## wpiea2022017-print-pdf

## Source details

**Canonical URL:** [wpiea2022017-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022017-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022017-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022017-print-pdf.pdf.json)

---

### Interest rate channel
- Mechanism:
  - A monetary contraction (increase in the short-term interest rate) pushes up longer-term rates through the expectations-hypothesis of the term structure.
  - With sticky prices, this raises the real interest rate and leads firms and consumers to cut investment and durable purchases, depressing output and prices.

### Credit channel
- Conceptual core and mechanism:
  - Monetary contraction reduces firms’ net worth, making borrowers less able to pledge collateral and increasing agency problems.
  - Lenders respond with a higher external financing premium, reducing investment and output (Bernanke and Gertler (1989); Bernanke, Gertler, and Gilchrist (1999)).
- Empirical proxies and findings:
  - Proxies for financial constraints: firm size, age, leverage, dividend payout, asset tangibility, etc.
  - Empirical observations cited:
    - Unsecured debt is strongly procyclical in U.S. data (Azariadis, Kaas, and Wen, 2016).
    - The unsecured credit spread goes up in recessions (Benmelech, Kumar, and Rajan, 2020).
    - “Excess sensitivity” of financially constrained firms to monetary shocks documented (Bernanke, Gertler, and Gilchrist (1999, pp. 1374-5)).
  - Channel delineation:
    - Balance sheet (borrower) vs. bank lending (lender) views are not separately identified in the paper due to lack of bank lending data; focus is on the broad credit channel.
  - Modeling implication:
    - Observed procyclicality of unsecured debt and flight to quality motivate models with both secured and unsecured financing.

### Exchange rate channel
- Mechanisms and pricing regimes:
  - Monetary contractions typically appreciate the home currency, potentially reducing net exports (Taylor, 1995).
  - If prices set in producer’s currency: appreciation contracts output more in export-dependent industries.
  - If many tradables are priced in U.S. dollars (dominant currency pricing): home-currency appreciation has little direct impact on external demand for exports; net exports may fall mainly via higher imports (Gopinath et al., 2020).

### The cost channel
- Concept and distinctive prediction:
  - Firms borrowing working capital face the cost of borrowing as an input to production; higher interest rates act like adverse cost-push shocks.
  - Prediction: contractionary monetary shock can increase prices for products of firms that rely more heavily on external financing (Ravenna and Walsh, 2006).
- Paper objective (differential industry approach):
  - Use industry heterogeneity to identify which transmission channels matter by estimating industry-specific responses to monetary policy shocks.

### Econometric specification and identification
- Panel: unbalanced panel of 105 countries and 22 manufacturing industries over 1973-2019.
- Baseline specifications:
  - Output growth:
    - Y_{i,c,t+1} = α_{i,c} + α_{i,t} + α_{c,t} + β (X_i × MPShock_{c,t}) + γ Mshare_{i,c,t} + ε_{i,c,t+1}
  - Price-deflator analogue:
    - π_{i,c,t+1} = α_{i,c} + α_{i,t} + α_{c,t} + β (X_i × MPShock_{c,t}) + γ π_{i,c,t} + ε_{i,c,t+1}
- Identification:
  - Main parameter β interpretable as ∂^2 Y_{i,c,t+1} / ∂ X_i ∂ MPShock_{c,t}; β < 0 means larger negative output response in industries scoring higher on X.
  - Fixed effects: α_{i,c} (industry-country), α_{i,t} (industry-time), α_{c,t} (country-time).
  - Standard errors clustered at the country-by-time treatment level.

### Hierarchical construction of monetary policy shock series (MPShock_{c,t})
- Prioritized sources (positive values correspond to monetary contractions):
  i. High-frequency (Kuttner-style) shocks where available (selected studies for U.S., eurozone, U.K., Canada, Norway, Sweden, Brazil, India, Japan, Korea).
  ii. One-day change in 3-month swap yield around policy decision days (Bloomberg).
  iii. One-day change in short-term domestic government bond yield around decision days (1, 3, 6, or 12-month tenors).
  iv. Bloomberg survey: realized policy rate minus prior expectation.
  v. Residuals from country-by-country Taylor rule OLS when i)-iv) unavailable:
    - ∆R_t = θ + Σ_{j=1}^3 κ_j g_{t-j/3} + Σ_{j=1}^3 λ_j π_{t-j/3} + Σ_{j=1}^3 τ_j ∆a_{t-j/3} + Σ_{j=1}^3 ρ_j ∆R_{t-j/3} + ε_{t}^R
    - Evidence: correlation of 0.43 between U.S. high-frequency shocks and U.S. Taylor residuals from (2).
  vi. If country pegged exchange rate to an anchor currency and i)-v) unavailable: use anchor-country shock scaled by Chinn-Ito capital account openness (0 to 1).
- Proxies i)-iv) calculated on announcement days and annualized by cumulation to match annual dependent variables.
- Sourcing composition (Table 1 reallocated to original sources):
  - High-frequency studies: Share (I) 8.8% ; Share (II) 57.8%
  - Change in swap yields: Share (I) 5.4% ; Share (II) 5.4%
  - Change in bond yields: Share (I) 1.7% ; Share (II) 1.7%
  - Survey-based measure: Share (I) 2.2% ; Share (II) 2.2%
  - Taylor residuals: Share (I) 23.3% ; Share (II) 32.9%
  - Taken from anchor countries: 58.6% (breakdown: High-frequency studies 49.0%; Taylor residuals 9.6%)
  - Aggregate note: almost 60% of all 5,433 observations originate from pre-existing high-frequency studies; one third are estimated Taylor residuals.

### Validation and robustness
- Panel VAR using constructed shock series yields puzzle-free impulse responses for cyclical real GDP and GDP deflator (Figure 1).
  - IRFs from panel fixed effect estimator of Cagala and Glogowsky (2014).
  - Cyclical components via Hamilton (2018) filter at country level.
  - Shock series used as instrument in recursive VAR following Plagborg-Møller and Wolf (2021).
- Combination of hierarchical shocks and triple fixed effects mitigates reverse causality and many omitted variable concerns.

