## Carbon Risk in Loan Pricing: Commitment Channels and Real Effects — Working Paper No. WP/2025/250

## Source details

**Canonical URL:** [Carbon Risk in Loan Pricing: Commitment Channels and Real Effects — Working Paper No. WP/2025/250](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025250-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025250-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025250-source-pdf.pdf.json)

---

### Data sources and primary variables
- Syndicated loan data: Refinitiv LoanConnector DealScan (formerly LPC DealScan). Key loan-level variable: all-in-spread-drawn (AISD) measured in basis points (bps), defined as the amount the borrower pays over LIBOR for each dollar drawn, incorporating both the loan spread and any annual or facility fees charged by the lending syndicate.
- DealScan observation unit: unique lender–borrower–tranche combination; for most analyses data are aggregated at the lead arranger–loan tranche level.
- Carbon emissions data: Urgentem. Reports absolute emissions (in tCO2e) and emission intensity (in tCO2e per $1 million of revenue) for scope 1, scope 2, and scope 3. Analysis focuses on scope 1 and scope 2 emissions and on carbon emission intensity.
- Firm balance sheet and controls: Compustat (leverage, firm size, profitability, capital expenditure).
- Probability of default (PD): National University of Singapore Credit Research Initiative (NUS-CRI).
- Environmental expenditure (annual): Refinitiv (capital spending related to environmental protection and reduction of adverse environmental impacts).
- Lender “green” classification: Science Based Targets initiative (SBTi) membership (merged via fuzzy name matching).
- Monetary policy shocks: Jarociński and Karadi (2020), aggregated monthly shocks to quarterly frequency by summing within each quarter and normalized by subtracting the mean and dividing by the standard deviation.

### Data cleaning, matching, sample construction, and variable definitions
- DealScan cleaning: remove incomplete or duplicate entries, exclude loans with inconsistencies in borrower identity or loan terms, and filter out non-lender roles within syndicates.
- Lead arranger identification: follow Bharath et al. (2011) criteria; loan tranches are arranged by an average of 3.7 lead arrangers in the sample.
- Urgentem data frequency alignment: annual emission values repeated across the four quarters for quarterly analysis.
- Matching procedure (three steps):
  1. Direct matching using Legal Entity Identifiers (LEIs).
  2. Indirect matching using ISINs with linking tables (GLEIF and ANNA; Dealscan–Worldscope linking table from Beyhaghi et al. (2021); Dealscan–Compustat linking table from Chava and Roberts (2008)).
  3. Fuzzy name matching followed by manual verification.
- Resulting matched sample: 2,746 unique firms.
- Carbon intensity (Carbonbt(y)): scope 1 and scope 2 carbon emissions per $1 million of revenue (expressed in thousand tCO2e/$1 million of revenue in tables).
- AISD dependent variable: all-in-spread-drawn for loan tranche l arranged by lender i for borrower b in quarter t.
- Borrower-level controls (Borrowerbt): PD in quarter t or balance-sheet variables (leverage, size, profitability) lagged by one quarter.
- Loan-level controls (Loanl): loan maturity, loan tranche amount, covenant, secured (collateral), leveraged, term loan, and loan purpose (six categories).
- Fixed effects: lender-time fixed effects (αit), industry fixed effects (φIndustry, Major Industry Group classification in DealScan, 30 categories), location fixed effects (λLocation, nine U.S. census divisions).
- Standard error clustering: clustered at the lender level.

