## wpiea2025025-print-pdf

## Source details

**Canonical URL:** [wpiea2025025-print-pdf](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025025-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025025-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025025-print-pdf.pdf.json)

---

### Full Information (section 3.1)
- Model setup and objective
  - Two-period partial equilibrium investment model with exogenous aggregate demand shocks.
  - Firm maximizes d_0 + 1/(1+r) d_1 and chooses next period capital k and debt b in period 0; k_0 and b_0 at steady state and taken as given.
  - Production: DRS production function, productivity normalized to one.
  - Net worth:
    - n_0 = p k_0^θ + q(1−δ) k_0 − b_0
    - n_1 = p k^θ + q(1−δ) k − b
    - θ < 1. Steady state p̄ = q̄ = 1. q changes one-to-one with aggregate shock; p has elasticity ξ with respect to aggregate shock.
  - Firm takes q and p as given when deciding at t = 0, after shock realization.

- Financing and frictions
  - Borrowing constraint: b ≤ b̃(q), borrowing limit depends on aggregate shock.
  - Debt subsidized with rate R < 1 + r; debt preferred to equity until limit reached.
  - Equity payout deviation cost: φ_d/2 (d_0 − d̄)^2; φ_d governs degree of financial friction.
  - Constrained firm maximization (full information):
    - max_k π(k,q) = d_0 + 1/(1+r) d_1
    - subject to d_0 + φ_d/2 (d_0 − d̄)^2 = n_0 − q k + 1/R b̃(q), d_1 = n_1.
  - Unconstrained (Modigliani-Miller) case: d_0 = n_0 − q k and d_1 = n_1.

- Log-linearized solution under full information
  - Define π̂(k̂, q̂) = π(k̄_e k̂, q̄_e q̂) with k̂, q̂ log-deviations from steady state.
  - Second-order Taylor expansion yields:
    - k̂⋄ = (π̂_12)/|π̂_11| q̂
  - Where π̂_11 = π_11 k̄^2 and π̂_12 = π_12 k̄ q̄.
  - Interpretation: π̂_12/|π̂_11| is the pass-through of shock q̂ to optimal capital k̂⋄ under full information.

### Rational Inattention (section 3.2)
- Inattentiveness framework
  - Shock realizes at beginning of period 0 but firm may not fully observe it; firm chooses signals s about q and pays information cost measured by mutual information I(q;s).
  - Firm optimization with inattentiveness:
    - max_{s ∈ Γ} E[ π(k*, q) − μ I(q;s) | s^{−1} ]
    - k* = arg max_k E[ π(k,q) | s ]
    - μ is marginal cost of information; Γ set of available signals.
  - Signal assumption (Máckowiak and Wiederholt (2009) style):
    - ŝ = q̂ + ε, with q̂ ∼ N(0, σ_q^2), ε ∼ N(0, σ_ε^2), independent.

- Log-linearized decision under attention
  - Using second-order expansion:
    - k̂* = E(k̂⋄ | s) = (π̂_12)/|π̂_11| E(q̂ | s)
    - With Gaussian signals, k̂* = (π̂_12)/|π̂_11| (σ_q^2/(σ_q^2 + σ_ε^2)) ŝ
  - Information flow κ = I(q;s) yields signal-to-noise ratio:
    - σ_q^2 / σ_ε^2 = 2^{2κ} − 1
  - Equivalent minimization for κ:
    - min_κ |π̂_11|^2 [ (π̂_12 / |π̂_11|)^2 2^{−2κ} ] + μ κ
  - First order condition gives optimal κ*:
    - κ* = 1/2 log( (σ_q^2 / (μ / log 2)) / |π̂_11| ( (π̂_12 / |π̂_11|)^2 ) )
  - Comparative statics: higher σ_q^2 and lower μ increase κ*; curvature |π̂_11| and pass-through π̂_12/|π̂_11| also affect κ*.

- Role of financial friction on curvature and pass-through
  - Curvature:
    - |π̂_11| = φ_d k̄^2 + 1/(1+r) θ(1−θ) k̄^θ
    - With k̄ = (θ/(r+δ))^{1/(1−θ)}, constrained firms (φ_d > 0) have higher curvature than unconstrained (φ_d = 0).
    - Higher curvature → higher profit loss from imperfect actions → greater incentive to acquire information.
  - Pass-through for constrained firms (expression as presented):
    - (π̂_12)/|π̂_11| = [ φ_d ( ξ k̄^{θ−δ} + 1/R ∂ b̃/∂q ) ]^{−1} + 1/(1+r) [ θ ξ k̄^{θ−1} + 1−δ ] / [ φ_d k̄ + 1/(1+r) θ(1−θ) k̄^{θ−1} ]
  - Unconstrained pass-through:
    - (π̂_12)/|π̂_11| = (ξ − 1)/(1 − θ)
  - Implication: if ξ > 1 unconstrained firms increase capital following expansionary shock; constrained firms’ pass-through can be negative if ∂ b̃/∂q < 0 is sufficiently negative and internal funds increase insufficiently.

- Implications for attentiveness and proposition
  - Optimal κ* depends on |π̂_11| and (π̂_12/|π̂_11|)^2 but not on the sign of pass-through.
  - Constrained firms will be more attentive (higher κ*) than unconstrained firms if curvature and pass-through effects make |π̂_11| (π̂_12/|π̂_11|)^2 larger for constrained firms.
  - Proposition 1: Financially constrained firms with φ_d > φ̄_d exhibit higher attentiveness compared to unconstrained firms. Threshold φ̄_d depends on net change in firm’s access to funds ( ξ k̄^{θ−δ} + 1/R ∂ b̃/∂q ) following a shock.

### Key quantitative expressions (from sections 3.1–3.2)
- Objective: d_0 + 1/(1+r) d_1
- Debt subsidy: R < 1 + r
- Equity payout deviation cost: φ_d/2 (d_0 − d̄)^2
- Pass-through under full information: k̂⋄ = (π̂_12)/|π̂_11| q̂
- Signal-to-noise ratio: σ_q^2 / σ_ε^2 = 2^{2κ} − 1
- Optimal κ*:
  - κ* = 1/2 log( (σ_q^2 / (μ / log 2)) / |π̂_11| ( (π̂_12 / |π̂_11|)^2 ) )
- Curvature: |π̂_11| = φ_d k̄^2 + 1/(1+r) θ(1−θ) k̄^θ
- Unconstrained pass-through: (π̂_12)/|π̂_11| = (ξ − 1)/(1 − θ)
- Steady-state capital: k̄ = ( θ / (r + δ) )^{1/(1−θ)}

### Firm’s Size and Willingness to Pay for Professional Forecasts (section 4.3)
- Concept and hypotheses
  - Survey question: managers report willingness to pay ($/year) for professional forecasts of inflation, unemployment rate, and GDP.
  - Higher willingness to pay may indicate greater importance of variable for firm decisions (→ higher attentiveness and lower nowcast errors), but relationship is ambiguous due to costs and scale effects.
  - Size-related channels:
    - Smaller firms might value GDP growth more; larger firms might value unemployment more.
    - Economies of scale can make fixed costs of accurate information more justifiable for larger firms.