### Industry characteristics used to probe channels
- Eight measures (normalized to zero mean and unit standard deviation across industries): External financial dependence (EFD), Asset tangibility (TAN), Investment intensity (INV), Labor intensity (LAB), Liquidity needs (LIQ), Capital depreciation (DEP), Durability dummy (DUR), Export intensity (EXP).
- Definitions and expected empirical implications:
  - EFD: share of capital expenditure not financed by cash flow (median U.S. firm). Credit channel: expected negative interaction with MPShock on output; cost channel (price-deflator): expected positive interaction.
  - TAN: share of tangible capital; higher TAN implies easier access to secured funding; expected positive interaction if credit-sheet channel important.
  - INV: gross investment/value-added; interest rate channel predicts vulnerability in high INV industries; collateral channel predicts resilience.
  - LAB: wages/value-added; tests both credit and cost channels—labor-intensive industries may face borrowing limitations and higher working-capital-driven cost pressures.
  - LIQ: inventories/sales from Raddatz (2006); proxies short-term working capital reliance; informative for cost channel.
  - DEP: industry depreciation rates from BEA; less durable capital reduces collateral pledgeability.
  - DUR: dummy = 1 if industry produces durable goods (12 of 22 industries); conventional interest channel predicts stronger effects for durable industries.
  - EXP: exports/value-added from Giovanni and Levchenko (2009); tests exchange rate channel (expected sign β: −).

### Export intensity (EXP): measurement and empirical findings
- Measurement and sample:
  - EXP from Giovanni and Levchenko (2009); industry-level averages of exports/value-added.
  - Industry growth from UNIDO covering 153 countries; baseline uses 22 manufacturing industries at two-digit INDSTAT2 2021 (ISIC Rev. 3).
  - Data processing: current local currencies deflated with Consumer Price Indices (Global Inflation Database); require ≥ 10 years consecutive data per industry; winsorize top/bottom 1% of growth variables.
  - Price-deflator sample is 30 percent smaller due to production index coverage; KLEMS used in alternative exercises.
  - Cross-country sample for shock proxies: 105 countries (33 advanced economies and 72 EMDEs).
  - U.S. excluded from regressions per Rajan-Zingales convention.
- Correlations (Table 3, exact values):
  - EXP with EFD: 0.338
  - EXP with TAN: -0.369
  - EXP with INV: -0.250
  - EXP with LAB: 0.278
  - EXP with LIQ: 0.239
  - EXP with DEP: 0.232
  - EXP with DUR: 0.396
- Baseline empirical findings (Table 4 summary):
  - Among eight characteristics, five (TAN, INV, LAB, DEP, DUR) are statistically significant in baseline OLS; EFD, LIQ, and EXP are insignificant.
  - Interpretation: EXP insignificance suggests industries more reliant on exports do not contract more after monetary tightening.
  - Robustness: EXP remains insignificant when restricting to floating-currency countries.
  - Dominant currency pricing interpretation: exchange rate channel may operate mainly via imports rather than exports (Gopinath et al., 2020).
- IV (2SLS) results:
  - Using composite instrument interacted with industry characteristics yields similar second-stage estimates to OLS; EXP becomes significant only at the 10% level.
- Additional robustness:
  - Inflation-targeting subsample, placebo/lead tests, and KLEMS-based price-deflator exercises do not provide robust support for EXP; cost channel not supported.

### Key empirical conclusions (from appendix and main results)
- Across sensitivity checks:
  - Credit channel is the most robust transmission channel, including stronger effects during downturns.
  - Interest rate channel is the next most important (durable-goods industries more affected).
  - No empirical support for the cost channel.
  - No robust evidence for an exchange rate channel operating via exports; any exchange rate effects likely operate through imports (consistent with dominant currency pricing).
- Credit-channel amplification and state dependence:
  - Credit channel stronger during downturns, tightened credit conditions, in less-developed financial markets, and during crises.
  - “Ability to secure external financing” (access to collateral) is more relevant than the “need for external financing.”
  - Empirical patterns highlight cyclical unsecured debt and flight to quality into secured lending in slowdowns, challenging models that rely solely on secured debt amplification.

### Selected key estimates (preserving reported values)
- Table 4 (Baseline: dependent = real value-added growth)
  - Monetary shock coefficient: -1.846*** (0.082)
  - Interaction ITI × MPShock: 0.110** (0.045)
  - Interaction LTL × MPShock: -0.157** (0.062)
  - Interaction EEMM × MPShock: -0.114** (0.056)
  - Interaction EEUR × MPShock: -0.082** (0.034)
  - R-squared: 0.305
  - Observations: 44,191
- Table 5 (Baseline: dependent = price index growth)
  - Baseline price coefficient: -0.158*** (0.013)
  - Interaction LTL × MPShock: -0.117* (0.061)
  - Interaction EEMM × MPShock: -0.059* (0.031)
  - R-squared: 0.346–0.347
  - Observations: 30,582
- Table A.4 (IV; dependent = real value-added growth)
  - Baseline monetary shock (푀ℎ푎푎푎푎푎푎): -2.024*** to -2.027*** with standard errors (0.111)–(0.112)
  - Selected interaction examples: LTL × MP: -0.126*** (0.044); EEMM × MP: -0.112*** (0.028)
  - F-statistics range reported (example): 5,358.595 to 6,662.989
  - Observations: 31,109
- Tables on state dependence and financial development (selected):
  - Table 7 monetary shock: -1.962*** (0.087); R-squared: 0.307; Observations: 42,   031
  - Table 8 monetary shock: -1.920*** to -1.918*** (0.083); interaction EEMM × MPShock: -0.252*** (0.053); R-squared: 0.306; Observations: 42,737
  - Table 9 monetary shock: -1.847*** to -1.846*** (0.082); Observations: 44,191
- Note: Standard errors clustered at country-time; *, **, *** denote significance at 10, 5, and 1 percent respectively.

### Data coverage and selected industry characteristic values (Table A.2 examples)
- Sample inclusion rule: industries with > 10 years of data.
- Country sample examples: Australia (Number of industries 22, Period 1975-2019, Group ADV); Brazil (22, 1996-2019, EMDE); China (22, 1986-2018, EMDE); India (22, 1974-2019, EMDE); Japan (22, 1990-2018, ADV); United Kingdom (22, 1974-2019, ADV).
- Selected industry indices (exact values preserved):
  - ISIC 15 Food products and beverages: EFD 0.110, TAN 0.373, INV 0.065, LAB 0.277, LIQ 0.100, DEP 7.090, DUR 0, EXP 0.212
  - ISIC 24 Chemicals and chemical products: EFD 0.500, TAN 0.287, INV 0.094, LAB 0.229, LIQ 0.145, DEP 8.154, DUR 0, EXP 0.401
  - ISIC 29 Machinery and equipment n.e.c.: EFD 0.600, TAN 0.195, INV 0.059, LAB 0.433, LIQ 0.170, DEP 8.832, DUR 1, EXP 3.878
  - ISIC 30 Office, accounting and computing machinery: EFD 0.960, TAN 0.208, INV 0.055, LAB 0.407, LIQ 0.200, DEP 9.381, DUR 1, EXP 0.484
  - ISIC 34 Motor vehicles, trailers and semi-trailers: EFD 0.360, TAN 0.264, INV 0.073, LAB 0.440, LIQ 0.180, DEP 10.559, DUR 1, EXP 1.499

*wpiea2022017-print-pdf*

### 1.   The interest rate channel. This channel is typically seen as the main “Keynesian” transmission

### 1.   The interest rate channel. This channel is typically seen as the main “Keynesian” transmission

### Interest rate channel
- Mechanism:
  - A monetary contraction (increase in the short-term interest rate) pushes up longer-term rates through the expectations-hypothesis of the term structure.
  - With prices being sticky, this raises the real interest rate.
  - Firms and consumers respond to the higher cost of capital by cutting back on investment and durable purchases, which depresses output and prices.