### Key dataset scope and summary statistics (selected from Table 1, Panel A–C and Table 2)
- Sample period for loan pricing regressions: 2010Q1 to 2023Q4.
- Observations in regressions (Table 3): 33,689; 33,378; 13,626; 29,130 (columns 1–4 respectively).
- Tranche characteristics (Panel A, Table 1):
  - All-in-spread-drawn (bps): Mean 223.26 Std.Dev. 137.17 p25 125.00 Median 175.00 p75 275.00
  - Maturity (years): Mean 4.50 Std.Dev. 1.77 p25 3.96 Median 5.00 p75 5.00
  - Amount ($ million): Mean 1047.67 Std.Dev. 1786.48 p25 250.00 Median 500.00 p75 1200.00
  - Log(amount): Mean 6.23 Std.Dev. 1.27 p25 5.52 Median 6.21 p75 7.09
  - Covenant (0/1): Mean 0.42 Std.Dev. 0.49 p25 0.00 Median 0.00 p75 1.00
  - Collateral (0/1): Mean 0.40 Std.Dev. 0.49 p25 0.00 Median 0.00 p75 1.00
  - Leveraged (0/1): Mean 0.50 Std.Dev. 0.50 p25 0.00 Median 1.00 p75 1.00
  - Term loan (0/1): Mean 0.42 Std.Dev. 0.49 p25 0.00 Median 0.00 p75 1.00
  - Green/sustainability-linked loan (0/1): Mean 0.01 Std.Dev. 0.12 p25 0.00 Median 0.00 p75 0.00
- Borrower characteristics (Panel B, Table 1):
  - Carbon intensity (thousand tCO2e/$1 million of revenue): Mean 0.25 Std.Dev. 0.80 p25 0.02 Median 0.04 p75 0.10
  - Emission reduction target (0/1): Mean 0.18 Std.Dev. 0.39 p25 0.00 Median 0.00 p75 0.00
  - PD (probability): Mean 0.37 Std.Dev. 2.78 p25 0.00 Median 0.02 p75 0.13
  - Leverage (total debt/total assets): Mean 0.31 Std.Dev. 0.22 p25 0.13 Median 0.29 p75 0.45
  - Total assets ($ million): Mean 10515.04 Std.Dev. 41650.10 p25 744.77 Median 2128.00 p75 6958.10
  - Log(total assets): Mean 7.80 Std.Dev. 1.59 p25 6.61 Median 7.66 p75 8.85
  - Return on assets: Mean 0.01 Std.Dev. 0.07 p25 0.00 Median 0.01 p75 0.02
- Lender characteristics (Panel C, Table 1):
  - SBTi member (0/1): Mean 0.02 Std.Dev. 0.14 p25 0.00 Median 0.00 p75 0.00
- Industry-level carbon intensity highlights (Table 2, scope 1 and 2 carbon intensity in thousand tCO2e per $1 million of revenue):
  - Utilities: Mean 1.96 Std.Dev. 2.21 p25 0.29 Median 1.05 p75 3.29
  - Mining: Mean 0.76 Std.Dev. 1.15 p25 0.30 Median 0.46 p75 0.62
  - Oil and Gas: Mean 0.49 Std.Dev. 0.37 p25 0.25 Median 0.46 p75 0.70
  - Financial Services (example low-carbon industry): Mean 0.01 Std.Dev. 0.03 p25 0.00 Median 0.00 p75 0.01
- Industry grouping: 30 industries classified into high-, medium-, and low-carbon groups; Figure 1 shows average carbon intensity from 2010 to 2023 with little fluctuation and limited changes in cross-industry group differences.

### Baseline and extended empirical specifications
- Baseline loan pricing model (Equation (1)):
  - Ylibt = β0 + β1 Carbonbt(y) + β2 Borrowerbt + β3 Loanl + αit + φIndustry + λLocation + εlibt
  - Key hypothesis: β1 > 0 indicates lenders charge a carbon premium to carbon-intensive borrowers.
- Extended regression with commitments:
  - Y_libt = β_0 + β_1 Carbon_bt(y) + β_2 Z_bt,it,l,t + β_3 Carbon_bt(y) × Z_bt,it,l + β_4 PD_bt + β_5 Loan_l + α_it + φ_Industry + λ_Location + ε_libt
  - Z_bt,it,l,t denotes borrower- or lender-commitment variables (Target, Disclosed, Green loan, lead arranger SBTi membership).
- Baseline specification uses PD to capture time-varying borrower credit risk and includes lender-time fixed effects to compare spreads charged by the same lender in the same quarter across borrowers.
- Standard errors clustered at the lender level unless otherwise stated.

