## wpiea2026127-source-pdf

## Source details

**Canonical URL:** [wpiea2026127-source-pdf](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026127-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026127-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026127-source-pdf.pdf.json)

---

### Main empirical finding: cross-sectional upstreamness differential
- Peak interaction coefficient: "2.8 percentage points larger than one at the 25th percentile, in response to a one percentage point monetary policy shock."
- Network-driven rigidity heterogeneity (the "flexibility channel") is identified as the primary driver of the cross-sectional pattern of price responses across industries.
- Cost propagation (cost-cascade) channel plays a secondary role and often partially offsets the flexibility channel because U.S. upstream sectors are intermediate-input intensive rather than labor-intensive.

### Theoretical mechanisms
- Flexibility channel
  - Production network determines which sectors are more flexible (upstream sectors selling predominantly to firms), producing larger price responses for those sectors.
  - With rigidity held fixed, a sector’s price tracks the flexible wage in proportion to its labor share.
- Cost-cascade channel
  - Upstream price adjustments propagate to downstream marginal costs via input linkages.
  - In U.S. data, this channel tends to tilt larger price responses toward downstream sectors, working against the upstream flexibility advantage.
- Asymmetric (sign-dependent) nominal rigidity
  - Asymmetric price rigidity (prices adjust more readily upward than downward) splits the duration-adjusted Leontief into distinct expansionary and contractionary matrices.
  - Model predicts upstreamness differential in price responses is substantially larger following expansionary shocks than contractionary shocks; empirical data are consistent with this.

### Empirical patterns by shock sign and real effects
- Asymmetric price responses
  - In response to a rate cut (expansionary shock), the peak interaction coefficient is more than twice the pooled peak.
  - Cumulating over the one-to-two-year window where responses diverge most, the upstreamness differential is large and negative following rate cuts but near zero following rate hikes, providing evidence of downward nominal rigidity.
  - Pattern corroborates the "rockets and feathers" observation that inflation is easier to generate than deflation.
- Real economy implications
  - Industries with more flexible prices absorb monetary shocks primarily through price changes, preserving real output.
  - Industries with sticky prices absorb shocks through quantity changes (quantity rationing).
  - Empirically, upstream sectors (more flexible) exhibit less negative real effects than downstream sectors.
  - No statistically significant real effects found in response to expansionary shocks; contractionary shocks lead upstream industries to cut prices and preserve output, while downstream industries absorb shocks via quantity rationing.
- Cross-sectional dispersion in prices and output is concentrated in contractionary episodes, consistent with asymmetric price rigidity mechanisms.

### Evidence for cost propagation through networks
- Three auxiliary empirical tests indicate buyer-seller price linkages transmit price adjustments across sectors:
  - Controlling for input-cost-weighted supplier prices absorbs most of the upstreamness interaction.
  - Interaction between upstreamness and the monetary shock is significantly stronger for industries with higher intermediate input shares.
  - An industry’s suppliers’ network positions predict its own price response.
- Conclusion: costs do propagate through the network even if they are not the primary driver of the upstreamness differential.

### Robustness and alternative specifications
- Results survive extensive robustness checks:
  - Alternative network measures (eigenvector centrality, intermediate input share) yield significant interaction coefficients of comparable magnitude.
  - Alternative monetary policy shocks (Jarociński and Karadi, 2020) yield comparable results.
  - Controlling for industry characteristics potentially correlated with upstreamness produces consistent results.
  - Results robust when constraining the sample to the manufacturing sector or when removing one industry at a time.
  - Permutation-based placebo tests indicate upstreamness—not an arbitrary industry ranking—generates the differential price response.

### Model structure and key equations (stylized)
- Multi-sector New Keynesian model with IO linkages; Calvo-style nominal rigidity with sector-specific θi and extension to sign-dependent reset probabilities.
- Households maximize lifetime utility:
  - E0 ∑t=0∞ βt [C1−σt /(1−σ) − L1+φt /(1+φ)]
- Composite consumption aggregates N sectors:
  - Ct = (∑i=1N ω1/ηi c(η−1)/ηit)η/(η−1)
- Firm production:
  - yijt = zit ℓαijt ∏k=1N mγikijk,t with αi + ∑N k=1 γik = 1
- Sectoral Calvo pricing: fraction 1−θi reset each period; slope of sector i Phillips curve λi ≡ (1−θi)(1−βθi)/θi.
- Sectoral Phillips curves (log-linearized):
  - πit = βEt πi,t+1 + λi [αi ŵt + ∑N k=1 γik p̂kt − p̂it]
- Matrix form:
  - πt = βEt πt+1 + Λ[A ŵt 1 + (Γ − I) p̂t], where Λ = diag(λ1, …, λN), A = diag(α1, …, αN), and Γ = [γik] is the IO matrix encoding the production network.

### Aggregate non-neutrality and network structure (selected propositions)
- Domar-weighted gross-output price index: P̂t = ∑i ξi p̂it (note ∑i ξi ≈ 1.8 in U.S. data).
- Proposition 2 (Network-adjusted rigidity): aggregate price response to a monetary shock equals ∑i ξi ∂p̂i,t+h / ∂εm t (Equation 9); real output effect approximated by Ŷt+h ≈ D̂t+h − P̂t+h (Equation 10). Output non-neutrality larger when network-weighted rigidity is high.
- Corollary 1 (Centrality-rigidity alignment): monetary policy has largest real effects when stickiest-price sectors are also most central (positive correlation between θi and ξi).
- Proposition 3 (Upstreamness and price dynamics), static case (β = 0): impact price response decomposes as ∂p̂it/∂εm t = ri × cmc i with ri = λi/(1 + λi); flexibility channel: ri increasing in Ui; cost-cascade channel: cmc i varies with Ui depending on IO linkages.
- Proposition 4 (Sign of the cost-cascade channel): first-order slope of price response on upstreamness from IO cost-cascade channel equals ¯r ˆ∆w Cov(αi, Ui) / Var(Ui) + O(¯r^2) (Equation 14); sign depends on Cov(αi, Ui).

### Data, sample, and shocks
- PPI data: monthly Bureau of Labor Statistics Producer Price Index at six-digit NAICS.
- Baseline sample: industries with BLS PPI coverage over February 1998 to December 2024; estimation sample is a monthly panel of 271 industries observed from February 1998 through December 2023, extended to end-2024 when using Jarociński and Karadi (2020) shocks.
- Monthly PPI inflation: mean 0.16% with a standard deviation of 1.8%.
- Upstreamness ranges from 1.00 to 3.77; aggregate row upstreamness 1.23, centrality 0.02.
- Preferred shock: Bauer and Swanson (2023) orthogonalized monetary policy surprise (MPS), aggregated to monthly frequency; approximately 8% of months contain two FOMC meetings.
- Robustness shock: Jarociński and Karadi (2020) shocks, available up to December 2024.

### Empirical strategy and identification
- Local projection (Jordà, 2005) horizons h = 0,…,36 months estimating:
  - ∆h pit = αh i + μh s(i),t + βh (Ui × εm t) + ρh ∆ pit−1 + uh it.
- Ui: upstreamness; εm t: Bauer‑Swanson monetary policy shock; industry fixed effects and sector×month fixed effects absorb main effects.
- βh interpretation: how cumulative price response to a monetary shock varies with sector position; under model βh predicted negative at medium-to-long horizons.
- Identification assumptions:
  1. Exogeneity of monetary shocks (after orthogonalization and controls).
  2. Stability of network position (rank correlation of upstreamness between 2002 and 2022 is 0.94).
- Protections: sector×time fixed effects, orthogonalization, placebo tests, and permutation checks.

### Key empirical magnitudes and timing
- Interaction coefficient dynamics (selected horizons):
  - h = 0: −0.013 (SE 0.010)
  - h = 12: −0.101*** (0.038)
  - h = 15: −0.105** (0.041)
  - h = 18: −0.113*** (0.039) (peak)
  - h = 21: −0.081** (0.039)
  - h = 24: −0.093** (0.043)
  - h = 36: 0.019 (0.039)
- Interpretation: peak interaction coefficient −0.113 at 18 months implies with interquartile upstreamness range ≈ 0.25 units, an industry at the 75th percentile exhibits a peak cumulative price response about 2.8 percentage points larger than an industry at the 25th percentile per unit of the monetary shock.
- Expansion vs contraction
  - Expansionary (rate cuts) peak: interaction coefficient reaches −0.25 about 27 months after the shock.
  - Contractionary interaction coefficient: generally not significant and noisier (standard errors on contractionary side over 50% larger at earlier horizons).
  - Wald tests reject symmetry after 20 months; mean expansionary coefficient statistically different from mean contractionary one over h = 12–24 and h = 0–36.