- Empirical evidence (summary of Table 6)
  - Dependent variables: log of willingness to pay ($/year) for forecasts of π, ue, GDP.
  - Key coefficients on log(employment):
    - Column (1) π: −0.000 (standard error 0.019)
    - Column (2) ue: −0.029 ∗∗ (standard error 0.013)
    - Column (3) GDP: 0.035 ∗∗∗ (standard error 0.011)
    - Column (4) π (with importance dummies): −0.017 ∗ (standard error 0.010)
    - Column (5) ue (with importance dummies): −0.026 ∗∗ (standard error 0.010)
    - Column (6) GDP (with importance dummies): −0.021 ∗∗ (standard error 0.010)
  - Importance-dummy coefficients (selected):
    - 1(π>ue): 0.383 ∗∗∗ (in π regression), −0.114 ∗∗∗ (in ue), −0.056 ∗∗ (in GDP)
    - 1(π>gdp): 0.417 ∗∗∗ (in π), 0.000 (in ue), −0.359 ∗∗∗ (in GDP)
    - 1(ue>gdp): −0.065 ∗∗∗ (in π), 0.228 ∗∗∗ (in ue), −0.084 ∗∗∗ (in GDP)
  - Estimation details:
    - Controls: log(age), log(#competitors), labor cost share, trade share, average margin; manager controls: tenure, education, income; sector fixed effects; sample weights; Huber robust regression; robust s.e. clustered at 3-digit ANZ SIC.
  - Goodness-of-fit (R2) by column: 0.578, 0.101, 0.110, 0.850, 0.429, 0.575.
  - Observations by column: 1082, 1112, 1094, 1108, 1111, 1105.

- Interpretation
  - Without importance controls (cols 1–3): negative coefficient for ue willingness-to-pay with firm size; positive for GDP; insignificant for π.
  - With importance rankings controlled (cols 4–6): coefficients on log(employment) reported as all negative and significant in text summary, interpreted as smaller firms more willing to pay once importance rankings controlled.
  - Importance dummies signs consistent with prediction that higher perceived importance → higher willingness to pay.

- Ambiguity and potential explanations
  - Higher willingness to pay could reflect greater attentiveness and lower nowcast errors, or reflect inability to pay/scale constraints limiting tracking despite importance.
  - Size-specific motives: smaller firms may value GDP more; larger firms may value unemployment more; economies of scale for fixed-cost information justify larger firms paying more.

### Appendix evidence (selected highlights relevant to attentiveness and firm size)
- Nowcast error of industry inflation (A.1, Table 9)
  - log(employment) coefficients in multiple columns: 0.276 ∗∗∗ (0.072); −0.044 ∗∗ (0.018); 0.085 ∗ (0.049); −0.048 ∗∗ (0.019).
  - Track dummy: −3.067 ∗∗∗ (0.058); −0.034 ∗∗ (0.015).
  - Observations reported: 10809, 5810, 65961 across columns. R2 reported: 0.6250.9310.8590.928.
- Number of products and nowcast error (A.2, Table 10)
  - log(employment): 0.237 ∗ (0.123); 0.132 ∗∗ (0.058); 0.068 (0.051).
  - log(#products): 0.047 (0.067); 0.007 (0.022); −0.061 ∗∗∗ (0.020).
  - Observations: 3833, 7237, 2387. R2: 0.6240.3700.336.
- Importance dummies and nowcast errors (A.3, Table 11)
  - log(employment) and importance dummy coefficients reported across inflation, unemployment, GDP regressions with observations and R2 listed.
- Forecast error and firm’s size (A.6, Table 14)
  - log(employment): 0.206 ∗∗ (0.094); 0.040 (0.028); −0.041 ∗ (0.024).
  - Observations: 1075, 1076, 1094. R2: 0.4440.1220.130.

### Parameterization (section 5.4)
- Parameters borrowed from literature (values as calibrated)
  - Cobb-Douglas capital share: θ = 0.36
  - Quarterly depreciation rate: δ = 0.025
  - Elasticity of substitution / demand elasticity: ε = 8.3 → steady state price markups of 13.7%
  - Equity adjustment cost parameter: φd = 0.4
  - Tax advantage: τ = 0.35
  - Aggregate capital adjustment cost parameter: φ = 4
  - Tightness of earning-based constraint parameter: θπ = 18.4
  - Discount factor: β = 0.995 → 2% annual real interest rate

- Calibrated parameters and empirical targets
  - Utility/disutility of labor parameter: ψ = 1.35 to obtain steady state labor L = 0.3
  - Share of constrained (junk-rated) firms: ω = 0.26 (observation period 1990−2008 from Compustat)
    - ω set by solving ω k_const.(1−ω) k_uncons. = (1/T) ∑_t (∑_i k_JR_i,t / ∑_i k_IG_i,t)
  - Information processing cost (per-period marginal cost of attention): μ = 0.0095 — calibrated by minimizing quadratic distance between model IRFs and empirical IRFs of capital for constrained and unconstrained firms using criterion (21).
  - Shock persistence (monetary shock): ρm = 0.9 from regression ∆(pce_t) = ρm ∆(pce_{t−1}) + εm
  - Standard deviation of monetary shock: σm = 0.006 (s.d. of εm)

- Optimal information capacities (calibrated)
  - Constrained firms: κ* = 1.1
  - Unconstrained firms: κ* = 0.97

- Nowcast errors of inflation (model)
  - Constrained firms: E|π_t − B_i(π_t)| = 0.60%
  - Unconstrained firms: E|π_t − B_i(π_t)| = 0.77%
  - Implied reduction in nowcast error due to financial friction: 0.17%
  - Empirical reference: column five of table (7) reports difference between nowcast error of large and small firms ≈ 0.26%, noted as comparable to model estimates.

- Calibration procedures and empirical normalization
  - μ calibrated by minimizing [Ψ̂ − Ψ(μ)]' V^−1 [Ψ̂ − Ψ(μ)] as in equation (21).
  - Monetary policy calibration uses nominal personal consumption expenditures (pce) data 1990–2008; monetary shock normalized so one standard deviation expansionary shock (27 basis points change in federal fund rate (APR)) is normalized to equal 1.

- Key model predictions from parameterization
  - Inattentiveness and financial frictions interact: constrained firms calibrated to be more attentive (κ* = 1.1 > 0.97).
  - Financial frictions reduce nowcast error in model by 0.17% (0.77% → 0.60%).
  - Heterogeneous firm responses to expansionary monetary policy:
    - Unconstrained firms increase capital following expansionary monetary shock.
    - Constrained firms contract due to tighter borrowing constraints and costly equity issuance.
  - Counterfactuals:
    - Raising μ from 0.0095 to 0.019 makes firm responses more sluggish and dampened for first ten quarters; at longer horizons indirect general-equilibrium effects can raise aggregate investment as demand peaks higher.
    - Increasing ω from 0.26 to 0.5 leads to more muted aggregate investment response: more firms contract, unconstrained firms increase investment but aggregate investment dampened.