### Key empirical findings — carbon premium and commitments (section 3.2, Table 4 selected)
- Carbon coefficients (Table 4, selected columns):
  - Column (1): 2.39** [1.049]
  - Column (2): 2.52** [0.993]
  - Column (3): 3.13*** [0.950]
  - Column (4): 3.24*** [0.942]
- Borrower commitment effects:
  - Target: Target = -22.19*** [2.443]; Target × Carbon = 5.40*** [1.703]
    - For high-emitting borrowers (90th percentile), estimated discount associated with a target ≈ 19 bps.
  - Disclosed: Disclosed = -15.70*** [2.471]; Disclosed × Carbon = 3.82** [1.841]
    - For high-emitting borrowers (90th percentile), estimated discount ≈ 14 bps.
  - Green loan: Green loan = -11.13** [4.418]; Green loan × Carbon = 7.94** [3.379]
    - For high-emitting borrowers (90th percentile), estimated discount ≈ 7 bps.
- Lender commitment effects:
  - SBTi member × Carbon: 7.94*** [2.129]
    - Carbon-intensive borrowers face higher loan spreads when borrowing from SBTi signatories: carbon premium increases by 4 bps for high-emitting borrowers (90th percentile) borrowing from SBTi signatories.
  - UNEP FI membership: no statistically significant impact on spreads found.
- Sample and model statistics (Table 4):
  - Observations: 29,130 (columns 1–4)
  - R-squared: 0.6130 (col 1); 0.6110 (col 2); 0.6100 (col 3); 0.610 (col 4)
  - Controls: PD included; loan controls included; lender-time FE, Industry FE, Location FE included.
  - Standard errors clustered at the lender level; significance: ***, **, * at 1%, 5%, 10%.

### Real effects: investment, innovation, and liquidity (section 3.3, Tables 5–7, A.5)
- Empirical strategy:
  - Firm-quarter regressions of log(CAPEX), log(R&D expense), log(Cash) on Carbon, commitments, Carbon×commitment, borrower controls, borrower FE, time FE; standard errors clustered at borrower level.
- Capital expenditure (Table 5):
  - Target coefficient: 0.08* (column 1); Target×Carbon = -0.07***.
  - Quantitative example: 90th percentile of carbon intensity in the firm-level dataset = 0.28; for a high-emitting borrower at that percentile, committing to an emission reduction target is associated with a 6% increase in capital expenditure.
  - Disclosed: Disclosed = 0.07*; Disclosed×Carbon = -0.10***.
  - Green loan: Green loan = 0.20 (not statistically significant); Green loan×Carbon = -2.55***.
  - Post SBTi lead: level not significant; Post SBTi lead×Carbon = -0.09* (carbon-intensive borrowers reduce investment if a prior lead arranger becomes green).
  - Observations = 19,574; R-squared = 0.907–0.908.
- R&D expense (Table 6):
  - Carbon coefficients: 0.03*** (col 1); 0.18** (col 2); 0.02 (col 3); 0.02*** (col 4).
  - Target×Carbon = -0.26***; Disclosed×Carbon = -0.17*; Post SBTi lead×Carbon = -0.26***.
  - Observations = 22,709; R-squared = 0.980.
- Cash and short-term investments (liquidity hoarding, Table 7):
  - Target = -0.10** (column 1); Target×Carbon not significant (0.06).
  - Disclosed = -0.05 (not significant); Disclosed×Carbon = 0.01.
  - Green loan = -0.28** (column 3).
  - Post SBTi lead = -0.21*** (column 4).
  - Observations = 45,483; R-squared = 0.880.
- Environmental expenditure (Table A.5):
  - Column (1): Carbon 0.28* [0.170]; Observations 448; R-squared 0.9000.
  - Column (2): Target 0.32* [0.186]; Target×Carbon -0.24** [0.105]; Observations 448; R-squared 0.8970.
  - Column (4): Green loan 0.94** [0.387]; Green loan×Carbon -1.03*** [0.126]; Post SBTi lead×Carbon -0.39*** [0.127]; Observations 448; R-squared 0.905.
  - Note: limited sample for environmental expenditure regressions, but broadly consistent with CAPEX and R&D results.