### Credit channel
- Conceptual core:
  - Associated with Bernanke and Gertler (1989) and Bernanke, Gertler, and Gilchrist (1999).
  - A monetary contraction reduces firms’ net worth (the sum of liquid assets and marketable collateral, less outstanding obligations), making borrowers less able to pledge collateral.
  - Reduced collateral increases agency problems: firm owners have less “skin in the game,” raising the incentive for risky investments not aligned with lenders’ interests.
  - Lenders compensate via a higher interest rate (the “external financing premium”), reducing firm investment and output.
- Empirical approach and proxies:
  - The external financing premium is hard to observe directly; literature relies on proxies for financial constraints (firm size, age, leverage, dividend payout, etc.).
  - Firms that face greater difficulties in pledging collateral (e.g., smaller size, nature of assets) are more vulnerable to this mechanism (Gertler and Gilchrist, 1994).
  - Downturns induce creditors to “flee to quality” (credit flowing away from borrowers without much collateral; Bernanke, Gertler, and Gilchrist, 1996).
  - Empirical findings cited:
    - Unsecured debt is found to be strongly procyclical in U.S. data (Azariadis, Kaas, and Wen, 2016).
    - The unsecured credit spread goes up in recessions (Benmelech, Kumar, and Rajan, 2020).
  - Firms with fewer collateralizable assets are expected to be more sensitive to monetary contractions and economic slowdowns; Bernanke, Gertler, and Gilchrist (1999, pp. 1374-5) describe this as “excess sensitivity” to monetary shocks for more financially constrained firms.
- Channel delineation:
  - Occasionally separated into the balance sheet channel (borrower’s side) and the bank lending channel (lender’s side).
  - The authors do not separately identify these two channels due to lack of bank lending data.
  - Quoted observation: “In practice, the distinction between the balance sheet and the bank lending view becomes blurred when the correlation between dependence on external funds and dependence on bank loans is high, or when banks are the predominant source of external finance.” (Braun and Larrain (2005: 1102))
  - Rationale: Many emerging/developing economies in the sample do not have well-developed corporate bond markets like the United States, so focusing on the broad credit channel is a reasonable approach.
- Modeling implications:
  - Observed patterns (procyclicality of unsecured debt, flight to quality) challenge Kiyotaki and Moore (1997)-style models that emphasize secured debt as the main amplifier.
  - These observations motivate models featuring both secured and unsecured financing (e.g., Azariadis, Kaas, and Wen (2016) and Luk and Zheng (2022)).

### Exchange rate channel
- Mechanism:
  - Monetary contractions typically appreciate the home currency, which can reduce net exports and aggregate demand (Taylor, 1995).
  - The impact varies with the currency in which prices are set.
- Pricing in producer’s currency:
  - If prices are set in the producer’s (exporter’s) currency, depreciation makes the exporter’s good cheaper for importers, so monetary tightening (appreciation) should contract output more in more export-dependent industries.
- Pricing in U.S. dollars:
  - Many traded goods are priced in U.S. dollars (even when exporter and importer are countries where the U.S. dollar is not legal tender; Gopinath et al., 2020).
  - In that case, appreciation of the home currency has no direct impact on external demand for home exports, because the exchange rate between importer currencies and the U.S. dollar is not affected.
  - Net exports may still fall, but mainly through higher imports rather than through a loss of external demand for exports.

*Italic source: wpiea2022017-print-pdf - 1.   The interest rate channel. This channel is typically seen as the main “Keynesian” transmission*

### 4. The cost channel. The last theoretical channel we investigate is the cost channel of monetary

### 4. The cost channel

### Concept and implications
- The cost channel: when firms pay factors of production (wages, inventories, etc.) before receiving sales revenue, they borrow working capital; the cost of borrowing becomes an input to production, so interest rate increases act like adverse cost-push shocks.
- Distinctive prediction: a contractionary monetary shock will increase prices for products produced by firms that rely more heavily on external financing.
- Policy relevance: a flipped (price-increasing) response of prices to monetary shocks has important implications for optimal monetary policy (Ravenna and Walsh, 2006).

### Objective of the paper (differential industry approach)
- Aim: analyze industry-specific responses to monetary policy shocks to shed light on the importance of transmission channels (cost/credit/other channels).
- Rationale: channels differ across industries depending on industry-specific characteristics; a “differential” approach using industry heterogeneity can uncover which channels matter.

### Econometric specification and identification
- Estimated panel specification for an unbalanced panel of 105 countries and 22 manufacturing industries over the period 1973-2019:
  - Baseline (output growth):
    - Y_{i,c,t+1} = α_{i,c} + α_{i,t} + α_{c,t} + β (X_i × MPShock_{c,t}) + γ Mshare_{i,c,t} + ε_{i,c,t+1}
  - Price-deflator analogue (testing cost channel):
    - π_{i,c,t+1} = α_{i,c} + α_{i,t} + α_{c,t} + β (X_i × MPShock_{c,t}) + γ π_{i,c,t} + ε_{i,c,t+1}
- Main parameter of interest: β (coefficient on X_i × MPShock_{c,t}), interpreted as a difference-in-differences measuring the differential impact of monetary contractions in industries with characteristic X_i.
  - Differentiation yields β = ∂^2 Y_{i,c,t+1} / ∂ X_i ∂ MPShock_{c,t}.
  - When β < 0: a monetary contraction (MPShock > 0) has a larger negative effect on output growth in industries scoring higher on X.
- Fixed effects included:
  - industry-country fixed effects (α_{i,c}): control for persistent industry-in-country factors.
  - industry-time fixed effects (α_{i,t}): control for global industry-specific shocks (e.g., oil price shocks affecting certain industries).
  - country-time fixed effects (α_{c,t}): control for all macro developments affecting country c in year t (including aggregate policy effects).
- Standard errors clustered at the treatment level (country by time), following Abadie et al. (2017).