- Empirical link
  - A one standard deviation increase in aggregate information rigidity λ_t (from Coibion and Gorodnichenko (2015)) dampens the capital response to a one s.d. expansionary monetary shock by about 0.5%.

### Policy-relevant implications (drawn from model and calibration)
- Lower volatility of monetary shocks predicted to reduce aggregate attentiveness (trend referenced to Bauer and Swanson (2023)), increasing sluggishness in investment responses.
- Periods of heightened financial constraints (higher ω):
  - Monetary policy transmission via investment channel is less effective because larger share of firms contract despite expansionary policy.
  - Targeted measures to alleviate financial constraints (improve credit access, reduce equity issuance costs) could enhance monetary policy effectiveness by enabling constrained firms to respond more to shocks.
- Importance of accounting for both firms’ financial conditions and information-processing capacities when evaluating or designing monetary policy interventions.

### Appendix highlights (methods and definitions relevant to replication)
- Mutual information and entropy definitions for Gaussian processes used to quantify information (section D).
  - Entropy H(X) = 1/2 log_2[(2πe)^N det Ω_X].
  - Mutual information I(X;Y) = H(X) − H(X|Y).
  - For stationary Gaussian processes I({X_t};{Y_t}) = lim_{T→∞} 1/T [ 1/2 log_2( det Ω_X / det Ω_{X|Y} ) ].
- Firm-level sample construction for panel local projections (Appendix B)
  - Data sources: Compustat and CRSP. Sample period: 1990Q1 to 2007Q4 (financial crisis observations omitted).
  - Exclusions: non-US incorporated firms; utilities (SIC 4900–4999), finance/insurance/real estate (SIC 6000–6999), public administration (SIC 9100–9999); observations with non-positive or negative values for key accounting series as listed.
  - Firms with less than 40 observations excluded unless otherwise stated.
- Dependent variable construction and control variable definitions (B.1.1, B.1.2) including fixed capital via perpetual inventory method and leverage, liquid asset ratio, Tobin’s q construction.
- S&P long-term issuer rating group definitions: Junk-Rated = BB+ or lower for at least a month in a quarter; Investment-Grade = A- or above without BB+ or lower in same quarter; BBB-only quarters excluded to reduce misclassification.

*Source: IMF working paper chapter and appendices (wpiea2025025-print-pdf).*

### 3.1    Full Information

### 3.1    Full Information

### Model setup and objective
- Two-period partial equilibrium investment model with exogenous aggregate demand shocks.
- Firm maximizes d_0 + 1/(1+r) d_1.
- Firm chooses next period capital k and debt b in period 0. k_0 and b_0 are at steady state and taken as given.
- Production: DRS production function, productivity normalized to one.
- Net worth:
  - n_0 = p k_0^θ + q(1−δ) k_0 − b_0
  - n_1 = p k^θ + q(1−δ) k − b
  - θ < 1. p and q are relative prices of output and capital with steady state levels p̄ = q̄ = 1.
- p and q change exogenously following aggregate shocks; q changes one-to-one with the aggregate shock. p has elasticity ξ with respect to the aggregate shock. q is treated as exogenous; p reacts to changes in q. Firm takes q and p as given when deciding at t = 0, after shock realization.

### Financing and frictions
- Firms use internal and external funds. Borrowing constraint: b ≤ b̃(q), borrowing limit depends on aggregate shock.
- Debt subsidized with rate R < 1 + r, so debt preferred to equity and used to full capacity before issuing equity.
- Equity payout target at t = 0 denoted d̄; deviation cost φ_d/2 (d_0 − d̄)^2. φ_d governs degree of financial friction; higher φ_d implies costlier equity issuance/share repurchases.
- Constrained firm maximization under full information:
  - max_k π(k,q) = d_0 + 1/(1+r) d_1
  - subject to d_0 + φ_d/2 (d_0 − d̄)^2 = n_0 − q k + 1/R b̃(q), d_1 = n_1
- Unconstrained (Modigliani-Miller) case: d_0 = n_0 − q k and d_1 = n_1.

### Log-linearized solution under full information
- Define π̂(k̂, q̂) = π(k̄_e k̂, q̄_e q̂) with k̂, q̂ log-deviations from steady state.
- Second-order Taylor expansion yields derivative condition:
  - k̂⋄ = (π̂_12)/|π̂_11| q̂
- Where π̂_11 = π_11 k̄^2 and π̂_12 = π_12 k̄ q̄.
- π̂_12/|π̂_11| is the pass-through of shock q̂ to optimal capital k̂⋄ under full information.

---

### 3.2    Rational Inattention

### Inattentiveness framework
- Shock realizes at beginning of period 0 but firm does not fully observe it. Firm forms posterior beliefs by choosing signals s that carry information about q; processing information is costly.
- Information cost measured by mutual information function I(q;s).
- Firm optimization with inattentiveness:
  - max_{s ∈ Γ} E[ π(k*, q) − μ I(q;s) | s^{−1} ]
  - k* = arg max_k E[ π(k,q) | s ]
  - μ is marginal cost of information; Γ is set of available signals.
- Assumptions on Γ (as in Máckowiak and Wiederholt (2009)):
  - ŝ = q̂ + ε, with q̂ ∼ N(0, σ_q^2), ε ∼ N(0, σ_ε^2), independent.

### Log-linearized decision under attention
- Using second-order expansion:
  - k̂* = E(k̂⋄ | s) = (π̂_12)/|π̂_11| E(q̂ | s)
  - With Gaussian signals, k̂* = (π̂_12)/|π̂_11| (σ_q^2/(σ_q^2 + σ_ε^2)) ŝ
- Information flow κ = I(q;s) yields signal-to-noise ratio:
  - σ_q^2 / σ_ε^2 = 2^{2κ} − 1
- Equivalent minimization for κ (trade-off between loss from suboptimal capital and information cost):
  - min_κ |π̂_11|^2 [ (π̂_12 / |π̂_11|)^2 2^{−2κ} ] + μ κ
  - First order condition gives optimal κ*:
    - κ* = 1/2 log( (σ_q^2 / (μ / log 2)) / |π̂_11| ( (π̂_12 / |π̂_11|)^2 ) )
    - Interpreted: higher shock volatility σ_q^2 and lower cost μ increase κ*. Curvature |π̂_11| and pass-through π̂_12/|π̂_11| also affect κ*.

### Role of financial friction on curvature and pass-through
- Curvature:
  - |π̂_11| = φ_d k̄^2 + 1/(1+r) θ(1−θ) k̄^θ
  - With k̄ = (θ/(r+δ))^{1/(1−θ)}, constrained firms (φ_d > 0) have higher curvature around steady state than unconstrained (φ_d = 0).
  - Higher curvature → higher profit loss from imperfect actions → greater incentive to acquire information.
- Pass-through for constrained firms:
  - (π̂_12)/|π̂_11| = [ φ_d ( ξ k̄^{θ−δ} + 1/R ∂ b̃/∂q ) ]^{−1} + 1/(1+r) [ θ ξ k̄^{θ−1} + 1−δ ] / [ φ_d k̄ + 1/(1+r) θ(1−θ) k̄^{θ−1} ]
  - For unconstrained firms (φ_d = 0):
    - (π̂_12)/|π̂_11| = (ξ − 1)/(1 − θ)
  - If ξ > 1, unconstrained firms increase capital following expansionary shock. For constrained firms, pass-through sign can be negative if borrowing constraint tightens (∂ b̃/∂q < 0) sufficiently and internal funds increase is insufficient.