### Quantitative fit and mechanism decomposition
- Calibration inputs:
  - External estimate of δ (upstreamness–flexibility slope) ˆδ = −0.152 from Pasten et al. (2020); average monthly reprice frequency 14.8% implies ˆ̄θ = 0.85 (mean price duration ≈ seven months).
  - Regression on PPI panel yields ˆδ = −0.166 (robustness).
- Calibrated model reproduces near-zero impact, hump-shaped build-up, peak at eighteen months:
  - Model peak: −0.105 captures 93% of empirical peak −0.113; both peak at roughly one and a half years.
- Mechanism decomposition at peak:
  - Flexibility channel accounts for roughly 120% of total interaction.
  - Cost-cascade channel contributes about −50%.
  - Interaction of the two channels makes up remaining +30%.
- Aggregate implication: at widest divergence (~one year after shock) cumulative price response in network economy is roughly 8% smaller in magnitude than representative-firm benchmark (Γ = 0, homogeneous θi), because IO cost-cascade offsets part of rigidity heterogeneity amplification.

### Pricing-friction variants and asymmetry foundations
- Menu-cost (Ss) model extension with strategic complementarities and IO linkages:
  - Menu-cost model peak: −0.125 (h = 15) vs Calvo model −0.105 and empirical −0.113.
  - Menu-cost model introduces endogenous extensive-margin selection reinforcing flexibility channel and provides structural basis for asymmetry (asymmetric Ss bands).
- Both time-dependent (Calvo) and state-dependent (menu-cost) frameworks generate hump-shaped interaction and asymmetry (expansionary shocks produce larger upstream-downstream differentials).

### Input intensity and supplier-neighbor evidence
- Triple interaction test (Ui × α^II_i × εm t) finds β̂_2 generally negative and significant in second year, indicating industries that are both upstream and input-intensive respond more strongly.
- Supplier-weighted upstreamness Ū_supplier_i = ∑k γik U_k interaction with εm t is negative and significant in the second year, showing suppliers’ network positions predict own-industry price responses.
- Table 2 (18-month horizon) selected coefficients:
  - Upstreamness × MP shocks: −0.113*** (0.039)
  - Centrality (eigenvector): −0.384*** (0.131)
  - Intm. Share: −0.728*** (0.280)
  - JK MP Shocks: −0.153** (0.059)
  - Observations: 69,772; R^2 ≈ 0.003–0.004.
- Table 3 (18-month horizon) robustness to industry characteristics: upstreamness × MP shocks remains significant across multiple controls (e.g., with labor share, energy share, intermediate share, trade exposure; baseline −0.113*** (0.039); all confounders included −0.090*** (0.034)).

### Robustness and validation checks
- Pre-trend/placebo: interaction at negative horizons small and indistinguishable from zero; permutation placebo: actual estimate lies in extreme tail of permutation distribution.
- Subsamples and leave-one-out: manufacturing-only and leave-one-out across 271 industries produce consistent estimates; no single industry drives results.
- Alternative shocks and measures: results hold using Jarociński and Karadi (2020) shocks and alternative network measures (eigenvector centrality, intermediate input share).
- Specification variants: adding richer lag structure, macro controls, using log(upstreamness) yield consistent interaction patterns.

### Policy implications
- Production network shapes the cross-sectional distribution of price rigidity; structural changes to the IO matrix can alter monetary policy potency even if average price rigidity is unchanged.
- Policies or shifts that alter IO structure—trade agreements, vertical integration, platform-mediated disintermediation—change B2B vs B2C composition and thus reshape rigidity distribution across network positions.
- Secular increase in services share of GDP shifts activity toward more downstream and more price-rigid sectors, increasing the wedge between upstream and downstream rigidity and potentially strengthening asymmetry.
- Network amplification flattens the aggregate Phillips curve: denser IO networks and rigid upstream sectors reduce κnet relative to no-network benchmark, implying monetary policy must move interest rates by more to achieve a given change in inflation.
- Central banks face structural asymmetry: rate cuts generate larger cross-sectional price dispersion than rate hikes; average rigidity is not a sufficient statistic for monetary potency—tracking network evolution matters for inflation monitoring.

*Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026127-source-pdf.pdf*

### 2.8 percentage points larger than one at the 25th percentile, in response to a one percentage point

### wpiea2026127-source-pdf - 2.8 percentage points larger than one at the 25th percentile, in response to a one percentage point

### Main empirical finding: cross-sectional upstreamness differential
- The peak interaction coefficient is "2.8 percentage points larger than one at the 25th percentile, in response to a one percentage point monetary policy shock."
- A network-driven rigidity heterogeneity (the "flexibility channel") is identified as the primary driver of the cross-sectional pattern of price responses across industries.
- The cost propagation (cost-cascade) channel plays a secondary role and often partially offsets the flexibility channel because U.S. upstream sectors are intermediate-input intensive rather than labor-intensive.

### Theoretical mechanisms
- Flexibility channel:
  - Production network determines which sectors are more flexible (upstream sectors selling predominantly to firms), producing larger price responses for those sectors.
  - With rigidity held fixed, a sector’s price tracks the flexible wage in proportion to its labor share.
- Cost-cascade channel:
  - Upstream price adjustments propagate to downstream marginal costs via input linkages.
  - In U.S. data, this channel tends to tilt larger price responses toward downstream sectors, working against the upstream flexibility advantage.
- Asymmetric (sign-dependent) nominal rigidity:
  - Extending the model to allow asymmetric price rigidity (prices adjust more readily upward than downward) splits the duration-adjusted Leontief into distinct expansionary and contractionary matrices.
  - The model predicts that upstreamness differential in price responses is substantially larger following expansionary shocks than contractionary shocks; the data are consistent with this.

### Empirical patterns by shock sign and real effects
- Asymmetric price responses:
  - In response to a rate cut (expansionary shock), the peak interaction coefficient is more than twice the pooled peak.
  - Cumulating over the one-to-two-year window where responses diverge most, the upstreamness differential is large and negative following rate cuts but near zero following rate hikes, providing evidence of downward nominal rigidity.
  - This pattern corroborates the "rockets and feathers" aggregate observation that inflation is easier to generate than deflation.
- Real economy implications:
  - Industries with more flexible prices absorb monetary shocks primarily through price changes, preserving real output.
  - Industries with sticky prices absorb shocks through quantity changes (quantity rationing).
  - Empirically, upstream sectors (more flexible) exhibit less negative real effects than downstream sectors.
  - No statistically significant real effects are found in response to expansionary shocks; contractionary shocks lead upstream industries to cut prices and preserve output, while downstream industries absorb shocks via quantity rationing.
- Cross-sectional dispersion in prices and output is concentrated in contractionary episodes, consistent with asymmetric price rigidity mechanisms.

### Evidence for cost propagation through networks
- Three auxiliary empirical tests indicate buyer-seller price linkages transmit price adjustments across sectors:
  - Controlling for input-cost-weighted supplier prices absorbs most of the upstreamness interaction.
  - The interaction between upstreamness and the monetary shock is significantly stronger for industries with higher intermediate input shares.
  - An industry’s suppliers’ network positions predict its own price response.
- These findings imply costs do propagate through the network even if they are not the primary driver of the upstreamness differential.

### Robustness and alternative specifications
- Results survive extensive robustness checks:
  - Alternative network measures (eigenvector centrality, intermediate input share) yield significant interaction coefficients of comparable magnitude.
  - Alternative monetary policy shocks (Jarociński and Karadi, 2020) yield comparable results.
  - Controlling for industry characteristics potentially correlated with upstreamness produces consistent results.
  - Results are robust when constraining the sample to the manufacturing sector or when removing one industry at a time.
  - Permutation-based placebo tests indicate upstreamness—not an arbitrary industry ranking—generates the differential price response.