### Endogeneity concerns and hierarchical shock identification
- Challenge: monetary policy is endogenous; changes in policy may be responses to macro developments that differentially affect industries.
- Mitigation:
  - The interaction approach combined with the triple fixed effects reduces reverse causality concerns (it would be implausible that monetary policy is set to target cross-industry growth differentials).
  - Further effort to purge endogeneity by constructing a monetary policy shock series via a hierarchical approach (i-vi), prioritizing high-frequency identification where available.
- Hierarchical approach to obtain MPShock_{c,t} (positive values correspond to monetary contractions):
  i. Where available, take shocks from high-frequency studies (Kuttner-style). Examples used: Bauer and Swanson (2022) for the U.S. (1988-2019); Jarociński and Karadi (2020) for the eurozone (1999-2016); Cesa-Bianchi, Thwaites, and Vicondoa (2020) for the U.K. (1997-2015); Champagne and Sekkel (2018) for Canada (1974-2015); Holm, Paul, and Tischbirek (2021) for Norway (1990-2018); Amberg et al. (2022) for Sweden (1999-2018); Alberola et al. (2021) for Brazil (2001-2017); Lakdawala and Sengupta (2021) for India (2003-2020); Kubota and Shintani (2022) for Japan (1992-2020); Ahn, Kim, and Lee (2021) for Korea (1990-2018).
  ii. If i) not available, proxy by the one-day change in the 3-month swap yield (Bloomberg) around monetary policy decision days (yield close of day T minus yield close of day T-1).
  iii. If i) and ii) not available, proxy by the one-day change in the short-term domestic government bond yield around decision days (tenors with original maturity less than a year: 1, 3, 6, or 12-month).
  iv. If i)-iii) not available, use Bloomberg’s survey of financial market participants: shock = realized policy rate minus prior expectation.
  v. If i)-iv) not available, use residuals from an estimated Taylor rule (country-by-country OLS) as proxy for shocks:
    - Estimated rule:
      ∆R_t = θ + Σ_{j=1}^3 κ_j g_{t-j/3} + Σ_{j=1}^3 λ_j π_{t-j/3} + Σ_{j=1}^3 τ_j ∆a_{t-j/3} + Σ_{j=1}^3 ρ_j ∆R_{t-j/3} + ε_{t}^R
    - Variables: ∆R_t (change in policy rate), g_t (real growth), π_t (inflation), ∆a_t (change in logged nominal exchange rate vs. U.S. dollar). Three lags included for all variables.
    - Justification: Taylor residuals can approximate shocks in data-scarce developing countries; evidence: correlation of 0.43 between U.S. high-frequency shocks and U.S. Taylor residuals obtained from (2).
  vi. If i)-v) not available and the country pegged its exchange rate to an anchor currency, use the estimated shock in the anchor country scaled by the Chinn-Ito measure of capital account openness (0 to 1) to reflect imported monetary policy strength.
- Proxies i)-iv) are calculated on announcement days and annualized by cumulation to match annual dependent variables.

### Sourcing and composition of the shock series
- Table 1 (sourcing of monetary policy shocks) as presented:
  - High-frequency studies: Share (I) 8.8% ; Share (II) 57.8%
  - Change in swap yields: Share (I) 5.4% ; Share (II) 5.4%
  - Change in bond yields: Share (I) 1.7% ; Share (II) 1.7%
  - Survey-based measure: Share (I) 2.2% ; Share (II) 2.2%
  - Taylor residuals: Share (I) 23.3% ; Share (II) 32.9%
  - Taken from anchor countries: 58.6%
    - High-frequency studies: 49.0%
    - Change in swap yields: 0.0%
    - Change in bond yields: 0.0%
    - Survey-based measure: 0.0%
    - Taylor residuals: 9.6%
    - n/a
- Aggregate: when shocks are reallocated to original sources (Column (II)), almost 60% of all 5,433 observations originate from pre-existing high-frequency studies; only a third of shocks are estimated Taylor residuals.

### Validation and robustness checks
- Using the constructed monetary policy shock series in a panel VAR yields puzzle-free impulse responses for the cyclical components of real GDP and the GDP deflator (Figure 1), supporting the credibility of the shock series.
  - IRFs are generated using the panel fixed effect estimator of Cagala and Glogowsky (2014).
  - Cyclical components obtained by applying the Hamilton (2018) filter at the country level.
  - The monetary policy shock series is used as an instrument for the true shock in a recursive VAR (instrument ordered first), following Plagborg-Møller and Wolf (2021).
- The paper emphasizes that, despite identification difficulties even with high-frequency methods, the combination of hierarchical shock construction and triple fixed effects (industry-country, industry-time, country-time) mitigates concerns about reverse causality and many forms of omitted variable bias.

### Data on industry characteristics (setup for subsequent analysis)
- Industry characteristics: eight measures (external financial dependence, asset tangibility, durability of output, etc.) mostly from Samaniego and Sun (2015), constructed at three-digit ISIC using U.S. firm-level data and aggregated to two-digit ISIC using industry-level average value-added as weights (per Choi, Furceri, and Jalles (2022)).
- Rationale: U.S.-based industry characteristics serve as technological benchmarks in a relatively frictionless environment and help mitigate reverse causality concerns (industry-country fixed effects capture persistent cross-country industry differences).

*Source: wpiea2022017-print-pdf — 4. The cost channel.*

### 1.    External  financial  dependence  (EFD).  Following  Rajan  and  Zingales  (1998),  dependence  on

### 1.    External  financial  dependence  (EFD).  Following  Rajan  and  Zingales  (1998),  dependence  on

### Industry characteristics and their expected roles in monetary transmission
- External financial dependence (EFD)
  - Definition: Dependence on external finance in each industry is proxied by that share of capital expenditure that is not financed by cash flow from operations; the industry value is that of the median U.S. firm in each industry.
  - Use: Captures firms’ need for external financing and is employed, inter alia, by Dedola and Lippi (2005) to test the credit channel of monetary policy.
  - Expected empirical implication:
    - Credit channel: negative sign on the interaction term between external financial dependence and the monetary policy shock (firms relying more heavily on external funding are more vulnerable to an increase in the premium driven by a monetary tightening).
    - Cost channel (when dependent variable is the price deflator): positive interaction term (firms with heavy reliance on external funding are more likely to raise the prices of their products following a monetary contraction).

- Asset tangibility (TAN)
  - Definition: Share of tangible capital in a firm’s total assets; proxies fraction of firm assets that can be pledged as collateral (Hart and Moore, 1994).
  - Use: Industry-level values are used in the Rajan-Zingales-based literature to proxy for the importance of the credit channel (Braun and Larrain, 2005; Aghion, Hemous, and Kharroubi, 2014).
  - Expected empirical implication:
    - Firms with plenty of tangible (collateralizable) assets find it easier to obtain external funding after a monetary tightening since secured credit tends to be more stable over the cycle (Azariadis, Kaas, and Wen, 2016; Benmelech, Kumar, and Rajan, 2020).
    - Bernanke, Gertler, and Gilchrist (1999: 1374-5) showed firms without access to secured credit show excessive sensitivity to monetary policy shocks.
    - Thus, industries high on asset tangibility can be expected to suffer less following a contraction, implying that the interaction term between asset tangibility and ∆MPR should be positive if the credit sheet channel is important.