### Implications for attentiveness and Proposition
- Optimal information κ* does not depend on sign of pass-through; depends on |π̂_11| (curvature) and (π̂_12/|π̂_11|)^2.
- Constrained firms will be more attentive (higher κ*) than unconstrained firms if curvature and pass-through effects make |π̂_11| (π̂_12/|π̂_11|)^2 larger for constrained firms.
- Proposition 1: Financially constrained firms with φ_d > φ̄_d exhibit higher attentiveness compared to unconstrained firms. Threshold φ̄_d depends on net change in firm’s access to funds ( ξ k̄^{θ−δ} + 1/R ∂ b̃/∂q ) following a shock. (Proof in Appendix F.1.)
- Intuition: Variation in internal funds plus external borrowing determines response size; larger variation requires larger pass-through and greater attentiveness. Even if financial friction damps sensitivity, sufficiently high friction increases curvature dominance and raises attentiveness.

### Extension and empirical motivation
- Assumption of exogenous ξ may be restrictive; sales for constrained and unconstrained firms could react differently to shocks.
- Section (5) extends the two-period model to a fully dynamic general equilibrium framework where firms set prices, hire employees, and adjust borrowing constraints.
- Before the dynamic model, the paper leverages firm expectation surveys for New Zealand and the United States to provide evidence on how financial conditions influence incentives to acquire information.

---

### Key quantitative expressions and parameters (as presented)
- Objective: d_0 + 1/(1+r) d_1
- Debt subsidy: R < 1 + r
- Equity payout deviation cost: φ_d/2 (d_0 − d̄)^2
- Pass-through under full information: k̂⋄ = (π̂_12)/|π̂_11| q̂
- Signal-to-noise ratio: σ_q^2 / σ_ε^2 = 2^{2κ} − 1
- Optimal κ*:
  - κ* = 1/2 log( (σ_q^2 / (μ / log 2)) / |π̂_11| ( (π̂_12 / |π̂_11|)^2 ) )
- Curvature: |π̂_11| = φ_d k̄^2 + 1/(1+r) θ(1−θ) k̄^θ
- Unconstrained pass-through: (π̂_12)/|π̂_11| = (ξ − 1)/(1 − θ)
- Steady-state capital: k̄ = ( θ / (r + δ) )^{1/(1−θ)}

*Source: IMF working paper chapter "3.1 Full Information" (wpiea2025025-print-pdf).*

### 4.3    Firm’s Size and Willingness to Pay for Professional Forecasts

### 4.3    Firm’s Size and Willingness to Pay for Professional Forecasts

### Concept and hypotheses
- Survey question: managers report how much they are willing to pay ($/year) for professional forecasts of inflation, unemployment rate, and GDP.
- Interpretation:
  - Higher willingness to pay for a professional forecast of a variable can indicate that the variable is more important or relevant in a firm’s decision-making, which should raise incentives to be attentive and reduce nowcast errors.
  - Relationship is ambiguous: a firm may rate a variable important but not track it closely because of costs of attentiveness; smaller firms might pay less simply because of resource constraints, or they may pay more if the variable is particularly critical to them.
  - Size-related channels considered:
    - Smaller firms might value GDP growth more; larger firms might value labor market conditions (unemployment) more.
    - Larger firms can benefit from economies of scale, making a fixed cost for accurate information more justifiable.

### Empirical evidence (summary of Table 6: Firms Size and Willingness to Pay for Professional Forecast)
- Dependent variables: log of willingness to pay ($/year) for a professional forecast for variable X (π, ue, GDP).
- Key coefficient estimates on log(employment):
  - Column (1) π: −0.000 (standard error 0.019)
  - Column (2) ue: −0.029 ∗∗ (standard error 0.013)
  - Column (3) GDP: 0.035 ∗∗∗ (standard error 0.011)
  - Column (4) π (with importance dummies): −0.017 ∗ (standard error 0.010)
  - Column (5) ue (with importance dummies): −0.026 ∗∗ (standard error 0.010)
  - Column (6) GDP (with importance dummies): −0.021 ∗∗ (standard error 0.010)
- Importance-dummy coefficients (selected, from columns that include them):
  - 1(π>ue): 0.383 ∗∗∗ (in π regression), −0.114 ∗∗∗ (in ue), −0.056 ∗∗ (in GDP)
  - 1(π>gdp): 0.417 ∗∗∗ (in π), 0.000 (in ue), −0.359 ∗∗∗ (in GDP)
  - 1(ue>gdp): −0.065 ∗∗∗ (in π), 0.228 ∗∗∗ (in ue), −0.084 ∗∗∗ (in GDP)
- Model controls and estimation details:
  - Firm controls: log(age), log(#competitors), labor cost share, trade share, and average margin.
  - Manager controls: tenure, education, and income.
  - Sector fixed effects included.
  - Sample weights applied to all specifications.
  - Huber robust regression; robust standard errors clustered at the 3-digit ANZ SIC level reported in parentheses.
  - Significance notation: ***, **, * denotes statistical significance at 1%, 5%, and 10% levels respectively.
- Goodness-of-fit (R2) reported by column:
  - Columns (1)–(6) R2: 0.578, 0.101, 0.110, 0.850, 0.429, 0.575 respectively.
- Observations by column:
  - Columns (1)–(6) Observations: 1082, 1112, 1094, 1108, 1111, 1105 respectively.

### Interpretation of empirical results
- Without importance controls (columns 1–3), the author reports:
  - A negative coefficient for unemployment (ue) willingness to pay with firm size.
  - A positive coefficient for GDP willingness to pay with firm size.
  - An insignificant coefficient for inflation (π).
- After controlling for importance rankings (columns 4–6):
  - Coefficients on log(employment) are reported as all negative and significant in the text summary, suggesting smaller firms are more willing to pay for a professional forecast once importance rankings are controlled for.
  - Signs of importance dummies are consistent with predictions: if a firm finds a variable more important for business decisions, it is willing to pay more for its professional forecast.

### Ambiguity and potential explanations
- Ambiguity in the relationship between willingness to pay and nowcast error:
  - Higher willingness to pay could indicate greater attentiveness and thus smaller nowcast errors.
  - Alternatively, resource constraints and costs of being attentive may prevent firms from tracking even important variables.
- Size-specific motives:
  - Smaller firms may place higher willingness-to-pay on some aggregate variables (e.g., GDP) if those variables more directly affect their outcomes.
  - Larger firms may pay relatively more to track labor-market variables (unemployment) because labor market conditions are crucial to their profitability.
  - Economies of scale may make it rational for larger firms to pay fixed costs for accurate information.

### Estimation notes (relevant to interpretation)
- Results use Huber robust regression.
- Sample weights and sector fixed effects applied.
- Robust standard errors clustered at the 3-digit ANZ SIC level are reported.
- Importance dummies are constructed from the survey ranking question: “Which macroeconomic variables are most important to you in making your business decisions (rank them)? a. Unemployment rate b. GDP c. Inflation d. None”.
- The willingness-to-pay regressions control for the same baseline covariates used in prior analyses (table (4) baseline model).