### Model structure and key equations (stylized)
- Multi-sector New Keynesian model with IO linkages; Calvo-style nominal rigidity with sector-specific θi and extension to sign-dependent reset probabilities.
- Households maximize lifetime utility:
  - E0 ∑t=0∞ βt [C1−σt /(1−σ) − L1+φt /(1+φ)]
- Composite consumption aggregates N sectors:
  - Ct = (∑i=1N ω1/ηi c(η−1)/ηit)η/(η−1)
- Firm production:
  - yijt = zit ℓαijt ∏k=1N mγikijk,t with αi + ∑N k=1 γik = 1
- Sectoral Calvo pricing: fraction 1−θi reset each period; slope of sector i Phillips curve λi ≡ (1−θi)(1−βθi)/θi.
- Sectoral Phillips curves (log-linearized):
  - πit = βEt πi,t+1 + λi [αi ŵt + ∑N k=1 γik p̂kt − p̂it]
- Matrix form:
  - πt = βEt πt+1 + Λ[A ŵt 1 + (Γ − I) p̂t], where Λ = diag(λ1, …, λN), A = diag(α1, …, αN), and Γ = [γik] is the IO matrix encoding the production network.

### Policy implications
- If the cross-sectional distribution of price rigidity is shaped by the production network, structural changes to the IO matrix can alter monetary policy potency even if average price rigidity is unchanged.
- Any policy or structural shift that alters the IO matrix—trade agreements, vertical integration, platform-mediated disintermediation—changes the composition of B2B versus B2C activity and thus reshapes the distribution of rigidity across network positions.
- The secular increase in the services share of GDP shifts activity toward sectors that are more downstream and more price-rigid, increasing the wedge between upstream and downstream rigidity and potentially strengthening the documented asymmetry.

*Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026127-source-pdf.pdf*

### 2.3    Aggregate non-neutrality and network structure

### 2.3    Aggregate non-neutrality and network structure

### Definitions and core objects
- Domar weights: ξi is sector i’s total sales (to both final consumers and other firms) to aggregate GDP. In the steady state, ξ = (I−Γ′)−1 ω, where ω is the vector of final demand shares.
- Upstreamness: Ui = e′i (I−A)−1 1, where A is the output-allocation matrix with aik = Zik/Yi. Ui = 1 for a sector that sells entirely to final consumers; higher Ui indicates more upstream positions.
- Aggregate price index (Domar-weighted gross-output price index): P̂t = ∑i ξi p̂it. Note: Because ∑i ξi ≈ 1.8 in U.S. data, P̂t is a gross-output-weighted price index that counts intermediate transactions.

### Aggregate non-neutrality: propositions and implications
- Proposition 2 (Network-adjusted rigidity):
  - Aggregate price response to a monetary shock εm t at horizon h:
    - ∂P̂t+h / ∂εm t = ∑i=1^N ξi ∂p̂i,t+h / ∂εm t. (Equation 9)
  - First-order approximation for real output effect:
    - Ŷt+h ≈ D̂t+h − P̂t+h. (Equation 10)
  - Implication: Output non-neutrality is larger when P̂t+h responds slowly to the monetary shock—i.e., when network-weighted rigidity is high.
- Corollary 1 (Centrality-rigidity alignment):
  - Aggregate price level is slowest to adjust, and monetary non-neutrality is greatest, when the correlation between θi and ξi is positive. Thus monetary policy has largest real effects when the stickiest-price sectors are also the most central in the production network.
- Relation to single-sector Calvo:
  - In a one-sector economy, a single θ parameter determines non-neutrality. In a network economy, what matters is not ¯θ but the Domar-weighted average ˜θ ≡ ∑i ξi θi.

### Cross-sectional predictions and maintained assumption
- Empirical motivation: B2B prices are renegotiated more frequently than consumer-facing prices.
- Assumption 1: More upstream sectors have weakly more flexible prices: θi is weakly decreasing in Ui, so λi = (1−θi)(1−βθi)/θi is weakly increasing in Ui. Equivalently, the price-adjustment probability ri ≡ λi/(1 + λi) is weakly increasing in Ui.
- Proposition 3 (Upstreamness and price dynamics), static case (β = 0), following a contractionary monetary shock (εm t > 0):
  - (i) General network: Impact price response of sector i decomposes as
    - ∂p̂it / ∂εm t = ri × cmc i, (Equation 11)
    - where ri = λi/(1 + λi) and cmc i > 0 is the marginal cost response cmc i = αi ŵt + ∑k γik p̂kt.
    - Flexibility channel: ri increasing in Ui provides force toward larger price responses for more upstream sectors.
    - Cost-cascade channel: how cmc i varies with Ui depends on IO linkages; a quantitative question resolved in calibration.
  - (ii) Chain economy: For a chain of K sectors with uniform input share γ and uniform r:
    - Price response of sector k:
      - xk = r [ (1−γ) + γ(1−r)(rγ)^{k−1} ] / (1− rγ). (Equation 12)
      - xk is strictly decreasing in k (position toward final demand).
    - Upstream-downstream price differential x1 − xK:
      - DK = rγ(1−r) [1− φ^{K−1}] / [1− φ], where φ = rγ. (Equation 13)
      - Differential is increasing in K and converges to rγ(1−r)/(1−φ) as K → ∞.
  - (iii) Chain economy: Output. Real output response of sector k satisfies ŷkt ≈ d̂kt − p̂kt; more downstream sectors have smaller |p̂kt|, so their output gap remains open longer and peak real effect occurs at a later horizon.

### Cost-cascade channel sign and condition
- Proposition 4 (Sign of the cost-cascade channel): Hold flexibility homogeneous at ¯r and consider impact price response to a nominal demand impulse ˆ∆w. To first order in ¯r, the cross-sectional regression slope of the price response on upstreamness attributable to the IO cost-cascade channel is
  - ∂/∂U b p_cost i = ¯r ˆ∆w Cov(αi, Ui) / Var(Ui) + O(¯r^2). (Equation 14)
- Implication: Cost-cascade channel reinforces the flexibility channel if Cov(αi, Ui) > 0 and offsets it if Cov(αi, Ui) < 0. Thus whether upstream sectors respond more depends also on how labor shares αi covary with upstreamness Ui.

### Two-sector chain example and intuition
- IO matrix for a two-sector chain (N = 2):
  - Γ = [[0, 0], [γ, 0]]. (Equation 15)
- Sector labor shares: α1 = 1 (sector 1 uses only labor); α2 = 1−γ.
- Output-allocation matrix yields upstreamness:
  - (I−A)^{-1} = [[1, γ], [0, 1]] → U1 = 1 + γ, U2 = 1.
- Static Phillips curves (β = 0):
  - π1t = βEt π1,t+1 + λ1(ŵt − p̂1t). (Equation 16)
  - π2t = βEt π2,t+1 + λ2[(1−γ) ŵt + γ p̂1t − p̂2t]. (Equation 17)
- Static impact responses (letting ˆ∆w ≡ ∂ ŵt / ∂εm t):
  - ∂p̂1t / ∂εm t = λ1/(1 + λ1) ˆ∆w. (Equation 18)
  - ∂p̂2t / ∂εm t = λ2/(1 + λ2) [ (1−γ) ˆ∆w + γ · λ1/(1 + λ1) ˆ∆w ]. (Equation 19)
- Intuition:
  - Flexibility force: upstream sector passes larger share of marginal cost change into price because λ1 > λ2 under Assumption 1.
  - Cost-linkage force: downstream marginal cost depends on sticky upstream price p̂1t weighted by γ; sign depends on labor-share structure (e.g., if upstream α1 is high, cost-linkage reinforces flexibility force).