### Monetary policy interactions and time variation (section 3.3, Table 8 and Table 9)
- Monetary policy shocks (MPS) interacted with Carbon and commitments:
  - Carbon coefficient (Table 8, col 1) = 3.64***.
  - MPS×Carbon (β4) = 3.16*** (col 1): carbon premium increases when monetary policy tightens.
    - Example magnitude: a one standard deviation monetary policy shock is associated with a 1.6 bps additional increase in loan spreads for borrowers at the 90th percentile of carbon intensity.
  - MPS×Target = -2.69* (mitigating effect); MPS×Target×Carbon = 3.61** (mitigating effect diminishes with carbon intensity).
  - MPS×Disclosed = -4.10***; MPS×Disclosed×Carbon = 0.62 (not significant).
  - MPS×Green Loan = -12.42***; MPS×Green Loan×Carbon = 24.63*.
  - SBTi member×Carbon = 8.55***; MPS×SBTi member×Carbon = -12.23***:
    - Following monetary tightening, the increase in loan spreads for carbon-intensive firms is smaller when borrowing from SBTi members; however, SBTi members demand significantly higher loan spreads when monetary policy tightens, raising borrowing costs for carbon-intensive firms from SBTi members.
  - Observations = 29,130; R-squared ranges 0.610–0.613.
- Time variation in the carbon premium (rolling regressions and event interactions):
  - Rolling six-year windows (2010–2015 to 2018–2023): carbon premium positive and increasing early, then declines; final window 2018–2023 coefficient turns negative but insignificant.
  - Event-interaction results (Table 9):
    - Paris withdrawal×Carbon: coefficients not significant (1.47; 1.64).
    - Post COVID (loans originated March 2020 to March 2022): Post COVID×Carbon = -4.76*** (col 3) and -5.08*** (col 4) — post–COVID-19 decline in carbon premium.
    - Post IRA (after Inflation Reduction Act in July 2022): Post IRA×Carbon = -8.69* and -9.02* — IRA associated with weakening of the carbon premium.
  - Interpretation: decline in carbon premium likely reflects pandemic-related expansionary policies and COVID-19 support measures improving financing conditions for carbon-intensive industries, and IRA provisions (tax credits, grants, loan guarantees) encouraging low-carbon investment.

### Overall conclusions and substantive findings (section 6)
- Robust evidence that lenders charge higher loan spreads to borrowers with higher carbon intensity (positive and significant Carbon coefficients across specifications and robustness checks).
- Borrower commitments matter:
  - Disclosure, emission reduction targets, and green/sustainability-linked tranches are associated with lower loan spreads.
  - However, the mitigating discounts from commitments decline with higher borrower carbon intensity (significant positive interaction terms Target×Carbon, Disclosed×Carbon, Green loan×Carbon).
- Lender commitments matter:
  - SBTi membership by lead arrangers is associated with higher carbon premia for carbon-intensive borrowers.
- Real effects:
  - Borrower commitments associated with higher capital expenditure and R&D spending, though these positive effects weaken or reverse as borrower carbon intensity increases.
  - Committed borrowers hold lower liquidity (lower cash and short-term investments); relationships with lenders that commit to environmental objectives can reduce investment for carbon-intensive firms.
- Monetary policy:
  - Monetary tightening amplifies the carbon premium; some mitigating effects of borrower commitments under tightening, but complex heterogeneity by lender commitments (SBTi interactions).
- Time trends:
  - Carbon premium has diminished in recent years; evidence suggests COVID-19 policy responses and the Inflation Reduction Act contributed to this decline.

*Source: 2.1 Data Sources and Variables; 3.2 Impact of Environmental Commitments on the Carbon Premium; 3.3 Impact of Environmental Commitments on Corporate Behavior — wpiea2025250-source-pdf*

### 2.1  Data Sources and Variables

### 2.1  Data Sources and Variables

### Data sources and primary variables
- Syndicated loan data: Refinitiv LoanConnector DealScan (formerly LPC DealScan). Key loan-level variable: all-in-spread-drawn (AISD) measured in basis points (bps), defined as the amount the borrower pays over LIBOR for each dollar drawn, incorporating both the loan spread and any annual or facility fees charged by the lending syndicate.
- DealScan observation unit: unique lender–borrower–tranche combination. For most analyses data are aggregated at the lead arranger–loan tranche level.
- Carbon emissions data: Urgentem. Reports absolute emissions (in tCO2e) and emission intensity (in tCO2e per $1 million of revenue) for scope 1, scope 2, and scope 3. Analysis focuses on scope 1 and scope 2 emissions and on carbon emission intensity.
- Firm balance sheet and controls: Compustat (leverage, firm size, profitability, capital expenditure).
- Probability of default (PD): National University of Singapore Credit Research Initiative (NUS-CRI).
- Environmental expenditure (annual): Refinitiv (capital spending related to environmental protection and reduction of adverse environmental impacts).
- Lender “green” classification: Science Based Targets initiative (SBTi) membership (merged via fuzzy name matching).
- Monetary policy shocks: Jarociński and Karadi (2020), aggregated monthly shocks to quarterly frequency by summing within each quarter and normalized by subtracting the mean and dividing by the standard deviation.