- Investment intensity (INV)
  - Definition: Ratio of gross investment over value-added; computed for U.S. industries using the two-digit UNIDO data.
  - Use: Appears in Dedola and Lippi (2005) and Peersman and Smets (2005).
  - Expected empirical implications:
    - Interest rate channel: capital-intensive industries (high investment intensity) are more vulnerable to a monetary tightening via an increase in the user cost of capital.
    - Collateral channel: industries with a higher investment ratio have plenty of collateral and can cope better after a monetary tightening due to greater access to secured funding.
    - Investigating the interaction term on investment intensity helps assess whether the ability to secure external financing is of first-order relevance to monetary transmission.
  - Data note: The UNIDO INDSTAT3 dataset used in Samaniego and Sun (2015) was discontinued. Currently, only the INDSTAT2 and INDSTAT4 databases are provided by UNIDO.

- Labor intensity (LAB)
  - Definition: Ratio of total wages and salaries over total value-added in the United States, using UNIDO data.
  - Use: Tests both the credit and cost channels of monetary policy.
  - Expected empirical implications:
    - Credit channel (ability to secure financing): labor-intensive industries are more likely to suffer from monetary contractions because labor input typically cannot serve as collateral (suggesting a weaker ability to secure external financing).
    - Alternative view: labor-intensive industries are less dependent on external financing for investment (Ilyina and Samaniego, 2011), so a significant LAB would suggest that the ability to secure external financing (rather than the need to obtain it) is important.
    - Cost channel (prices): firms with higher labor intensity face more pressure on production costs following a monetary tightening (greater wage bill-induced borrowing requirements), which would lead to higher prices for their products.

- Liquidity needs (LIQ)
  - Definition / source: Taken from Raddatz (2006); measured by the ratio of inventories to sales, proxying reliance on short-term working capital to maintain inventories.
  - Role: Captures different dimensions of credit constraints relative to technological characteristics; not associated with collateral pledgeability and is likely a short-run phenomenon.
  - Expected empirical implications:
    - Helps distinguish the relative importance between the “ability to secure” and the “need to obtain” external funds.
    - When prices are the dependent variable, LIQ is informative on the relevance of the cost channel, which concerns short-term borrowing requirements.

- Capital depreciation (DEP)
  - Definition: Computed using industry-specific rates of depreciation from the BEA’s capital flow tables; based on resale value of capital goods and reflects physical and economic depreciation.
  - Role: Similar to asset tangibility in testing the credit channel because less durable capital stocks are not readily collateralizable (suggesting a weaker ability to secure external financing).

- Durability (DUR)
  - Definition / construction: Binary dummy equal to one if the industry produces durable goods; durability defined by the economic destination of production from the national accounts statistics. 12 out of 22 industries fall into this category.
  - Use: Follows Dedola and Lippi (2005) and Peersman and Smets (2005).
  - Expected empirical implication:
    - Conventional interest channel predicts a stronger effect of monetary policy on industries producing more durable goods, as such purchases are often financed by credit and thus more sensitive to interest rates.

*wpiea2022017-print-pdf*

### 8.    Export  intensity  (EXP). Data  on  this  characteristic  are  taken  from Giovanni  and  Levchenko

### 8. Export intensity (EXP)

### Definition and data
- Export intensity (EXP) is taken from Giovanni and Levchenko (2009), who calculated industry-level averages of the ratio of industry-level exports to value-added.
- Purpose: used to test the exchange rate channel of monetary policy — industries more reliant on exports may suffer more when domestic monetary tightening leads to an appreciation.
- Limitation: this characteristic only tests the export-related part of the exchange rate channel; the dataset spans only industry value-added growth and cannot analyze import effects (e.g., imports of final consumer goods).

### Measurement and sample details
- Industry-level growth outcome source: United Nations Industrial Development Organization (UNIDO) database, covering 153 different countries.
- Baseline industry growth measured by value-added growth for 22 manufacturing industries at the two-digit INDSTAT2 2021, ISIC Revision 3.
- Data processing steps:
  - Use data reported in current local currencies, deflated using Consumer Price Indices from the Global Inflation Database.
  - Require at least ten years of consecutive data for each industry.
  - Top and bottom one percent of the growth variables are winsorized.
- Price-deflator construction and alternatives:
  - Industry price index created by dividing value-added by the production index (as in Samaniego and Sun (2015)); the sample size for this deflator is smaller by 30 percent due to narrower production index coverage.
  - KLEMS (“EU KLEMS” and “World KLEMS”) used in separate exercises; KLEMS has price deflators taken directly from National Accounts and covers services, but has coarser manufacturing disaggregation (12 sectors vs. UNIDO’s 22/23).
  - Original INDSTAT2 includes 23 manufacturing industries; the “manufacture of recycling” industry is excluded due to insufficient observations.
- Cross-country sample for analyses using proxies for monetary policy shocks: 105 countries (33 advanced economies and 72 emerging market and developing economies).
- Rajan-Zingales convention: the U.S. is not included in regressions to alleviate reverse causality.

### Theoretical placement and expected sign
- Table mapping (excerpted):
  - EXP corresponds to the Exchange rate channel.
  - Expected sign on the interaction term (β) for EXP: −

### Correlations and normalization
- Table 3 correlation entry for EXP with other industry characteristics (exact values preserved):
  - with EFD: 0.338
  - with TAN: -0.369
  - with INV: -0.250
  - with LAB: 0.278
  - with LIQ: 0.239
  - with DEP: 0.232
  - with DUR: 0.396
- Observed multicollinearity: some industry characteristics are highly correlated (e.g., INV and TAN with correlation 0.813) which prevents simultaneous inclusion of multiple measures in the same regression.
- To ease comparison across channels, each measure X is normalized to have a zero mean and unit standard deviation over all industries.