### Asymmetric price rigidity and network amplification
- Empirical motivation: Prices adjust asymmetrically; firms are less likely to lower prices than to raise them (Peltzman 2000; Nakamura and Steinsson 2008).
- Asymmetric Calvo pricing (reduced-form): Firms reset price with probability 1 − θs i, s ∈ {+,−}, where s = + if p̂∗it > p̂it and s = − if p̂∗it < p̂it. Asymmetric rigidity corresponds to θ−i > θ+i.
- State-dependent Phillips curve slopes:
  - λs i = (1−θs i)(1−βθs i)/θs i, s ∈ {+,−}. (Equation 21)
  - Asymmetric rigidity (θ−i > θ+i) ⇒ λ−i < λ+i: Phillips curve is flatter when firms need to cut prices than to raise them.
- Sectoral Phillips curve with state s:
  - πit = βEt πi,t+1 + λs t i [ αi ŵt + ∑k γik p̂kt − p̂it ]. (Equation 22)
- Proposition 5 (Sign-dependent network propagation), static economy (β = 0):
  - Impact price vector to a nominal demand impulse of sign s:
    - p̂s = Ms α ˆ∆w, Ms ≡ (I − Rs Γ)^{-1} Rs. (Equation 23)
  - Properties:
    - (i) Nesting: If r+ i = r− i for all i, Ms are identical and responses independent of sign.
    - (ii) Asymmetry: If r+ i ≥ r− i for all i with strict inequality for some i, then M+ ≥ M− elementwise, so |p̂+ i| ≥ |p̂− i| for every sector: price responses uniformly larger after expansionary shocks; cross-sectional dispersion with respect to upstreamness is wider.
  - Ms is a sign-dependent dampened Leontief matrix; asymmetry arises because R+ ≥ R− implies M+ ≥ M− elementwise via the Neumann series Ms = ∑_{k=0}^∞ (Rs Γ)^k Rs.
- Proposition 6 (Asymmetric network amplification: chain economy), asymmetric Calvo with θ−i > θ+i for all i, chain of K sectors, β = 0:
  - (i) Upstream-downstream price differential under shock sign s ∈ {+,−}:
    - Ds K = rs γ(1−rs) [1− (φs)^{K−1}] / [1− φs], where rs = λs/(1 + λs) and φs = rs γ. (Equation 24)
  - (ii) Ratio of expansionary to contractionary differential:
    - RK = [ r+ (1− r+) / ( r− (1− r−) ) ] × [ ∑_{j=0}^{K−2} (φ+)^{j} / ∑_{j=0}^{K−2} (φ−)^{j} ]. (Equation 25)
    - Under restriction r+, r− < 1/2 with r+ > r−:
      - (a) RK > 1 for all K ≥ 2;
      - (b) RK is strictly increasing in K;
      - (c) RK converges to finite limit R∞ = [ r+ (1− r+) (1− φ−) ] / [ r− (1− r−) (1− φ+) ].
  - Interpretation:
    - First factor r(1−r) captures own-flexibility asymmetry and operates even for K = 2.
    - Second factor (ratio of partial geometric sums) captures network amplification; equals one at K = 2, increases with K, and converges to (1−φ−)/(1−φ+) > 1.
    - Longer chains amplify asymmetry because stronger expansionary cascade (φ+ > φ−) cumulates over more stages, but amplification is bounded.
- Testable prediction:
  - Interaction between upstreamness and monetary shocks should be driven predominantly by expansionary episodes.
  - Asymmetry should be stronger for industries with higher upstreamness (more deeply embedded in the production network), because difference in cascade factors cumulates over more stages.

### Data note
- Empirical analysis combines three data sources: industry-level producer prices, U.S. IO tables, and a measure of monetary policy stance.

*Source: 2.3 Aggregate non-neutrality and network structure (from the supplied PDF content).*

### 3.1    Producer price indices

### 3.1    Producer price indices

### PPI data and sample
- Monthly Producer Price Index (PPI) data from the Bureau of Labor Statistics.
- Baseline sample covers all industries with BLS Producer Price Index coverage at the six-digit NAICS level over the period February 1998 to December 2024.
- Included sectors: manufacturing (NAICS 311–339), mining, utilities, wholesale trade, retail trade, transportation, information, and selected service sectors.
- The panel is unbalanced: some series (e.g., Food Manufacturing, NAICS 311) available from the late 1990s; others (e.g., Wholesale Trade subsectors) begin around 2003 when BLS expanded PPI coverage.
- Of the 274 industries in the BEA IO tables, three are excluded for having less than 10 years of PPI data; the estimation sample is a monthly panel of 271 industries observed from February 1998 through December 2023, extended to end-2024 when using the Jarociński and Karadi (2020) shocks.
- Monthly PPI inflation averages 0.16% with a standard deviation of 1.8%.

### Table 1 — industries and network position (summary)
- Total industries: 274
- Upstreamness and centrality reported by upstreamness quartile (N per quartile indicated):
  - Q1 (Most downstream) (N = 69)
    - Min: Direct life insurance carriers — Upstreamness 1.00, Centrality 0.00
    - Max: Fan, blower, air purification equipment mfg — Upstreamness 1.03, Centrality 0.00
  - Q2 (N = 68)
    - Min: Cut stone and stone product manufacturing — Upstreamness 1.03, Centrality 0.00
    - Max: Clay product and refractory manufacturing — Upstreamness 1.11, Centrality 0.00
  - Q3 (N = 68)
    - Min: Dry, condensed, and evaporated dairy products — Upstreamness 1.11, Centrality 0.00
    - Max: Architectural & structural metals mfg — Upstreamness 1.28, Centrality 0.01
  - Q4 (Most upstream) (N = 69)
    - Min: Portfolio management and investment advice — Upstreamness 1.28, Centrality 0.06
    - Max: Iron and steel mills and ferroalloy mfg — Upstreamness 3.77, Centrality 0.08
- Aggregate row: Total industries: 274 — Upstreamness 1.23, Centrality 0.02
- Notes on computation:
  - Upstreamness computed following Antràs et al. (2012) from the BEA Detailed IO tables.
  - Centrality is eigenvector centrality computed as the dominant eigenvector of the direct requirements matrix.
  - N is the number of industries in the quartile. The full sample includes all industries with BLS PPI coverage.

### IO tables and network position
- Production network constructed using the BEA 2012 Detailed IO tables (Detailed-level Use table) mapped to six-digit NAICS via a crosswalk.
- Technical coefficients matrix Γ = [γik] obtained by normalizing rows, where γik is the cost share of inputs from sector k in sector i’s total intermediate expenditure.
- Upstreamness definition (following Antràs et al. 2012):
  - Ui = 1×Fi/Yi + 2×∑k aik Fk/Yi + 3×∑k∑j aij ajk Fk/Yi + ··· = e′i (I − A)−1 1,
    where Fi is final demand for industry i’s output, Yi is total output, and A is the output-allocation matrix with aik = Zik/Yi.
- Upstreamness ranges from 1.00 to 3.77 in the full sample.
- Eigenvector centrality vi satisfies vi = (1/λmax) ∑k γki vk, where λmax is the largest eigenvalue of Γ′.
- Alternative raw measure: share of intermediate inputs in total purchases (used in robustness tests).
- Network features (Figure 1 description):
  - Panel (a): Network visualized aggregated to three-digit NAICS for legibility; node size ∝ eigenvector centrality; color = upstreamness quartile; directed edges from supplier to customer with edge width ∝ value of flow from BEA Detailed IO tables. Network is dense; most sectors have multiple suppliers and customers.
  - Panel (b): Cross-sectional upstreamness distribution is right-skewed with a modal value near 1.0 and a long right tail beyond 3.5; quartile boundaries divide sample into four groups with meaningfully different production chain positions.

### Monetary policy shocks
- Preferred shock: orthogonalized monetary policy surprise (MPS) of Bauer and Swanson (2023), available up to December 2023.
  - Construction: (1) high-frequency changes in federal funds futures prices in a narrow window around FOMC announcements isolate market’s revision of expected policy path; (2) raw surprises orthogonalized with respect to a set of macroeconomic and financial variables to purge the “Fed information effect.”
  - Available at FOMC-meeting frequency; aggregated to monthly frequency by summing surprises within each calendar month (months without FOMC meetings receive a shock of zero).
  - In months with two FOMC meetings, the monthly shock is the sum of both surprises; in the sample, approximately 8% of months contain two meetings.
- Robustness shock: Jarociński and Karadi (2020) shocks, available up to December 2024.
  - Identification: sign-restriction decomposition on high-frequency changes in interest rates and stock prices around FOMC announcements; median-rotation estimator isolates monetary policy component (higher rates, lower stocks) and purges information shocks (higher rates, higher stocks).

### Sample construction and descriptive statistics
- Estimation sample: monthly panel of 271 industries from February 1998 through December 2023; extended to end-2024 with Jarociński and Karadi (2020) shocks.
- Three BEA industries excluded for having less than 10 years of PPI data.
- Table A2 (Appendix A-IV) reports summary statistics.
- Monthly PPI inflation: mean 0.16%, standard deviation 1.8%.
- Industries display substantial heterogeneity in upstreamness.