*Italic: Source: wpiea2025025-print-pdf — 4.3    Firm’s Size and Willingness to Pay for Professional Forecasts*

### 5.4    Parameterization

### 5.4    Parameterization

### Parameters borrowed from the literature
- Cobb-Douglas capital share: θ = 0.36 (Jermann and Quadrini (2012))
- Quarterly depreciation rate: δ = 0.025 (Jermann and Quadrini (2012))
- Elasticity of substitution / demand elasticity: ε = 8.3 (Jermann and Quadrini (2012)) resulting in steady state price markups of 13.7%
- Equity adjustment cost parameter: φd = 0.4 (Jermann and Quadrini (2012))
- Tax advantage: τ = 0.35 (Jermann and Quadrini (2012))
- Aggregate capital adjustment cost parameter: φ = 4 (Bernanke, Gertler and Gilchrist (1999))
- Tightness of earning-based constraint parameter: θπ = 18.4 (Drechsel (2022))
- Discount factor: β = 0.995 resulting in a 2% annual real interest rate

### Calibrated parameters and empirical targets
- Utility / disutility of labor parameter: ψ = 1.35 to obtain steady state labor L = 0.3
- Share of constrained (junk-rated) firms in dataset: ω = 0.26 (observation period 1990−2008 from Compustat)
  - ω is set by solving ω k_const.(1−ω) k_uncons. = (1/T) ∑_t (∑_i k_JR_i,t / ∑_i k_IG_i,t)
- Optimal information / inattentiveness calibration targets and shocks:
  - Information processing cost (per-period marginal cost of attention): μ = 0.0095 — calibrated by minimizing the distance between model IRFs and empirical IRFs of capital for constrained and unconstrained firms using criterion (21): min_μ [Ψ̂ − Ψ(μ)]' V^−1 [Ψ̂ − Ψ(μ)]
  - Shock persistence (monetary shock): ρm = 0.9 from regression ∆(pce_t) = ρm ∆(pce_{t−1}) + εm
  - Standard deviation of monetary shock: σm = 0.006 (s.d. of εm)
- Optimal information capacities:
  - Constrained firms: κ* = 1.1
  - Unconstrained firms: κ* = 0.97
- Nowcast errors of inflation (model):
  - Constrained firms: E|π_t − B_i(π_t)| = 0.60%
  - Unconstrained firms: E|π_t − B_i(π_t)| = 0.77%
  - Implied reduction in nowcast error due to financial friction: 0.17%
- Empirical reference: column five of table (7) reports the difference between nowcast error of large and small firms ≈ 0.26%, noted as comparable to model estimates

### Calibration procedures and empirical normalization
- μ calibrated by minimizing quadratic distance between empirical IRFs (Ψ̂) and model IRFs Ψ(μ), weighting by V (diagonal matrix of empirical IRF variances) as in equation (21)
- Monetary policy calibration: use nominal personal consumption expenditures (pce) data 1990–2008 and estimate ∆(pce_t) = ρm ∆(pce_{t−1}) + εm to obtain ρm and σm
- Monetary shock normalization in empirical regressions: one standard deviation expansionary shock (27 basis points change in federal fund rate (APR)) is normalized to equal 1

### Key model predictions (summary of implications from parameterization)
- Inattentiveness and financial frictions interact:
  - Constrained firms are calibrated to be more attentive (higher κ*) than unconstrained firms because they face higher costs from suboptimal decisions.
  - Financial frictions reduce nowcast error in the model by 0.17% (0.77% → 0.60%).
- Heterogeneous firm responses to expansionary monetary policy (as implied by calibrated parameters and subsequent IRFs):
  - Unconstrained firms increase capital following an expansionary monetary shock due to access to cheaper funds.
  - Constrained firms contract because tighter borrowing constraints and costly equity issuance make raising funds infeasible/expensive.
- Aggregate implications of varying μ and ω (reported counterfactuals):
  - Increasing the cost of information (μ from 0.0095 to 0.019) makes firm responses more sluggish and dampened for the first ten quarters; for longer horizons indirect general-equilibrium effects can dominate, raising aggregate investment as demand peaks higher.
  - Increasing the share of constrained firms (ω from 0.26 to 0.5) leads to a more muted aggregate investment response: more firms contract, unconstrained firms increase investment but aggregate investment is dampened due to the larger mass of contracting constrained firms and higher attentiveness in the economy.
- Empirical local-projection evidence linked to calibrated model:
  - A one standard deviation increase in aggregate information rigidity λ_t (from Coibion and Gorodnichenko (2015)) dampens the capital response to a one s.d. expansionary monetary shock by about 0.5%

### Policy-relevant implications drawn from calibration and model behavior
- Lower volatility of monetary shocks is predicted to reduce aggregate attentiveness (referenced trend in Bauer and Swanson (2023)), which increases sluggishness in investment responses to shocks.
- During periods of heightened financial constraints (higher ω):
  - Monetary policy transmission via the investment channel is less effective because a larger share of firms contracts despite expansionary policy.
  - Targeted measures to alleviate financial constraints (improve access to credit, reduce equity issuance costs) could enhance monetary policy effectiveness by enabling constrained firms to respond more to shocks.
- Considering both firms’ financial conditions and information-processing capacities is important when evaluating or designing monetary policy interventions.

*Source: Section 5.4 (Parameterization) of the supplied IMF working paper content unit.*