### Empirical findings regarding EXP and the exchange rate channel
- Baseline results (from Table 4 summary):
  - Among eight industry characteristics (EFD, TAN, INV, LAB, LIQ, DEP, DUR, EXP), five (TAN, INV, LAB, DEP, and DUR) are statistically significant; EFD, LIQ, and EXP are insignificant in the baseline OLS results.
  - Interpretation: the insignificant coefficient on the interaction term with EXP suggests that industries more reliant on exports do not contract more following a monetary tightening.
  - Robustness: estimates for EXP remain insignificant when focusing solely on countries with floating currencies (classification from Ilzetzki, Reinhart, and Rogoff, 2019).
  - Interpretation in pricing frameworks: this is consistent with “dominant currency pricing” models, which predict that the exchange rate channel (if important) mostly runs through imports rather than exports (Gopinath et al., 2020).
- Instrumental variable (2SLS) approach:
  - Treating the constructed monetary policy shock as a noisy proxy and using a composite instrument interacted with industry characteristics yields second-stage estimates similar to OLS.
  - Change: the export channel (EXP) becomes statistically significant, but only at the 10% level.
- Robustness checks that continue to find EXP generally insignificant:
  - Use of inflation-targeting subsample (adoption dating from Ha, Kose, and Ohnsorge, 2019) confirms baseline findings.
  - Placebo/lead tests (regressing current industry growth on future ∆MP) show essentially none of the interaction variables using the lead of monetary policy shocks are statistically significant, supporting the parallel trends assumption.
  - Using price deflators from KLEMS to test the cost channel still does not provide evidence supporting the cost channel; EXP remains insignificant in cost-channel exercises.

### Broader interpretation and relation to channels
- Main empirical conclusion about channels:
  - Strong empirical support for the credit channel and the conventional interest rate channel (DUR and INV in part).
  - No robust empirical support for the exchange rate channel (EXP) in baseline analyses focused on industry value-added growth.
- Explanation for null finding on EXP:
  - Dataset limitation: industry-level value-added growth does not capture the import-side mechanism through which exchange rate changes might affect industries (dominant currency pricing literature suggests imports rather than export volumes/prices drive the exchange rate channel).
  - IV evidence: only weak evidence (10% significance) under IV suggests limited or context-dependent role for EXP in the studied regressions.

### Related empirical design and sensitivity notes (relevant to EXP interpretation)
- State-dependence and financial conditions:
  - The paper estimates a smooth transition model with θ = 1.5 to allow for state-dependent effects; credit-channel interaction terms are larger during bad times, while proxies for other channels (including EXP) do not exhibit business-cycle dependency.
  - Using a quantity-based credit gap (private bank credit to GDP cyclical deviation via Hamilton filter) to proxy financial conditions, interaction terms capturing the credit channel are stronger during credit contractions; EXP does not show amplification in these exercises.
- Financial development and crises:
  - Adding interactions with financial development (private bank credit to GDP) and crisis dummies shows the credit channel tends to be stronger in less-developed financial markets and during crises; financial development does not materially affect strength of the exchange rate channel as captured by EXP.

*Italic: Source — wpiea2022017-print-pdf, Section 8. Export intensity (EXP). Data on this characteristic are taken from Giovanni and Levchenko (2009).*

### appendix shows that monetary policy shocks bring about distinct effects from crises.

### Appendix: monetary policy shocks bring about distinct effects from crises

### Main conclusions on transmission channels
- The paper presents new evidence on the empirical relevance of various transmission channels of monetary policy using a panel dataset of monetary policy shocks covering over 170 countries.
- The approach combines non-traditional shock identification with a set of fixed effects to mitigate endogeneity, focusing on differential industry-level impacts rather than the overall stance of monetary policy.
- Across sensitivity tests:
  - The credit channel of monetary policy is the most robust transmission channel (including stronger effects during bad times).
  - The interest rate channel is the next most important, with industries producing durable goods more heavily affected.
  - No evidence was found supporting the cost channel of monetary policy.
  - No evidence was found for an exchange rate channel running through exports; exchange rate effects on exports are minor, consistent with "dominant currency pricing" (Gopinath et al., 2020), with any exchange rate channel mostly working through imports.

### Credit channel: amplification, collateral, and unsecured lending
- The credit channel is amplified during:
  - economic downturns,
  - periods of tightened credit conditions,
  - in countries with less developed financial markets,
  - in countries experiencing crises.
- Distinction identified:
  - The "ability to secure external financing" (i.e., access to collateral) is more relevant for monetary transmission than the "need for external financing".
  - Firms with high external financing needs do not suffer more following monetary contractions conditional on having access to collateral.
- Implications for theory and modeling:
  - Highlights importance of cyclical fluctuations in unsecured debt and a "flight to quality" from unsecured into secured lending during slowdowns.
  - Challenges models where secured debt alone drives/amplifies the business cycle (e.g., Kiyotaki and Moore (1997)) and calls for models featuring both secured and unsecured lending.

### Key empirical estimates (selected table highlights)
- Table 4 (Baseline: dependent variable = real value-added growth)
  - 푀푀ℎ푎푎푎푎푎푎 coefficient: -1.846*** (0.082) across columns (I)–(VIII).
  - Interaction 퐼퐼푇푇퐼퐼×푀푀푀푀푀푀: 0.110** (0.045) in reported column.
  - Interaction 퐿퐿푇푇퐿퐿×푀푀푀푀푀푀: -0.157** (0.062).
  - Interaction 퐸퐸퐸퐸푀푀×푀푀푀푀푀푀: -0.114** (0.056).
  - Interaction 퐸퐸푈푈푅푅×푀푀푀푀푀푀: -0.082** (0.034).
  - R-squared: 0.305.
  - Observations: 44,191.
- Table 5 (Baseline: dependent variable = price index growth)
  - 퐿퐿.퐸퐸푎푎퐷퐷 coefficient: -0.158*** (0.013) across columns (I)–(VIII).
  - Interaction 퐼퐼푇푇퐼퐼×푀푀푀푀푀푀: 0.071 (0.045).
  - Interaction 퐿퐿푇푇퐿퐿×푀푀푀푀푀푀: -0.117* (0.061).
  - Interaction 퐸퐸퐸퐸푀푀×푀푀푀푀푀푀: -0.059* (0.031).
  - R-squared: 0.346–0.347.
  - Observations: 30,582.
- Table 6 (Robustness: firm size as measure of financial constraints)
  - 푀푀ℎ푎푎푎푎푎푎: -1.844*** (0.082).
  - 푀푀푆푆푧푧푎푎
    퐺퐺퐺퐺퐺퐺퐺퐺 × 푀푀푀푀푀푀: 0.159** (0.067).
  - R-squared (top panel): 0.305.
  - Observations (top panel): 44,040.
  - For price growth: 퐿퐿.퐸퐸푎푎퐷퐷 = -0.158*** (0.013); R-squared = 0.347; Observations = 30,578.
- Table 7 (Role of the state of the business cycle; dependent = real value-added growth)
  - 푀푀ℎ푎푎푎푎푎푎: -1.962*** (0.087) across columns (I)–(VIII).
  - Selected triple interactions:
    - 퐼퐼푇푇푇푇×푀푀푀푀푀푀×퐶퐶퐸퐸퐸퐸퐶퐶... : 0.121*** (0.042).
    - 퐿퐿푇푇퐿퐿×푀푀푀푀푀푀×퐶... : -0.110** (0.053).
    - 퐸퐸퐸퐸푀푀×푀푀푀푀푀푀×퐶... : -0.146** (0.069).
  - R-squared: 0.307.
  - Observations: 42,   031 (formatted in source).
- Table 8 (Role of financial development; dependent = real value-added growth)
  - 푀푀ℎ푎푎푎푎푎푎: -1.920*** to -1.918*** (0.083).
  - 퐼퐼푇푇푇푇×푀푀푀푀푀푀: 0.170*** (0.057).
  - 퐼퐼푇푇푇푇×푀푀푀푀푀푀×퐹퐹퐸퐸: -0.003** (0.001).
  - 퐼퐼푇푇 퐼퐼×푀푀푀푀푀푀: 0.224*** (0.070).
  - 퐸퐸퐸퐸푀푀×푀푀푀푀푀푀: -0.252*** (0.053).
  - 퐸퐸퐸퐸푀푀×푀푀푀푀푀푀×퐹퐹퐸퐸: 0.004*** (0.001).
  - R-squared: 0.306.
  - Observations: 42,737.
  - Note: 퐹퐹퐸퐸 denotes financial development measured by the private bank credit to GDP ratio.
- Table 9 (Role of financial crises; dependent = real value-added growth)
  - 푀푀ℎ푎푎푎푎푎푎: -1.847*** to -1.846*** (0.082).
  - 퐸퐸퐹퐹퐸퐸×푀푀푀푀푀푀: 0.047 (0.048).
  - 퐸퐸퐹퐹퐸퐸×푀푀푀푀푀푀×퐶퐶푎푎푆푆퐸퐸푆푆퐸퐸: -0.080 (0.062).
  - 퐼퐼푇푇푇푇×푀푀푀푀푀푀×퐶퐶푎푎푆푆퐸퐸푆푆퐸퐸: 0.129* (0.069).
  - 퐼퐼푇푇 퐼퐼×푀푀푀푀푀푀×퐶... : 0.161* (0.096).
  - 퐸퐸퐸퐸푀푀×푀푀푀푀푀푀×퐶... : -0.205*** (0.066).
  - R-squared: 0.305.
  - Observations: 44,191.