### Empirical strategy — estimation
- Local projection method of Jordà (2005) to estimate impulse response functions for horizons h = 0, 1, ..., H with H = 36 months.
- Estimated equation for each horizon h:
  - ∆h pit = αh i + μh s(i),t + βh (Ui × εm t) + ρh ∆ pit−1 + uh it,
    where ∆h pit = ln(PPIi,t+h) − ln(PPIit) is cumulative log price change from month t to t + h; at h = 0 the dependent variable is contemporaneous monthly PPI inflation πit.
- Regressor definitions and fixed effects:
  - Ui is upstreamness of industry i.
  - εm t is the Bauer-Swanson monetary policy shock.
  - αh i are six-digit NAICS industry fixed effects.
  - μh s(i),t are three-digit NAICS sector × month fixed effects.
  - ∆pit−1 denotes the lagged PPI inflation.
- Interpretation:
  - Coefficient βh captures how the cumulative price response to a monetary shock varies with sector position in the production network.
  - Under the model (Proposition 3), βh is predicted to be negative at medium-to-long horizons: more upstream sectors exhibit larger cumulative price declines after contractionary monetary shocks.
- Identification of βh:
  - Main identification comes from within-sector variation in upstreamness across six-digit industries interacted with time-series monetary shock variation; main effects of Ui and εm t absorbed by industry and sector×time fixed effects respectively.
  - Sector×time fixed effects μh s(i),t absorb time-varying shocks common within three-digit NAICS sectors.
  - A variance decomposition indicates 56% of total cross-sectional variance in upstreamness is within three-digit sectors, with a median within-sector standard deviation of 0.15 and a median within-sector range of 0.40 (overall standard deviation 0.35), implying broad-based identification rather than a few within-sector comparisons.

### Empirical strategy — identification assumptions and threats
- Two core assumptions:
  1. Exogeneity of monetary policy shocks: conditional on industry fixed effects and lagged macro controls, Ui × εm t is uncorrelated with the error term.
  2. Stability of network position: upstreamness Ui is stable over the sample period and reflects technological characteristics rather than endogenous responses to monetary conditions.
- Supporting evidence and protections:
  - FOMC targets aggregate inflation (CPI or PCE), not sector-specific producer prices.
  - Bauer-Swanson orthogonalization removes the component of policy surprises correlated with the Fed’s private information.
  - Sector×time fixed effects absorb time-varying shocks common within three-digit NAICS sectors including sectoral demand shifts, commodity price movements, and trade shocks.
  - Rank correlation of industry-level upstreamness between BEA annual IO tables for 2002 and 2022 is 0.94, supporting stability.
  - The 2012 BEA Detailed table is used for baseline analysis.
- Discussed threats and mitigations:
  - Concern: simultaneous shocks (e.g., oil price shocks) that both trigger Fed actions and differentially affect upstream industries. Mitigation: sector×time fixed effects and orthogonalization.
  - Concern: upstream vs downstream differences in demand elasticities. Mitigation: industry fixed effects and focus on interaction term.
  - Concern: industry demand/productivity/cost shocks correlated with both monetary shock and upstreamness. Mitigation: orthogonalization, sector-time fixed effects, and robustness checks.
  - Concern: influential observations in smaller samples addressed in robustness analysis (outlier sensitivity).
  - Note: relaxing the PPI data-length threshold to 60 months adds three additional industries without materially affecting estimates.

### Relationship with the theoretical model
- βh measures cross-sectional variation in cumulative price responses across industries with different upstreamness.
- Model forces shaping βh (Proposition 3):
  - Dominant force: systematic relationship between network position and intrinsic price rigidity — upstream sectors sell to other firms and B2B prices adjust more frequently than consumer-facing prices, generating correlation between upstreamness and the Calvo parameter so more upstream sectors are inherently more flexible.
  - Secondary force: cost-cascade channel — each sector’s marginal cost depends on suppliers’ prices, which are sticky, introducing additional cross-sectional variation.
- Methodological notes:
  - Equation 27 is a reduced-form specification motivated by the model; it does not impose structural cross-equation restrictions.
  - PPI data correspond to gross-output prices (inclusive of intermediate input costs), aligning with the model’s gross-output price indices and the sectoral Phillips curve emphasis on cost-cascade channels. A value-added price index would strip out intermediate input costs and thus would not capture the cost-cascade channel emphasized by the model.

*Source: wpiea2026127-source-pdf - 3.1    Producer price indices*

### 4.2    Asymmetric effects of expansionary and contractionary shocks

### 4.2    Asymmetric effects of expansionary and contractionary shocks

### Theory and empirical test for asymmetry
- Baseline specification constrains the interaction coefficient βh to be the same for positive and negative monetary policy shocks, but the model implies potential asymmetry because network-determined flexibility interacts with asymmetric price adjustment.
- Asymmetric price rigidity hypothesis:
  - Price increases propagate more readily than price decreases.
  - When the Fed cuts rates (expansionary), upstream firms—being more flexible—raise prices readily, producing a large upstreamness differential.
  - When the Fed tightens (contractionary), the flatter downward Phillips curve slope attenuates the cross-sectional differential.
- Empirical test augmentation (Equation 28):
  - Define D+ t = 1{εm t > 0} (contractionary months) and D− t = 1{εm t < 0} (expansionary months).
  - Local projection estimated for each horizon h = 0, 1, ..., H:
    - ∆h pit = αh i + μh s(i),t + βh,+ (Ui × εm t × D+ t) + βh,− (Ui × εm t × D− t) + ρh ∆ pit−1 + uh it.
  - Key coefficients:
    - βh,+ captures upstreamness effect on price responses during contractionary episodes.
    - βh,− captures upstreamness effect during expansionary episodes.
- Sample composition over the baseline sample:
  - 118 months feature contractionary shocks (εm t > 0).
  - 93 months feature expansionary shocks (εm t < 0).
  - Mean magnitudes: 0.040 and −0.044, respectively.
  - An additional 100 months have zero shock (no FOMC meeting).

### Empirical findings: network position and price dynamics
- Interaction-coefficient dynamics (estimated from Equation 27):
  - Following a contractionary monetary shock, industries with higher upstreamness experience larger cumulative price declines.
  - The interaction coefficient is near zero at impact, becomes progressively more negative and significant, reaching the peak one year and a half after the shock, and begins to dissipate after 27 months.
  - Interpretation: upstream prices are more flexible (or downstream prices are stickier due to accumulated nominal rigidities across stages).
  - Hump-shaped pattern consistent with Proposition 3: transient medium-run upstreamness effect before long-run uniform adjustment.
- Quartile responses (Panel (b) of Figure 2):
  - Four upstreamness quartiles display monotonic ordering: Q4 > Q3 > Q2 ≈ Q1 ≈ 0.
  - Q4 (most upstream) includes petroleum refining, basic chemicals, iron and steel mills, other primary materials producers—largest cumulative price adjustment.
  - Q1 and Q2 (most downstream) include food manufacturing, apparel, furniture, retail-facing sectors—responses small and statistically indistinguishable from zero at all horizons.
  - Timing: most upstream industries begin showing cumulative responses around the 6-month horizon; gap widens through 12 months before narrowing at longer horizons.
- Magnitude and economic interpretation:
  - Peak interaction coefficient: −0.113 one year and a half after the shock.
  - Interquartile range of upstreamness: approximately 0.25 units.
  - Thus, per unit of the monetary shock, an industry at the 75th percentile of upstreamness exhibits a peak cumulative price response about 2.8 percentage points larger than an industry at the 25th percentile.
  - Quartile estimates: Q4 shows large and significant cumulative responses; Q1 statistically indistinguishable from zero.

### Quantitative fit: calibrated multi-sector model
- Calibration inputs and discipline:
  - Use BEA IO matrix and simulate model-implied path of interaction coefficients βh,model to compare with estimated β̂h.
  - Pricing parameters:
    - ̄θ is the average Calvo parameter governing baseline frequency of price adjustment.
    - δ captures cross-sectional variation in price rigidity: θi = ̄θ + δ(Ui − ̄U).
    - A negative δ implies more upstream sectors have more flexible prices (consistent with Assumption 1).
  - Demand impulse ˆwt: disciplined to empirical transmission of the monetary shock to real activity (hump-shaped, troughing at seventeen months).
- External discipline from Pasten et al. (2020) sectoral price-adjustment frequencies:
  - Estimated slope mapping frequency to upstreamness: ˆδ = −0.152 (more upstream sectors reprice more frequently).
  - Average monthly frequency: 14.8% implies ˆ̄θ = 0.85, a mean price duration of about seven months.
  - Regression on author’s own PPI panel yields nearly identical slope: ˆδ = −0.166 (robustness check).
- Model fit to empirical interaction coefficient:
  - Calibrated model reproduces three salient features: near-zero impact response, gradual build-up to a peak at eighteen months, and subsequent reversion.
  - Model peak: −0.105, capturing 93% of empirical peak −0.113; both peak at roughly one and a half years.
  - Near-zero impact response arises because the demand impulse is small on impact and builds over time.
- Mechanism decomposition (Panel (b) of Figure 3):
  - Two channels:
    - Flexibility channel: systematic variation in price rigidity across network positions (heterogeneous θi, Γ = 0 for isolation).
    - Cost-cascade channel: propagation of cost shocks along production structure (homogenize θi = ̄θ but retain Γ).
  - At the peak horizon:
    - Flexibility channel accounts for roughly 120% of the total interaction coefficient.
    - Cost-cascade channel contributes about −50%.
    - Interaction of the two channels makes up remaining +30%.
  - Interpretation: holding rigidity constant, IO linkages attenuate the upstreamness differential rather than amplify it; heterogeneous rigidity channel must dominate for observed pattern to emerge.
  - Empirical covariance: Corr(Ui, αi) = −0.90 (upstream sectors are intermediate-input intensive rather than labor intensive), so cost linkages work against the flexibility advantage of upstream sectors.
- Aggregate implication:
  - Compare network calibration (full Γ and heterogeneous θi = ˆ̄θ + ˆδ(Ui − ̄U)) to representative-firm benchmark (Γ = 0, homogeneous θi = ˆ̄θ).
  - At point of widest divergence, around one year after the shock, cumulative price response in the network economy is roughly 8% smaller in magnitude than in the representative-firm benchmark.
  - The gap is modest because IO cost-cascade channel offsets much of the amplification from rigidity heterogeneity.