### References

### References

### Key citations
- Afrouzi, Hassan.2016. “Strategic Inattention, Inflation Dynamics, and the Non-Neutrality of Money.”
- Bauer, Michael D, and Eric T Swanson.2023. “An alternative explanation for the “fed information effect”.” American Economic Review, 113(3): 664–700.
- Bayarmagnai, Gerelmaa.2023. “Rational inattention to inflation among New Zealand households.” Reserve Bank of New Zealand.
- Bernanke, Ben S., Mark Gertler, and Simon Gilchrist.1999. “The financial accelerator in a quantitative business cycle framework.”
- Caglio, Cecilia, Matthew Darst, and Sebnem Kalemli-Özcan.2021. “Risk-Taking and Monetary Policy Transmission: Evidence from Loans to Smes and Large Firms.” SSRN Electronic Journal.
- Charoenwong, Ben, Yosuke Kimura, Alan Kwan, and Eugene Tan.2021. “Capital Budgeting, Uncertainty, and Misallocation.” Working Paper, 1–45.
- Christiano, Lawrence J, Martin Eichenbaum, and Charles L Evans.2005. “Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy.” Journal of Political Economy, 113(1): 1–45.
- Clementi, Gian Luca, and Berardino Palazzo.2019. “Investment and the Cross-Section of Equity Returns.” Journal of Finance, 74(1): 281–321.
- Coibion, Olivier, and Yuriy Gorodnichenko.2012. “What Can Survey Forecasts Tell Us about Information Rigidities?” Journal of Political Economy, 120(1): 116–159.
- Coibion, Olivier, and Yuriy Gorodnichenko.2015. “Information rigidity and the expectations formation process: A simple framework and new facts.” American Economic Review, 105(8): 2644–2678.
- Coibion, Olivier, Yuriy Gorodnichenko, and Saten Kumar.2018. “How do firms form their expectations? New survey evidence.” American Economic Review, 108(9): 2671–2713.
- Cover, Thomas M., and Joy A. Thomas.2006. Information Theory Elements of.
- Drechsel, Thomas.2022. “Earnings-Based Borrowing Constraints and Macroeconomic Fluctuations.”
- Duygan-Bump, Burcu, Alexey Levkov, and Judit Montoriol-Garriga.2015. “Financing constraints and unemployment: Evidence from the Great Recession.” Journal of Monetary Economics, 75: 89–105.
- Flynn, Joel P., and Karthik A. Sastry.2021. “Attention Cycles.”
- Gertler, Mark, and Simon Gilchrist.1994. “Monetary Policy, Business Cycles, and the Behavior of Small Manufacturing Firms.” 2.
- Golosov, Mikhail, and Robert E. Lucas Jr.2007. “Menu Costs and Phillips Curves.” Journal of Political Economy.
- Hadlock, Charles J, and Joshua R Pierce.2010. “New evidence on measuring financial constraints: Moving beyond the KZ index.” Review of Financial Studies, 23(5): 1909–1940.
- Hennessy, Christopher A., and Toni M. Whited.2005. “Debt Dynamics.” The Journal of Finance, 60(3): 1129–1165.
- Jeenas, Priit.2018. “Monetary Policy Shocks, Financial Structure, and Firm Activity: A Panel Approach.”
- Jermann, Urban, and Vincenzo Quadrini.2012. “Macroeconomic effects of financial shocks.”
- Jordà, Òscar.2005. “Estimation and inference of impulse responses by local projections.” American Economic Review, 95(1): 161–182.
- Kermani, Amir, and Yueran Ma.2021. “Two Tales of Debt.” SSRN Electronic Journal.
- Kiyotaki, Nobuhiro, and John Moore.1997. “Credit Cycles.” 2, The University of Chicago Press.
- Lian, Chen, and Yueran Ma.2021. “Anatomy of Corporate Borrowing Constraints.” Quarterly Journal of Economics, 136(1): 229–291.
- Máckowiak, Bartosz, and Mirko Wiederholt.2009. “Optimal sticky prices under rational inattention.” American Economic Review, 99(3): 769–803.
- Máckowiak, Bartosz, and Mirko Wiederholt.2015. “Real Business Cycle Dynamics under Rational Inattention.” 1(14089).
- Mankiw, N. Gregory, and Ricardo Reis.2002. “Sticky information versus sticky prices: A proposal to replace the new Keynesian Phillips curve.” Quarterly Journal of Economics, 117(4): 1295–1328.
- Mckay, Alisdair, and Johannes F Wieland.2021. “Lumpy Durable Consumption Demand and the Limited Ammunition of Monetary Policy.”
- Nakamura, Emi, and Jón Steinsson.2010. “Monetary non-neutrality in a multisector menu cost model.” Quarterly Journal of Economics, 125(3): 961–1013.
- Ottonello, Pablo, and Thomas Winberry.2018. “Financial Heterogeneity and the Investment Channel of Monetary Policy.” 1–72.
- Ottonello, Pablo, and Thomas Winberry.2020. “Financial Heterogeneity and the Investment Channel of Monetary Policy.” Econometrica, 88(6): 2473–2502.
- Rezghi, Abolfazl.2020. “Monetary Policy Shocks, Financial Frictions, and Investment.”
- Romer, Christina D, and David H Romer.2004. “A New Measure of Monetary Shocks: Derivation and Implications.” American Economic Review, 94(4): 1055–1084.
- Sims, Christopher A.2003. “Implications of rational inattention.” Journal of Monetary Economics, 50(3): 665–690.
- Song, Wenting, and Samuel Stern.2020. “Firm inattention and the efficacy of monetary policy: A text-based approach.” Available at SSRN 3900641.
- Tenreyro, Silvana, and Gregory Thwaites.2016. “Pushing on a string: Us monetary policy is less powerful in recessions.” American Economic Journal: Macroeconomics, 8(4): 43–74.
- Vavra, Joseph.2014. “Inflation Dynamics and Time-Varying Uncertainty.” The Quarterly Journal of Economics, 129(No. 1 (February 2014)): 215–258.
- Wang, Zijian.2022. “Rational Inattention , Financial Heterogeneity and Effectiveness of Monetary Policy.”
- Whited, Toni M, and Guojun Wu. n.d.. “Financial Constraints Risk.”
- Woodford, Michael.2001. “Imperfect Common Knowledge and the Effects of Monetary Policy.” NBER Working Paper Series.
- Yang, Choongryul.2020. “Rational Inattention, Menu Costs, and Multi-Product Firms: Micro Evidence and Aggregate Implications.”
- Zorn, Peter.2020. “Investment under Rational Inattention : Evidence from US Sectoral Data.”

*References list from the content unit.*

### Appendix A — Additional Survey Evidence

### A.1 Nowcast Error of Industry Inflation (Table 9)
- Regression: Huber robust regression; dependent variables are absolute firm errors about past 12-month headline and industry inflation from wave 4 survey.
- Key coefficients (standard errors in parentheses):
  - log(employment): 0.276 ∗∗∗ (0.072); −0.044 ∗∗ (0.018); 0.085 ∗ (0.049); −0.048 ∗∗ (0.019) across columns (1)–(4).
  - Track dummy: −3.067 ∗∗∗ (0.058); −0.034 ∗∗ (0.015).
- Controls: Firm Controls Yes; Manager Controls Yes; Sector FE Yes.
- Observations: 10809, 5810, 65961 (reported across columns).
- R2: 0.6250.9310.8590.928 (reported across columns).
- Notes: Sample weights applied. Robust standard errors clustered at the 3-digit ANZ SIC level. Track dummy equals one if manager tracks the headline inflation.

### A.2 Number of products and Nowcast Error (Table 10)
- Regression: Huber robust regression; dependent variables are absolute firm errors about past 12-month inflation, unemployment rate, and GDP growth rate from wave 4 survey. Number of products from wave 2 survey.
- Key coefficients (standard errors in parentheses):
  - log(employment): 0.237 ∗ (0.123); 0.132 ∗∗ (0.058); 0.068 (0.051).
  - log(#products): 0.047 (0.067); 0.007 (0.022); −0.061 ∗∗∗ (0.020).
- Controls: Firm Controls Yes; Manager Controls Yes/No (varies by column); Sector FE Yes.
- Observations: 3833, 7237, 2387.
- R2: 0.6240.3700.336.
- Notes: Sample weights applied. Robust standard errors clustered at the 3-digit ANZ SIC level.