### Data cleaning, matching, and sample construction
- DealScan cleaning: remove incomplete or duplicate entries, exclude loans with inconsistencies in borrower identity or loan terms, and filter out non-lender roles within syndicates.
- Lead arranger identification: following Bharath et al. (2011), classify a lender as a lead arranger if designated “Lead Arranger” in DealScan or assigned roles including administrative agent, agent, bookrunner, (mandated) arranger, (mandated) lead arranger, or syndication agent. In the sample, loan tranches are arranged by an average of 3.7 lead arrangers.
- Urgentem data frequency alignment: annual emission values repeated across the four quarters for quarterly analysis.
- Matching procedure (three steps):
  1. Direct matching using Legal Entity Identifiers (LEIs).
  2. Indirect matching using ISINs with linking tables (ISIN-to-LEI relationship file from GLEIF and ANNA; Dealscan–Worldscope linking table from Beyhaghi et al. (2021); Dealscan–Compustat linking table from Chava and Roberts (2008)).
  3. Fuzzy name matching followed by manual verification for remaining unmatched firms.
- Resulting matched sample: 2,746 unique firms.

### Variable constructions and fixed effects
- Carbon intensity (Carbonbt(y)): scope 1 and scope 2 carbon emissions per $1 million of revenue (expressed in thousand tCO2e/$1 million of revenue in tables).
- AISD dependent variable: all-in-spread-drawn for loan tranche l arranged by lender i for borrower b in quarter t.
- Borrower-level controls (Borrowerbt): either PD in quarter t or balance-sheet variables (leverage, size, profitability) lagged by one quarter.
- Loan-level controls (Loanl): loan maturity, loan tranche amount, covenant, secured (collateral), leveraged, term loan, and loan purpose (six categories: corporate operating purposes; debt repayment; mergers and acquisitions; recapitalization; working capital financing; residual purposes).
- Fixed effects: lender-time fixed effects (αit), industry fixed effects (φIndustry, Major Industry Group classification in DealScan, 30 categories), location fixed effects (λLocation, nine U.S. census divisions).
- Standard error clustering: clustered at the lender level.

### Lender and borrower commitment indicators
- Emission reduction targets: Urgentem reports firms’ emission reduction targets including target start year (used to capture borrower climate commitments).
- Public disclosure indicator: Urgentem variable indicating whether emissions data are disclosed by the firm and publicly available.
- Green lenders: indicator equals 1 if lender has committed to or set a target for emissions reduction through SBTi.

### Frequency alignment and aggregation
- Emission and environmental expenditure data are annual; emissions values are repeated for quarters, environmental expenditure merged at annual frequency with Compustat company balance sheet information using ISINs.
- Monetary policy shocks: monthly shocks summed to quarterly frequency and normalized for interpretation.

