); α1 determines the rate of cost decline at the balanced growth path.

### Key simulation outcomes — marginal-cost pricing regulation
- Compared with the unregulated monopoly case:
  - The regulated network lowers card fees to both merchants and consumers.
  - Card adoption by both consumers and merchants increases.
  - The fraction of large merchants who serve both card and cash users increases substantially.
  - Consumer welfare in each period rises relative to the unregulated monopoly.
- Dynamic effect on R&D and social welfare:
  - Regulation removes profit for R&D so δt remains at δ0 (no cost decline).
  - Figure 12 shows that, comparing with the unregulated monopoly and the Ramsey social planner, marginal-cost pricing yields a lower level of social welfare in each period except for the initial few years.
- Present-value comparison:
  - Marginal-cost pricing maximizes consumer welfare in a static setting, yielding a higher present value of consumer welfare than the unregulated monopoly.
  - However, because it leaves no profit for the card network to conduct R&D, the present value of social welfare under marginal-cost pricing is lower than under the unregulated monopoly.

### Key simulation outcomes — merchant fee cap regulation (Cap 1% example)
- With a binding merchant fee cap at 1%:
  - The card network raises the consumer fee to make up lost revenue.
  - Consumer card adoption is lower compared with the unregulated monopoly.
  - Merchant adoption (including large merchants serving both card and cash users) is higher compared with the unregulated monopoly.
  - Card transaction share does not change much.
  - Markup (fm + fc − d) is lower, so network profit is reduced.
  - Network R&D spending is constrained in the first few years, leading to a slower decline in card service costs until the late stage of the sample period.
- Welfare dynamics:
  - Consumer welfare in each period is higher under the merchant fee cap regulation compared with the unregulated monopoly case.
  - Social welfare gets lower in the early years before it turns higher in the longer run.
- Present-value comparison (Figure 13 context):
  - Merchant fee cap regulation redistributes between network profit and consumer welfare without hurting network R&D much.
  - The lower the cap, the higher the consumer welfare.
  - Social welfare changes little compared with the unregulated monopoly and slightly increases in the cap value.

### Scenarios and comparative metrics reported
- Scenarios explicitly compared: Monopoly, Monopoly with Cap 0.5%, Cap 1%, Cap 2%, Marginal Cost Pricing, Social Planner (Ramsey social planner).
- Present values of consumer welfare and network profits are computed with the present value of the cash economy being normalized to 100.
- Figures referenced:
  - Figure 10: Policy Experiments (time series of fees, adoption, transaction share by scenario).
  - Figure 11: Policy Experiments (continued — shares of small and large merchants, R&D to income ratio R (%), Card Cost: d (%), Card Markup: fm+fc−d (%), Consumer Welfare time series).
  - Figure 12: Comparing Social Welfare of Policy Experiments (time series; shows marginal-cost pricing below monopoly and social planner except initial years).
  - Figure 13: Comparing Present Values of Consumer and Social Welfare (consumer welfare and network profits PVs across scenarios; cash economy normalized to 100).

### Policy implications and trade-offs
- Marginal-cost pricing:
  - Strength: maximizes consumer welfare in a static framework.
  - Weakness: eliminates network profits and thus R&D, potentially reducing social welfare in a dynamic setting.
- Merchant fee cap:
  - Strength: easier to implement, improves consumer welfare and does not substantially impair R&D dynamics; redistributes surplus from network profits to consumers.
  - Trade-off: lowering the cap raises consumer welfare but reduces network profit; social welfare changes little and can slightly increase with higher cap values.
- Ramsey social planner benchmark:
  - Would set lower usage fees and conduct more R&D than the monopoly, achieving higher adoption and usage and higher levels of consumer and social welfare.

*Source: wpiea2019057 - 5.3  Policy Experiments (IMF working paper chapter).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019057.pdf_