### A.3 Importance of An Aggregate Variable (Table 11)
- Regression: Huber robust regression; dependent variables are absolute firm errors about past 12-month inflation, unemployment rate and GDP growth rate from wave 4 survey.
- Key coefficients (standard errors in parentheses):
  - log(employment): 0.276 ∗∗∗ (0.072); 0.078 ∗∗∗ (0.024); 0.016 (0.026); 0.127 ∗∗ (0.050); 0.053 ∗∗ (0.024); 0.032 (0.026).
  - 1(π>ue): −1.069 ∗∗∗; 1.013 ∗∗∗; 0.004.
  - 1(π>gdp): −2.004 ∗∗∗; −0.866 ∗∗∗; 0.213 ∗∗∗.
  - 1(ue>gdp): 0.025; −0.235 ∗∗∗; 0.104 ∗∗.
- Observations: 10801, 10421, 1099108210601095 (reported across columns).
- R2: 0.6250.1180.1640.8360.2220.484.
- Notes: Importance dummies constructed from survey ranking question. Sample weights applied. Robust standard errors clustered at the 3-digit ANZ SIC level.

### A.4 Nowcast Error of Interest Rate and Exchange Rate (Table 12)
- Regression: Huber robust regression; dependent variables are absolute firms’ nowcast errors about interest rate, output gap, and exchange rate.
- Key coefficients (standard errors in parentheses):
  - log(employment): 0.198 ∗∗∗ (0.058); 0.383 (0.429); 0.005 ∗∗ (0.002).
  - log(age): −0.182 ∗∗∗ (0.053); 0.030 (0.323); −0.002 (0.002).
- Observations: 1075, 1076, 1094.
- R2: 0.4440.1220.130.
- Notes: Interest rate and output gap from wave 2; exchange rate from wave 4. Sample weights applied. Robust standard errors clustered at the 3-digit ANZ SIC level.

### A.5 Firm Age and Nowcast Error (Table 13)
- Regression: Huber robust regression; dependent variables are absolute firm nowcast errors of past 12-month inflation, unemployment rate and GDP growth rate from wave 4 survey.
- Key coefficients (standard errors in parentheses):
  - log(employment): 0.276 ∗∗∗ (0.072); 0.078 ∗∗∗ (0.024); 0.016 (0.026).
  - log(age): 0.256 ∗∗∗ (0.064); −0.009 (0.019); −0.038 (0.024); 0.350 ∗∗∗ (0.070); 0.015 (0.020); −0.031 (0.024).
- Observations: 10801, 10421, 1099108210401099 (reported across columns).
- R2: 0.6250.1180.1640.6230.1050.166.
- Notes: Sample weights applied. Robust standard errors clustered at the 3-digit ANZ SIC level.

### A.6 Forecast Error and Firm’s Size (Table 14)
- Regression: Huber robust regression; dependent variables are absolute firms’ forecast errors about inflation, unemployment rate and GDP growth rate in the next 12 months from wave 4 survey. Professional forecasters’ forecast used as reference.
- Key coefficients (standard errors in parentheses):
  - log(employment): 0.206 ∗∗ (0.094); 0.040 (0.028); −0.041 ∗ (0.024).
  - log(age): 0.340 ∗∗∗ (0.087); −0.033 (0.026); 0.030 (0.026).
- Observations: 1075, 1076, 1094.
- R2: 0.4440.1220.130.
- Notes: Sample weights applied. Robust standard errors clustered at the 3-digit ANZ SIC level.

### Appendix B — Sample Construction for Panel Local Projections

### B.1 Firm Level Data
- Data sources: Compustat and CRSP.
- Sample period: 1990Q1 to 2007Q4.
  - Observations during the financial crisis omitted to focus on effects of conventional monetary policy.
  - Excluding data before 1990 allows robustness checks to high frequency identified monetary shocks available from 1990. Baseline result broadly robust to including observations from 1985-1990.
- Firms/observations excluded from the sample:
  1. Firms that are incorporated outside US.
  2. Firms in utilities (SIC codes 4900–4999), finance, insurance, and real estate (SIC codes 6000–6999), and public administration sector (SIC codes 9100–9999).
  3. Observations with non-positive values for ATQ_i,t, PPENTQ_i,t, PPEGTQ_i,t, SALEQ_i,t, INV TQ_i,t, PRCCQ_i,t, and CSHOQ_i,t.
  4. Observations with negative values for DLCQ_i,t, DLTTQ_i,t, and CHEQ_i,t.

*Appendix and references content from the content unit.*

### 5.  Firms that have less than 40 observations in the sample, unless otherwise stated.

### 5.  Firms that have less than 40 observations in the sample, unless otherwise stated.

### Data treatment and sample restrictions
- Firms with less than 40 observations in the sample are excluded, unless otherwise stated.
- When running regression model 22, 1 percent of outlier observations at both tails of distribution for dependent variable ∆_h_ log(y_i,t+h_) are dropped.

### B.1.1 Dependent variables (construction and definitions)
- Fixed Capital:
  - Constructed using the perpetual inventory method.
  - The first observation k_i,0 is set as the earliest available value for PPEGTQ_i,t (Property, Plant and Equipment (Gross)) in the dataset.
  - Iterative construction:
    - k_i,t = k_i,t−1 + PPENTQ_i,t − PPENTQ_i,t−1
    - k_i,t is firm i’s capital at the end of period t; PPENTQ_i,t is the net value of Property, Plant and Equipment.
  - PPENTQ_i,t and PPEGTQ_i,t are deflated using the implied price index of gross value added in the U.S. non-farm business sector (BEA-NIPA Table 1.3.4 Line 3).
- Debt, Inventory, and Sale:
  - Total debt = DLCQ_i,t + DLTTQ_i,t (sum of short and long term debt).
  - Inventory = INVTQ_i,t in Compustat.
  - Sale = SALEQ_i,t.

### B.1.2 Control variables (baseline regressions)
- Size: measured as firm’s total assets (ATQ_i,t).
- Sales growth: log difference of deflated value of sale (SALEQ_i,t) between period t and t−4.
- Cash flow ratio:
  - (IBQ_i,t + DPQ_i,t) / ATQ_i,t
  - IBQ_i,t = Income Before Extraordinary Items; DPQ_i,t = Total Depreciation and Amortization.
- Leverage:
  - (DLCQ_i,t + DLTTQ_i,t) / ATQ_i,t
  - Outliers are dropped at 99% cutoff for each quarter.
- Liquid asset ratio:
  - CHEQ_i,t / ATQ_i,t
  - Outliers are dropped at 99% cutoff for each quarter.
- Capital share: firm’s fixed capital divided by firm’s size.
- Tobin’s q:
  - Market to book value of assets measured as (ATQ_i,t + PRCCQ_i,t × CSHOQ_i,t − CEQQ_i,t + TXDITCQ_i,t) / ATQ_i,t following Ottonello and Winberry (2018).
  - Observations are winsorized at 1% of each tail of the distribution.