### Key statistics and dataset scope (from Table 1 and related text)
- Sample period for loan pricing regressions: 2010Q1 to 2023Q4.
- Observations and sample sizes reported in regressions: Observations in Table 3: 33,689; 33,378; 13,626; 29,130 (columns 1–4 respectively).
- Tranche characteristics (Panel A, Table 1):
  - All-in-spread-drawn (bps): Mean 223.26 Std.Dev. 137.17 p25 125.00 Median 175.00 p75 275.00
  - Maturity (years): Mean 4.50 Std.Dev. 1.77 p25 3.96 Median 5.00 p75 5.00
  - Amount ($ million): Mean 1047.67 Std.Dev. 1786.48 p25 250.00 Median 500.00 p75 1200.00
  - Log(amount): Mean 6.23 Std.Dev. 1.27 p25 5.52 Median 6.21 p75 7.09
  - Covenant (0/1): Mean 0.42 Std.Dev. 0.49 p25 0.00 Median 0.00 p75 1.00
  - Collateral (0/1): Mean 0.40 Std.Dev. 0.49 p25 0.00 Median 0.00 p75 1.00
  - Leveraged (0/1): Mean 0.50 Std.Dev. 0.50 p25 0.00 Median 1.00 p75 1.00
  - Term loan (0/1): Mean 0.42 Std.Dev. 0.49 p25 0.00 Median 0.00 p75 1.00
  - Green/sustainability-linked loan (0/1): Mean 0.01 Std.Dev. 0.12 p25 0.00 Median 0.00 p75 0.00
- Borrower characteristics (Panel B, Table 1):
  - Carbon intensity (thousand tCO2e/$1 million of revenue): Mean 0.25 Std.Dev. 0.80 p25 0.02 Median 0.04 p75 0.10
  - Emission reduction target (0/1): Mean 0.18 Std.Dev. 0.39 p25 0.00 Median 0.00 p75 0.00
  - PD (probability): Mean 0.37 Std.Dev. 2.78 p25 0.00 Median 0.02 p75 0.13
  - Leverage (total debt/total assets): Mean 0.31 Std.Dev. 0.22 p25 0.13 Median 0.29 p75 0.45
  - Total assets ($ million): Mean 10515.04 Std.Dev. 41650.10 p25 744.77 Median 2128.00 p75 6958.10
  - Log(total assets): Mean 7.80 Std.Dev. 1.59 p25 6.61 Median 7.66 p75 8.85
  - Return on assets (net income/total assets): Mean 0.01 Std.Dev. 0.07 p25 0.00 Median 0.01 p75 0.02
- Lender characteristics (Panel C, Table 1):
  - SBTi member (0/1): Mean 0.02 Std.Dev. 0.14 p25 0.00 Median 0.00 p75 0.00
- Industry-level carbon intensity highlights (Table 2, scope 1 and 2 carbon intensity in thousand tCO2e per $1 million of revenue):
  - Utilities: Mean 1.96 Std.Dev. 2.21 p25 0.29 Median 1.05 p75 3.29
  - Mining: Mean 0.76 Std.Dev. 1.15 p25 0.30 Median 0.46 p75 0.62
  - Oil and Gas: Mean 0.49 Std.Dev. 0.37 p25 0.25 Median 0.46 p75 0.70
  - (Additional industries reported in Table 2 down to Financial Services: Mean 0.01 Std.Dev. 0.03 p25 0.00 Median 0.00 p75 0.01)
- Industry grouping and time series: 30 industries classified into three groups—high-, medium-, and low-carbon—based on mean carbon intensity; Figure 1 shows average carbon intensity from 2010 to 2023 with little fluctuation and limited changes in cross-industry group differences.

### Variable usage in empirical specifications
- Baseline loan pricing model (Equation (1)):
  - Ylibt = β0 + β1 Carbonbt(y) + β2 Borrowerbt + β3 Loanl + αit + φIndustry + λLocation + εlibt
  - Key hypothesis: β1 > 0 indicates lenders charge a carbon premium to carbon-intensive borrowers.
- Baseline specification uses PD to capture time-varying borrower credit risk and includes lender-time fixed effects to compare spreads charged by the same lender in the same quarter across borrowers.

*Source: 2.1  Data Sources and Variables (wpiea2025250-source-pdf)*

### 3.2  Impact of Environmental Commitments on the Carbon Pre-

### 3.2  Impact of Environmental Commitments on the Carbon Pre-

### Overview
- Environmental commitments by borrowers and lenders may influence how carbon risk is priced in loan contracts by signaling long-term risk management and alignment with regulatory or market expectations.
- The section extends the baseline model by interacting borrower carbon intensity, Carbon_bt(y), with variables capturing borrower and lender commitments to test their effect on the carbon premium.

### Empirical specification
- Extended regression:
  Y_libt = β_0 + β_1 Carbon_bt(y) + β_2 Z_bt,it,l,t + β_3 Carbon_bt(y) × Z_bt,it,l + β_4 PD_bt + β_5 Loan_l + α_it + φ_Industry + λ_Location + ε_libt
- Z_bt,it,l,t denotes borrower- or lender-commitment variables varying by borrower-time (bt), lender-time (it), or loan (l).
- Commitment variables analyzed:
  - Borrower emission reduction target (Target dummy).
  - Public disclosure of emissions (Disclosed dummy; disclosure category 1 or 2 for public, complete scope 1 and 2 emissions).
  - Green or sustainability-linked loan tranche (Green loan dummy).
  - Lead arranger being an SBTi member (green lender dummy).