### Empirical notes and significance
- Standard errors are clustered at the country-time level throughout.
- *, **, *** denote significance at 10, 5, and 1 percent, respectively.
- Dependent variables:
  - Real value-added growth (industry-country pair) based on equation (1) or (3) or (4) as noted.
  - Price index growth (industry-country pair) based on equation (1’).
- Business cycle expansions and contractions are identified using real GDP growth.
- Financial crisis dummy denoted as 퐶퐶푎푎푆푆퐸퐸푆푆퐸퐸 in the regressions.

*Source: wpiea2022017-print-pdf - appendix shows that monetary policy shocks bring about distinct effects from crises.*

### References

### References and Appendix (wpiea2022017-print-pdf)

### Sample coverage (Table A.1)
- Note: Only industries with more than ten years of data are included in the analysis.
- Baseline sample: country-level entries (selected examples with exact numbers preserved)
  - Albania: Number of industries 10, Period 2001-2019, Group EMDE
  - Algeria: Number of industries 8, Period 1985-2017, Group EMDE
  - Argentina not listed in excerpt; full list continues with entries such as:
  - Australia: Number of industries 22, Period 1975-2019, Group ADV
  - Belgium: Number of industries 22, Period 1985-2019, Group ADV
  - Brazil: Number of industries 22, Period 1996-2019, Group EMDE
  - Canada: Number of industries 22, Period 1975-2019, Group ADV
  - China: Number of industries 22, Period 1986-2018, Group EMDE
  - India: Number of industries 22, Period 1974-2019, Group EMDE
  - Japan: Number of industries 22, Period 1990-2018, Group ADV
  - Korea, Rep.: Number of industries 22, Period 1987-2019, Group ADV
  - United Kingdom: Number of industries 22, Period 1974-2019, Group ADV
  - United States not listed in excerpt; many other country entries with Number of industries and Periods (EMDE or ADV) are reported in Table A.1.
- Several small-sample countries shown (examples):
  - Botswana: Number of industries 4, Period 1993-2019, Group EMDE
  - Congo, Dem. Rep.: Number of industries 6, Period 2004-2009, Group EMDE
  - Niger: Number of industries 5, Period 1994-2018, Group EMDE

### Industry technological characteristics (Table A.2)
- Note: Indices EFD, TAN, LAB, DEP from Samaniego and Sun (2015); LIQ from Raddatz (2006); DUR dummy from Dedola and Lippi (2005); EXP from Giovanni and Levchenko (2009); INV computed using UNIDO for U.S. industries.
- Selected industry rows with exact index values:
  - ISIC 15 Food products and beverages: EFD 0.110, TAN 0.373, INV 0.065, LAB 0.277, LIQ 0.100, DEP 7.090, DUR 0, EXP 0.212
  - ISIC 16 Tobacco products: EFD -0.450, TAN 0.189, INV 0.032, LAB 0.117, LIQ 0.280, DEP 5.248, DUR 0, EXP 0.158
  - ISIC 17 Textiles: EFD 0.190, TAN 0.345, INV 0.072, LAB 0.458, LIQ 0.170, DEP 7.665, DUR 0, EXP 0.349
  - ISIC 20 Wood and products of wood and cork, except furniture: EFD 0.280, TAN 0.305, INV 0.081, LAB 0.467, LIQ 0.110, DEP 9.525, DUR 1, EXP 0.302
  - ISIC 24 Chemicals and chemical products: EFD 0.500, TAN 0.287, INV 0.094, LAB 0.229, LIQ 0.145, DEP 8.154, DUR 0, EXP 0.401
  - ISIC 29 Machinery and equipment n.e.c.: EFD 0.600, TAN 0.195, INV 0.059, LAB 0.433, LIQ 0.170, DEP 8.832, DUR 1, EXP 3.878
  - ISIC 30 Office, accounting and computing machinery: EFD 0.960, TAN 0.208, INV 0.055, LAB 0.407, LIQ 0.200, DEP 9.381, DUR 1, EXP 0.484
  - ISIC 33 Medical, precision and optical instruments, watches and clocks: EFD 0.960, TAN 0.181, INV 0.053, LAB 0.382, LIQ 0.210, DEP 9.210, DUR 1, EXP 1.642
  - ISIC 34 Motor vehicles, trailers and semi-trailers: EFD 0.360, TAN 0.264, INV 0.073, LAB 0.440, LIQ 0.180, DEP 10.559, DUR 1, EXP 1.499
  - ISIC 36 Furniture; manufacturing n.e.c.: EFD 0.370, TAN 0.245, INV 0.044, LAB 0.460, LIQ 0.169, DEP 8.968, DUR 1, EXP 0.623

### Robustness and econometric results (selected tables)
- Table A.3 (Alternative KLEMS data; interaction coefficients only)
  - Dependent variables: Value-added growth (I) and Price index growth (II)
  - EFD × MP: -0.179 (Value-added), 0.287 (Price index); R-squared 0.502 (I), 0.489 (II); Observations 8,155 (I), 7,880 (II)
    - Standard errors: (0.330) for Value-added, (0.290) for Price index
  - LAB × MP: -0.768** (Value-added), -0.533 (Price index)
    - Standard errors: (0.360) for Value-added, (0.397) for Price index
    - R-squared 0.502 (I), 0.489 (II); Observations 8,155 (I), 7,880 (II)
  - Note: Clustered standard errors at the country-time level in parentheses. *, **, *** denote significance at 10, 5, and 1 percent.