### B.2 S&P’s Long-Term Issuer Credit Rating (group definitions and treatment)
- Ratings availability: monthly frequency.
- Junk-Rated firm: firm with a rating equal to BB+ or lower for at least a month in a given quarter.
- Investment-Grade firm: firm with an A- or above rating condition on not having BB+ or lower rating in the same quarter.
- Firms with only BBB rating in a quarter are excluded from the sample to reduce misclassification between Junk-Rated and Investment-Grade groups.
- Baseline results are robust to including BBB firms as investment-grade.
- Note on rating definition change:
  - Prior to September 1, 1998: rating = “assessment of the creditworthiness of an obligor with respect to a senior or subordinated debt obligation”.
  - After September 1, 1998: rating = “opinion of an issuer’s overall creditworthiness”.
  - Because the analysis focuses on rating categories (Junk Rated vs. Investment-Grade) rather than specific ratings, this definitional change is considered alleviated; excluding BBB helps further lower misclassification risk.

### C Firm-level attentiveness and investment response to monetary shocks
- Regression specification (using attentiveness measure from Song and Stern (2020)):
  - ∆_h_ log(k_i,t+h_) = f_i,h + q_t+h + Trend +
    4∑_{j=1} [Θ′_h + Γ′_h ε^m_t] W_i,t−j +
    4∑_{j=1} [Ω′_h,j + Λ′_t ε^m_t] Y_t−j +
    (α_h + γ_h d_i,t−4 + β_h d_i,t) ε^m_t + u_i,h,t+h.
  - Y_t and W_t are vectors of aggregate and firm-level controls defined in Section 6.3.
  - d_i,t is measured annually; the regression uses quarterly data, so last year’s attentiveness d_i,t−4 is included as a control.
  - d_i,t equals one if firm i mentioned general economic conditions in its filing at time t, and zero otherwise.
- Interpretation:
  - Figure 7 (described): coefficient β_h for an expansionary monetary shock illustrated.
  - Higher values of d_i,t correspond to greater attentiveness; results suggest more attentive firms are more responsive to expansionary shocks.
  - Notes: point estimates and 90% confidence interval for β_h estimated from specification (24). Standard deviations are two-way clustered across both firms and time dimensions.

### D Quantifying information (mutual information function definitions)
- Information quantified as reduction in uncertainty (entropy) following Sims (2003) and literature on rational inattention.
- Entropy of normally distributed vector X = (X_1, ..., X_N):
  - H(X) = 1/2 log_2[(2πe)^N det Ω_X], where Ω_X is covariance matrix of X.
- Conditional entropy H(X|Y):
  - H(X|Y) = 1/2 log_2[(2πe)^N det Ω_{X|Y}], where Ω_{X|Y} is covariance matrix of X conditional on Y.
- Mutual information:
  - I(X;Y) = H(X) − H(X|Y) (and I(X;Y) = I(Y;X)).
- Mutual information for processes {X_t} and {Y_t}:
  - I({X_t};{Y_t}) = lim_{T→∞} 1/T [ H(X_0,...,X_{T−1}) − H(X_0,...,X_{T−1} | Y_0,...,Y_{T−1}) ].
  - For stationary Gaussian process {X_t,Y_t}:
    - I({X_t};{Y_t}) = lim_{T→∞} 1/T [ 1/2 log_2( det Ω_X / det Ω_{X|Y} ) ].
- Application: this definition is used for equation (19) where X_t = p^*_it and Y_t = s_it.

### E Derivation of the firm’s objective (outline and key steps)
- Manager values real dividend using stochastic discount factor Λ_t = β^t (C_t / C_0)^{−1}.
- Define g as discounted sum of dividends; ˆg as discounted sum in log-deviations; ̃g second-order Taylor approximation to ˆg.
- Vectors:
  - x′_t = (ˆp_it, ˆk_it+1)
  - z′_t = (ˆP_t, ˆW_t, ˆY_t, ˆq_t, ˆC_t, ˆr_t)
- Log-quadratic approximation to expected discounted sum of dividends given (equation (25)) contains terms:
  - β^{−1} h′_{−1} x_{−1} + 1/2 x′_{−1} H_{−1} x_{−1} + sum_{t=0}^∞ β^t [ linear and quadratic expectations in x_t and z_t ].
- Regularity conditions assumed to move expectation inside summation:
  - E_{i,−1}[ˆk^2_{j,−1}] < ∞ (26)
  - E_{i,−1}[ |ˆk_{j,−1} v_{n,0}| ] < ∞ for all n (27)
  - There exist δ < 1/β and A ∈ R such that for all m,n,t,τ = 0,1,2:
    - E_{i,−1}[ | v_{m,t} v_{n,t+τ} | ] < δ^t A (28)
- Definition of auxiliary process {x^*_t} satisfying (30) and use of iterated expectations yield equation (31).
- Expected loss from limited attention (difference between ̃g evaluated at realized x and at x^*) expressed in (32) and simplified through (33) to (34).
- Final expression (34):
  - E_{j,−1}[̃g(...) ] − E_{j,−1}[̃g(x^*_...,...)] = ∑_{t=0}^∞ β^t E_{j,−1}[ 1/2 (x_t − x^*_t)′ H_{x,0} (x_t − x^*_t) + (x_t − x^*_t)′ H_{x,1} (x_{t+1} − x^*_{t+1}) + (x_t − x^*_t)′ H_{x,2} (x_{t+2} − x^*_{t+2}) ].
- Second-order derivative matrices H_{x,0}, H_{x,1}, H_{x,2} defined as in (35)-(37) and computed (after algebra) as in (38)-(40).
  - Π_it = p_it P_t y_it − W_t P_t l_it
  - b_it = (1 + r_t) θ_π Π_it
- Evaluating these derivatives at steady state yields matrices for constrained and unconstrained firms.

### F Proof of Proposition (1) (key derivation and threshold)
- Comparison of attentiveness ratio ˆπ^2_12 / |ˆπ_11| between constrained and unconstrained firms.
- Define:
  - X = ξ ̄k θ^{−δ} ̄k + 1/R ∂ ̃b / ∂q (text uses notation: X = ξ ̄k θ − δ ̄k + 1/R ∂ ̃b / ∂q)
  - A = −1 + 1/(1+r) (θ ξ ̄k θ−1 + 1 − δ)
  - B = 1/(1+r) θ(1−θ) ̄k θ−2
- Constrained firm is more attentive iff:
  - (φ_d X + A)^2 / (φ_d + B) > A^2 / B
- Simplification yields threshold condition:
  - φ_d > A / X [ A X B − 2 ] (text: φ_d > A/X (A XB − 2) )
- Given A and B positive:
  - If X < 0 then right hand side positive and define ̄φ_d = A / X [ A X B − 2 ].
  - If X > 0 then A − 2 X B = 2 θ(1−θ) ̄k θ−1 /(1+r) [ ξ − 1/(2(1−θ)) − ( ξ /(r+δ) θ − δ + ̄1 / ̄k 1/R ∂ ̃b / ∂q |_{X} ̄k) ] (text gives expression).
  - If 0 < X < ̄k ξ^{−1}/[2(1−θ)] then ̄φ_d = A / X [ A X B − 2 ].
  - For X > ̄k ξ^{−1}/[2(1−θ)], ̄φ_d = 0.
- Final threshold definition:
  - ̄φ_d = max{ A / X [ A X B − 2 ], 0 }.

*Inattentiveness and the Investment Channel of Monetary Policy — Working Paper No. WP/2025/025*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025025-print-pdf.pdf_