### Robustness: pricing friction variants and selection effects
- Calvo vs. menu cost models:
  - Calvo framework imposes constant exogenous probability of price adjustment, missing nonlinear selection channel of menu cost models.
  - Extensive-margin selection effect in menu-cost/Ss pricing: shocks push some firms past adjustment thresholds, triggering further downstream adjustments; this reinforces the flexibility channel.
  - Multi-sector menu cost model with Ss pricing, strategic complementarities, and IO linkages:
    - Generates peak interaction coefficient of −0.125, compared to −0.105 for externally-calibrated Calvo model and −0.113 in the data.
    - The two pricing frictions bracket the data from below and above, indicating cross-sectional pattern robustness to pricing-friction choice.
  - Both time-dependent and state-dependent models generate the hump-shaped interaction between upstreamness and monetary non-neutrality.
  - The menu cost model also provides a structural foundation for asymmetry: asymmetric Ss adjustment bands produce endogenously sign-dependent rigidity—firms are more willing to raise prices than to cut them.

*Source: wpiea2026127-source-pdf - 4.2    Asymmetric effects of expansionary and contractionary shocks*

### 5.3    Asymmetric network amplification

### 5.3    Asymmetric network amplification

### Empirical findings on asymmetry
- Expansionary interaction coefficient is negative and significant at medium-to-long horizons, reaching a peak of −0.25 about 27 months after the shock.
- Contractionary interaction coefficient is generally not significant and considerably noisier (standard errors on the contractionary coefficient are over 50% larger than on the expansionary side at earlier horizons).
- Wald tests reject symmetry at longer horizons (i.e., after 20 months) but not at earlier ones; averaging interaction coefficients across horizons yields a mean expansionary coefficient statistically different from the mean contractionary one over both h = 12–24 and h = 0–36.
- Cross-sectional evidence (cumulative 2-year price response vs upstreamness):
  - During expansionary episodes, a clear negative relationship: more upstream industries exhibit larger cumulative price changes.
  - During contractionary episodes, the relationship is flatter and noisier; the vertical axis in the plotted scatter is truncated at ±1.5 and three observations fall outside this range (two expansionary, one contractionary).

### Interpretation and mechanism
- Pattern is consistent with model prediction (Proposition 6): asymmetry is driven predominantly by expansionary shocks because upward Phillips curve slope λ+ is relatively steep while downward slope λ− < λ+ is flatter.
- When the Fed eases (demand rises):
  - Upstream prices—being more flexible—increase readily.
  - At each stage, the Phillips curve slope is λ+, producing a large cross-sectional differential between upstream and downstream industries.
- When the Fed tightens (demand falls):
  - Each stage faces the flatter λ−, attenuating the differential response.
- Proposition 6 implies the ratio of the expansionary to contractionary differential exceeds one and increases with number of production stages, so asymmetry is amplified along longer supply chains.
- Macro implication: prices rise more readily than they fall; asymmetric transmission helps explain why disinflation is costly while inflation can accelerate rapidly. Central banks face a harder task when tightening than when easing—a feature standard models ignoring production structure cannot capture.

### Structural-parameter back-of-the-envelope
- Under the asymmetric Calvo model, asymmetry ratio R_K (Equation 25) decomposes into own-flexibility factor r+ (1− r+) / [ r− (1− r−) ] and a network factor growing with effective chain length  ̄K.
- Empirical inputs:
  - Ratio of expansionary peak coefficient (at h = 27) to pooled peak (at h = 18): |ˆβh,−27| / |ˆβh18| ≈ 2.2.
  - Median upstreamness  ̄K ≈ 2 (network factor equals one at  ̄K = 2).
- Implied own-flexibility factor is roughly 2.2 (illustrative, since coefficients measured at different horizons).
- Solving r+ (1− r+) / [ r− (1− r−) ] = 2.2 suggests the probability of upward price adjustment is roughly twice the probability of downward adjustment, in line with central estimates in Peltzman (2000) and Nakamura and Steinsson (2008) who document ratios close to two.

*Source: wpiea2026127-source-pdf - 5.3    Asymmetric network amplification*

### 5.6    Input intensity and network neighbors

### 5.6    Input intensity and network neighbors

### Input share amplification
- Theory: The cost-propagation channel implies the upstreamness effect should be amplified for industries with higher intermediate input shares because these industries are more exposed to upstream cost pass-through.
- Empirical specification: Augment baseline local projection with a triple interaction:
  - ∆^h p_it = α^h_i + μ^h_s(i),t + β_1 (U_i × ε^m_t) + β_2 [U_i × α^II_i × ε^m_t] + β_3 [α^II_i × ε^m_t] + ρ^h ∆ p_{i,t−1} + u^h_it.  (Equation 32)
  - The lower-order interaction α^II_i × ε^m_t is included so that β̂_2 captures the marginal effect of input intensity conditional on the direct effect of intermediate input share.
- Testable implication: If cost pass-through operates through buyer-seller linkages, β̂_2 should be negative (higher input intensity amplifies transmission).
- Empirical finding:
  - Panel (a) of Figure 7: the triple interaction coefficient β̂_2 is generally negative and statistically significant during the second year after the shock.
  - Interpretation: Industries that are both upstream and input-intensive respond more strongly; amplification builds over medium horizons as cost pass-through accumulates through the production chain.

### Supplier-weighted upstreamness (network neighbors)
- Theory: An industry’s price response should depend not only on its own upstreamness but also on the upstreamness of its suppliers if network propagation operates through supplier linkages.
- Empirical measure:
  - Ū_supplier_i = ∑_k γ_ik U_k,  (Equation 33)
    - where γ_ik is the share of industry i’s intermediate inputs purchased from industry k.
  - Replace own upstreamness with Ū_supplier_i and interact with ε^m_t in the local projection.
- Empirical finding:
  - Panel (b) of Figure 7: the interaction between supplier-weighted upstreamness and monetary policy shocks is negative and statistically significant during the second year after the shock.
  - Interpretation: Supplier network position—a variable determined entirely by the IO structure—predicts own-industry price responses, providing evidence consistent with network propagation (rather than an omitted characteristic of individual sectors).

### Key quantitative evidence (from figures and tables referenced)
- Timing: Significant negative interaction effects generally appear during the second year after the shock (months roughly corresponding to 12–24 after the shock; Figure 7 panels plot months 0–36).
- Table 2 (18-month horizon, dependent variable: 18-month cumulative price change):
  - Network measure × MP shocks:
    - Upstreamness: −0.113***  (0.039)
    - Centrality (eigenvector): −0.384***  (0.131)
    - Intm. Share: −0.728***  (0.280)
    - JK MP Shocks: −0.153**  (0.059)
  - Observations: 69,772 for each column.
  - R^2: 0.003 (columns 1–3), 0.004 (column 4).
  - Notes: Driscoll-Kraay standard errors in parentheses. ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.10.
- Table 3 (18-month horizon, robustness to industry characteristics):
  - Upstreamness × MP shocks:
    - Baseline (col 1): −0.113***  (0.039)
    - With labor share control (col 2): −0.104***  (0.037)
    - With energy share control (col 3): −0.105***  (0.038)
    - With intermediate share control (col 4): −0.102***  (0.036)
    - With trade exposure control (col 5): −0.112***  (0.039)
    - All confounders included (col 6): −0.090***  (0.034)
  - Additional interaction estimates (selected):
    - Labor share × MP shocks: 0.619***  (0.232) in one specification; 0.468**  (0.205) in another.
    - Energy share × MP shocks: −0.153  (0.122); −0.148  (0.120) (not significant).
    - Intm. share × MP shocks: −0.537**  (0.227); −0.361*  (0.201).
    - Trade exposure × MP shocks: −0.034  (0.094); −0.041  (0.091).
  - Observations: 69,772 across columns. R^2 ranges 0.003–0.004.
  - Notes: All specifications include industry fixed effects, sector-time fixed effects, and a PPI inflation lag. Driscoll-Kraay standard errors in parentheses. ∗∗∗ p < 0.01; ∗∗ p < 0.05; ∗ p < 0.10.