- Table A.4 (IV regression; dependent variable: real value-added growth; estimates based on equation (1))
  - Baseline monetary shock coefficient (푀ℎ푎푎푎푎푎푎): consistently around -2.024*** to -2.027*** across columns (I) to (VIII)
    - Exact reported values: -2.024***, -2.026***, -2.027***, -2.022***, -2.025***, -2.024***, -2.025***, -2.027***
    - Standard errors uniformly reported as (0.112) or (0.111) across columns
  - Selected interaction coefficients (with standard errors):
    - ITT × MP: 0.065** (0.028) in one specification
    - ITI × MP: 0.101*** (0.031) in one specification
    - LTL × MP: -0.126*** (0.044)
    - EEMM × MP: -0.112*** (0.028)
    - EEUR × MP: -0.085** (0.042)
    - EEXM × MP: -0.054* (0.028)
  - F-statistics reported per column: 5,938.960; 5,985.619; 6,415.074; 6,662.989; 6,214.009; 5,934.613; 5,593.524; 5,358.595
  - Observations: 31,109 (all columns)
  - Note: Clustered standard errors at the country-time level in parentheses. *, **, *** denote significance at 10, 5, and 1 percent.

- Table A.5 (Inflation-targeting countries subsample; dependent variable: real value-added growth)
  - Monetary shock coefficient (푀ℎ푎푎푎푎푎푎): -1.983*** to -1.988*** across columns
    - Exact values reported: -1.983*** (multiple columns), -1.988*** in one column; standard errors (0.149)
  - Selected interaction coefficients (with standard errors):
    - ITT × MP: 0.062** (0.031)
    - ITI × MP: 0.116*** (0.043)
    - LTL × MP: -0.162*** (0.061)
    - EEMM × MP: -0.121** (0.058)
    - EEUR × MP: -0.079*** (0.027)
  - R-squared: 0.373 (all columns)
  - Observations: 21,845 (all columns)

- Table A.6 (Placebo test; dependent variable: real value-added growth; lagged monetary policy changes replaced by forward variable)
  - Monetary shock coefficient (푀ℎ푎푎푎푎푎푎): -1.853*** across columns; standard error (0.084)
  - Selected interaction coefficients:
    - EFD × MP: 0.058** (0.026)
    - ITT × MP: -0.040 (0.049)
    - LTL × MP: -0.058 (0.070)
    - EEMM × MP: 0.002 (0.042)
  - R-squared: 0.324 (all columns)
  - Observations: 43,986 (all columns)

- Table A.7 (Role of the state of financial conditions; dependent variable: real value-added growth; equation (4))
  - Monetary shock coefficient (푀ℎ푎푎푎푎푎푎): -1.961*** to -1.963*** across columns; standard errors (0.088)
  - Selected triple-interaction coefficients and standard errors (MP × industry characteristic × financial state):
    - ITT × MP × CCECaaacCCS EEEE: 0.103*** (0.030)
    - ITI × MP × CCEC...: 0.174*** (0.039)
    - LTL × MP × CCEC...: -0.162*** (0.044)
    - EEMM × MP × CCEC...: -0.217*** (0.032)
    - EEUR × MP × CCEC...: -0.104*** (0.033)
  - R-squared: 0.310 (all columns)
  - Observations: 42,042 (all columns)
  - Note: Credit expansions and contractions identified using cyclical component of private bank credit-to-GDP ratio.

- Table A.8 (Same role using a one-sided HP filter)
  - Monetary shock coefficient (푀ℎ푎푎푎푎푎푎): -1.956*** to -1.958*** across columns; standard errors (0.087)
  - Selected significant triple-interaction coefficients:
    - ITT × MP × CCEC...: 0.095* (0.053)
    - ITI × MP × CCEC...: 0.174*** (0.065)
    - LTL × MP × CCEC...: -0.223** (0.095)
    - EEMM × MP × CCEC...: -0.194** (0.076)
  - Observations: 42,254 (all columns)
  - R-squared: 0.309 (all columns)

- Table A.9 (Controlling for financial development; dependent variable: real value-added growth)
  - Monetary shock coefficient (푀ℎ푎푎푎푎푎푎): -1.920*** to -1.918*** across columns; standard errors (0.083)
  - Selected interaction coefficients (with 퐹퐹퐸퐸 = private bank credit to GDP ratio):
    - EFD × ∆MP: 0.013 (0.028)
    - EFD × FFE: 0.019*** (0.006)
    - ITT × ∆MP: 0.062* (0.034)
    - ITI × ∆MP: 0.107** (0.044)
    - LTL × ∆MP: -0.157** (0.062)
    - EEMM × ∆MP: -0.113** (0.056)
    - EEUR × ∆MP: -0.080** (0.034)
  - R-squared: 0.306
  - Observations: 42,737

- Table A.10 (Controlling for crisis episodes; dependent variable: real value-added growth)
  - Monetary shock coefficient (푀ℎ푎푎푎푎푎푎): -1.846*** to -1.848*** across columns; standard errors (0.082)
  - Selected interaction coefficients with crisis dummy (퐶퐶푎푎푆푆퐸퐸푆푆퐸퐸):
    - EFD × Crisis: 1.299** (0.651)
    - ITT × Crisis: -1.019 (0.745)
    - ITI × Crisis: -1.040 (0.810)
    - EEUR × Crisis: -1.720** (0.713)
  - R-squared: 0.305
  - Observations: 44,191

### Figure A.1: Source of monetary policy shocks, 1973-2020 (labeling)
- Label mapping provided:
  - 1 (Red) = High-frequency shocks
  - 2 (Green) = Shocks implied by changes in swap yields
  - 3 (Blue) = Shocks implied by changes in bond yields
  - 4 (Grey) = Survey-based shock estimates
  - 5 (Yellow) = Taylor residuals
  - 61 (Red) = High-frequency shocks coming from anchoring currencies
  - 65 (Yellow) = Taylor residuals coming from anchoring currencies

*Italic: Source document — wpiea2022017-print-pdf (References and Appendix tables).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022017-print-pdf.pdf_