### Robustness checks and extensions (selected)
- Lags/controls/functional form:
  - Adding richer lag structure, lags of monetary shocks, macro controls (CPI inflation, industrial production growth, change in unemployment, M2 growth) yields virtually unchanged results.
  - Using log(upstreamness) yields consistent results with linear specification.
- Alternative network measures and shock identification:
  - Results hold using eigenvector centrality and intermediate input share as network measures.
  - Results hold using Jarociński and Karadi (2020) monetary policy shocks (decomposes high-frequency FOMC surprises into pure monetary policy and central bank information components) on an extended sample through December 2024.
- Industry characteristic confounds:
  - Horse-race regressions including labor intensity, energy cost share, intermediate input dependence, and trade exposure (exports to total output at six-digit NAICS) do not eliminate the upstreamness interaction; coefficient remains significant across specifications.
  - Energy share interaction is not significant at any horizon; including it leaves upstreamness coefficient at roughly 80% of baseline magnitude at the peak horizon in a separate two-variable specification.
- Pre-trend (placebo) checks:
  - Interaction estimated at negative horizons (h = −12, . . . , −2): coefficients are small and statistically indistinguishable from zero (Table A9), suggesting no pre-trend and supporting causal interpretation.
  - Permutation placebo: randomly permuting upstreamness rankings 200 times, actual estimate lies far in the tail of permutation distribution at significant horizons (Figure A4), indicating the actual IO pattern drives results.
- Subsamples and outliers:
  - Manufacturing subsample (209 industries, NAICS 311–339) yields consistent results (Table A10).
  - Leave-one-out analysis across 271 industries: distribution of estimates at 18 months is tightly concentrated around baseline; no single industry (e.g., Petroleum and Coal Products) drives results (Figure A3).
  - Sub-sample stability (estimates at 18 months):
    - Pre-crisis (through 2010): −0.073 (statistically significant)
    - Post-crisis (2011 onward): −0.118 (statistically significant)
    - Excluding COVID period (through December 2019): −0.104 (statistically significant) (Table A11).
- Demand cyclicality:
  - Controlling for early-sample (1998–2002) PPI volatility interacted with the monetary shock (sample falls to 42,000 observations) leaves upstreamness interaction with the same sign but reduced magnitude at h = 18: −0.055 (compared to −0.113 baseline); volatility control not significant. Reduction attributed to smaller sample, not omitted demand cyclicality.

### Implications and interpretation
- Main conclusion: The production network is a first-order determinant of how monetary policy transmits to prices—more upstream industries exhibit significantly larger cumulative price responses to identified monetary shocks.
- Mechanism:
  - A multi-sector New Keynesian model with IO linkages rationalizes results: network position determines customer composition (B2B vs B2C), which determines price flexibility.
  - Counterfactual decomposition shows network-determined rigidity heterogeneity (not cost-cascade propagation) is the primary driver of the cross-sectional pattern.
- Asymmetry:
  - Over the one-to-two-year window, the upstreamness differential following expansionary shocks is large and negative while the contractionary differential is near zero.
  - Asymmetric price rigidity interacting with network-determined rigidity differential explains why inflation accelerates more easily than it decelerates.
- Policy relevance:
  - For central banks, the production network creates a structural asymmetry in transmission: rate cuts generate larger cross-sectional price dispersion than rate hikes.
  - Because price rigidity distribution is endogenous to network structure, average rigidity is not a sufficient statistic for monetary potency; tracking network evolution matters for inflation monitoring.
  - Two long-running structural shifts highlighted:
    - Rising services share of GDP moves the economy toward sectors that are more downstream and stickier, widening the upstream-downstream wedge.
    - Changes in global value chains (lengthening or re-shoring) alter the number of intermediate links between producers and consumers and thus the scope for asymmetric rigidity to compound stage by stage.

*Source: 5.6 Input intensity and network neighbors (excerpt from the provided PDF content).*

### References

### wpiea2026127-source-pdf - References

### Theoretical contributions and formal results
- Sectoral pricing and Phillips curves
  - Calvo pricing with sector-specific θi yields sectoral Phillips curve slope λi = (1− θi)(1− βθi)/θi.  
  - Real marginal cost (log-linearized) for sector i: cmcit = αi ŵt + Σk γik p̂kt − p̂it.
- Network-adjusted rigidity and aggregate non-neutrality
  - Aggregate gross-output price index P̂t = Σi ξi p̂it (Domar weights ξi = Yi/GDP).  
  - Output non-neutrality requires P̂t+h ≠ D̂t+h. Slow Domar-weighted price adjustment (high-θi sectors with large ξi) maximizes the gap D̂t+h − P̂t+h and thus aggregate output non-neutrality.
  - Corollary (alignment): sectors with higher θi (more rigid) yield flatter λi and slower aggregate price adjustment when they carry large Domar weights.
- Upstreamness, IO structure, and price responses
  - Static factorization (β = 0): p̂t = (I − RΓ)−1 R α ŵt with R = diag(r1,..., rN), ri = λi/(1 + λi). Every sector’s price moves in same sign as wage: p̂it/ŵt > 0 for all i.
  - Two channels linking upstreamness Ui to |p̂it|: (1) flexibility channel (ri increasing in Ui under Assumption 1); (2) input-cost channel via γik and αi with ambiguous sign depending on data distribution. In calibrated U.S. data Corr(Ui, αi) = −0.90 (upstream sectors have lower labor shares), so input-cost channel partially offsets flexibility channel; flexibility channel is quantitatively dominant.
  - Chain-economy closed form (uniform r, chain with γ): xk ≡ p̂kt/ŵt solves xk = r[(1− γ) + γ xk−1]; solution xk = r[(1− γ) + γ(1− r)φk−1]/(1− φ) with φ ≡ rγ. Upstream-downstream differential DK = x1 − xK = rγ(1− r)[1 − φK−1]/(1 − φ).
  - Delayed peak output effect: downstream sectors (smaller |p̂kt|) have larger and later output gaps ŷkt ≈ d̂kt − p̂kt; peak real effect occurs later for downstream sectors.
- Sign-dependent and asymmetric propagation
  - Under state-dependent rigidities rsi (s ∈ {+, −}), matrix Ms = (I − Rs Γ)−1 Rs exists (Neumann series) and is non-decreasing in each rsi. If r+ i ≥ r− i for all i, then |p̂+ i | ≥ |p̂− i | for every sector i.
  - Chain economy asymmetry: DsK ≡ x s 1 − x s K = rs γ(1 − rs)[1 − (φs)K−1]/(1 − φs). Ratio RK ≡ D+ K /D− K = [r+(1− r+)]/[r−(1− r−)] × [ΣK−2 j=0 (φ+) j]/[ΣK−2 j=0 (φ−) j] (γ cancels). For r+, r− < 1/2 with r+ > r−: RK > 1, G(K) (network factor) strictly increasing in K, finite limit R∞ given by formula in text.
- Network Phillips curve (aggregate)
  - Sectoral system: πt = βEt πt+1 + Λ[A ŵt 1 + (Γ − I) p̂t]. Aggregate inflation πagg t = ξ′ πt. In static β = 0 case, p̂t = [I − (I − Λ) Γ]−1 Λα x t 1.
  - Aggregate Phillips curve: πagg t = βEt πagg t+1 + κnet xt with κnet = ξ′[I − (I − Λ)Γ]−1 Λα.
  - Properties: (i) κnet = κstandard when Γ = 0, where κstandard = ξ′ Λα; (ii) κnet < κstandard when Domar-weighted upstream sectors are more rigid (high ξi and high Ui); (iii) ∂κnet/∂ρ(Γ) < 0 (denser IO network reduces aggregate Phillips slope).
  - Policy implication (Remark 1): a longer, more interconnected supply chain flattens the aggregate Phillips curve, implying monetary policy must move interest rates by more to achieve a given change in inflation.

### Menu-cost model extension and mechanism robustness
- Model setup and strategic complementarities
  - Firms face menu cost Fi and follow asymmetric Ss rules with upper/lower bands y∗i ± bu i /bℓ i; desired prices incorporate strategic complementarity ψ: p∗ijt = (1 − ψ) mcijt + ψ Pit.
  - Log nominal marginal cost: mcijt = αi wt + Σk γik Pkt − zijt; idiosyncratic zijt follows random walk with volatility σz.
- Calibration (Table A1)
  - σz = 0.04 (Idiosyncratic volatility, monthly)  
  - μ = 0.002 (Trend inflation, monthly, ≈ 2.4% annual)  
  - demand path hump/trough h = 17 (Nominal demand impulse)  
  - ψ = 0.55 (Strategic complementarity)  
  - ̄θ = 0.85 (Avg. Calvo parameter for Ss calibration)  
  - δ = −0.152 (Upstreamness–flexibility slope)  
  - IO matrix from BEA Use Table aggregated to N = 36 summary sectors
- Key menu-cost model results (Figure A1 and text)
  - Peak interaction coefficients (horizon and magnitudes):
    - Menu cost model peak: −0.125 at h = 15.
    - Calvo model (externally calibrated) peak: −0.105 (timing in main text).
    - Empirical peak: −0.113 (data).
  - Mechanisms raising menu-cost peak relative to Calvo:
    - Endogenous rigidity dispersion (extensive-margin): upstream sectors adjust more frequently even with identical menu costs because they face larger, more volatile cost shocks.
    - Cascade amplification (extensive-margin selection): upstream adjustments propagate through IO matrix and can trigger downstream firms to cross Ss thresholds—absent in Calvo (where fraction adjusting is fixed = 1 − θi).
  - Quantitative fit: Calvo model captures 93% of empirical peak and matches timing; menu-cost model raises peak modestly and supports asymmetry results but lacks closed-form tractability.
  - Both frameworks yield a hump-shaped interaction coefficient peaking at medium horizons and reverting toward zero—robust feature across pricing-friction specifications.

### Empirical findings, magnitudes, and statistical evidence
- Baseline interaction estimates (Table A3, Panel A)
  - Interaction coefficient Upstreamness × MP shocks by horizon h (months):  
    - h = 0: −0.013 (SE 0.010)  
    - h = 3: −0.039 (0.054)  
    - h = 6: −0.066 (0.054)  
    - h = 9: −0.059 (0.047)  
    - h = 12: −0.101*** (0.038)  
    - h = 15: −0.105** (0.041)  
    - h = 18: −0.113*** (0.039)  
    - h = 21: −0.081** (0.039)  
    - h = 24: −0.093** (0.043)  
    - h = 27: −0.097** (0.048)  
    - h = 30: −0.035 (0.037)  
    - h = 33: 0.012 (0.036)  
    - h = 36: 0.019 (0.039)
  - Panel B: cumulative PPI responses by upstreamness quartile show most upstream (Q4) sectors have largest negative responses (e.g., at h = 18: −0.127*** (0.044) for Q4).
- Asymmetric specification (Table A3, Panel C)
  - Upstreamness × MP shocks × expansion dummy and × contraction dummy show large negative coefficients concentrated at medium horizons for expansion/contraction splits; sample values provided in table (e.g., Upstr.×MP×exp. at h = 18: −0.158* (0.093); Upstr.×MP×contr. at h = 0: −0.048** (0.021)).
- Real effects on industrial production (Table A4)
  - Symmetric Upstreamness × MP shocks on cumulative IP growth by horizon: e.g., h = 15: 0.041** (0.017); h = 18: 0.042** (0.017); h = 24: 0.047*** (0.017).
  - Asymmetric effects: Upstr.×MP×contr. shows strong positive IP effects at medium horizons (e.g., h = 18: 0.120*** (0.040)).
- Supply-chain cascade and input-share amplification (Tables A5–A6)
  - Supplier price is a large predictor: Supplier price coefficients significant and increasing across horizons (Panel A: 0.176*** at h = 0 up to 0.739*** at h = 18, with standard errors reported per horizon).
  - Triple interaction Upstr. × intermediate input share × MP shock (Table A6) shows significant negative coefficients at medium horizons (e.g., h = 18: −0.591*** (0.221)), indicating amplification where intermediate input shares are higher.
  - Supplier-weighted upstreamness interaction (Table A7) significant and negative at medium horizons (e.g., h = 15: −0.074** (0.031); h = 18: −0.079** (0.031)).
- Data summary statistics (Table A2)
  - Sample covers 271 six-digit NAICS industries over 1998:02–2023:12 (Jarociński-Karadi extendable to 2024:12). Selected statistics:
    - Monthly PPI inflation (N = 75,073): mean 0.002, Std. Dev. 0.018, P10 −0.006, P50 0.001, P90 0.013.
    - Bauer-Swanson orthogonalized MP surprises (N = 71,857): mean 0.003, Std. Dev. 0.049, P10 −0.038, P50 0.000, P90 0.053.
    - Upstreamness (N = 75,073): mean 1.222, Std. Dev. 0.352, P10 1.009, P50 1.103, P90 1.566.
    - Eigenvector centrality mean 0.016, Std. Dev. 0.067.
- Robustness and specification checks
  - Specification variants (Table A8) across lag structures, controls, and log transformations yield consistent interaction patterns; baseline results robust to these changes.
  - Pre-trend/placebo/leave-one-out tests:
    - Pre-trend coefficients at negative horizons (Table A9) show no systematic pre-trend (coefficients small and statistically insignificant across h = −12,…, −1).
    - Permutation placebo: 200 permutations of upstreamness show baseline estimates lie in extreme left tail of placebo distributions at key horizons (Figure A4).
    - Leave-one-out distributions concentrated closely around baseline estimates at key horizons (Figure A3).
  - Sub-sample stability (Table A11):
    - Interaction coefficient at h = 18 across sub-periods: Full sample −0.113*** (0.039); up to GFC −0.083* (0.048); post-GFC −0.118** (0.060); excl. COVID −0.103*** (0.037).
  - Manufacturing-only estimates (Table A10) qualitatively similar and in some cases larger magnitudes (e.g., symmetric Upstreamness × MP shocks at h = 18: −0.151*** (0.052)).

### Data construction and shock measures
- PPI data: industry-level PPI series from BLS via FRED; PPI ID convention PCU{NAICS}{NAICS} with six-digit NAICS codes. Estimation sample: 209 manufacturing (NAICS 311–339) plus 62 non-manufacturing industries; final matched sample 271 industries with PPI and IO measures.
- IO matrices and upstreamness:
  - BEA Detailed Use table matched to six-digit NAICS; output-allocation matrix A computed by normalizing each Use row by industry total output (aik = Zik/Yi). Ghosh inverse (I − A)−1 row sums yield upstreamness. Correlation of upstreamness with eigenvector centrality = 43%.
- Monetary policy shocks:
  - Baseline shock: Bauer and Swanson (2023) FOMC-meeting-level high-frequency surprises orthogonalized (“MPS_ORTH”), aggregated to monthly frequency by summing surprises within calendar months; months without meetings assigned zero. Series covers 1988–2023.
  - Alternative shock: Jarociński and Karadi (2020) information-purged shocks available up to end-2024; correlation ρ = 0.76 with Bauer-Swanson series over overlap.

### Key policy takeaways and implications
- Network amplification flattens the aggregate Phillips curve: denser IO networks and rigid upstream sectors reduce κnet relative to a no-network benchmark, implying monetary policy must enact larger interest-rate moves to achieve inflation targets.
- Upstream sectors exhibit stronger and earlier PPI responses to monetary policy surprises; downstream sectors experience later and more persistent real effects—policy assessments should account for cross-sector timing heterogeneity.
- State-dependent pricing (menu-cost Ss) introduces endogenous rigidity dispersion and extensive-margin cascades that amplify network effects beyond Calvo-based predictions; these mechanisms imply asymmetric propagation and larger responses during expansions versus contractions under plausible parameterizations.

*References and appendix material as presented in the source document.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026127-source-pdf.pdf_