### Key empirical findings — borrower commitments
- General: The carbon premium (β_1) remains positive and statistically significant when controlling for borrower or lender commitments.
- Emission reduction target:
  - Lenders charge lower spreads for borrowers with an emission reduction target (Target coefficient negative and significant).
  - Interaction Target × Carbon is negative and statistically significant, indicating the “commitment discount” for a target declines with borrower carbon intensity.
  - Quantitatively, for high-emitting borrowers (at the 90th percentile of carbon intensity), the estimated discount associated with a target is approximately 19 bps.
  - The discount diminishes as emissions increase (see Figure 2 marginal effect).
- Public disclosure of emissions:
  - Borrowers who disclose emissions face a spread discount (Disclosed coefficient negative and significant).
  - Interaction Disclosed × Carbon is positive and significant, implying the benefit of disclosure fades as carbon intensity increases.
  - For high-emitting borrowers (at the 90th percentile), the estimated discount is 14 bps.
- Green and sustainability-linked loans:
  - Firms borrowing via green or sustainability-linked loans face lower loan spreads (Green loan coefficient negative and significant).
  - Interaction Green loan × Carbon is positive and significant, indicating the discount from green borrowing declines with carbon intensity.
  - For high-emitting borrowers (at the 90th percentile), the estimated discount from borrowing through a green or sustainability-linked loan is 7 bps.

### Key empirical findings — lender commitments
- Lender commitment defined as SBTi membership (lead arranger SBTi signatory).
- The coefficient capturing differential spreads for green lenders is absorbed by lender-time fixed effects; interaction with Carbon captures whether that differential varies with carbon intensity.
- Results show carbon-intensive borrowers face higher loan spreads when borrowing from SBTi signatories.
  - The carbon premium increases by 4 bps for high-emitting borrowers (at the 90th percentile) borrowing from SBTi signatories.
- No statistically significant impact of UNEP FI membership on spreads was found.

### Numerical results from Table 4 (selected coefficients and statistics)
- Carbon:
  - Column (1): 2.39** [1.049]
  - Column (2): 2.52** [0.993]
  - Column (3): 3.13*** [0.950]
  - Column (4): 3.24*** [0.942]
- Target:
  - Target: -22.19*** [2.443]
  - Target × Carbon: 5.40*** [1.703]
- Disclosed:
  - Disclosed: -15.70*** [2.471]
  - Disclosed × Carbon: 3.82** [1.841]
- Green loan:
  - Green loan: -11.13** [4.418]
  - Green loan × Carbon: 7.94** [3.379]
- SBTi interaction:
  - SBTi member × Carbon: 7.94*** [2.129]
- Sample and model controls:
  - Observations: 29,130 (columns 1–3); 29,130 (column 4)
  - R-squared: 0.6130 (col 1); 0.6110 (col 2); 0.6100 (col 3); 0.610 (col 4)
  - PD: Yes (included)
  - Loan controls: Yes (tranche size, maturity, collateral, covenant status, leveraged status, and loan purpose)
  - Fixed effects: Lender-time FE Yes; Industry FE Yes; Location FE Yes
  - Standard errors clustered at the lender level (reported in brackets)
  - Significance notation: ***, **, * represent significance at the 1%, 5%, and 10% levels, respectively

### Interpretation and comparatives
- Borrower commitment signals (targets, disclosure, green loan tranches) are associated with lower spreads, but their mitigating effects on spreads decline with higher borrower carbon intensity.
- Lender commitments (SBTi membership) are associated with higher carbon premia for carbon-intensive borrowers, increasing the carbon premium for high-emitting borrowers by 4 bps at the 90th percentile.
- Results contrast with some prior findings in Europe (Altavilla et al., 2024), suggesting regional differences in how commitments interact with carbon intensity and perceived costs of decarbonization.

*Source: 3.2 Impact of Environmental Commitments on the Carbon Premium (wpiea2025250-source-pdf)*

### 3.3  Impact of Environmental Commitments on Corporate Behav-

### 3.3  Impact of Environmental Commitments on Corporate Behavior

### Mechanisms and hypotheses
- Environmental commitments can shape firm-level decisions by signaling credible long-term strategies, affecting:
  - real investment choices (capital expenditure),
  - innovation activity (R&D spending),
  - financial decisions (liquidity hoarding and cash holdings).
- Lower loan spreads for committed firms can:
  - reduce financing costs and support higher capital expenditures,
  - facilitate greater investment in R&D, particularly in low-carbon technologies,
  - reduce the need for precautionary liquidity hoarding, lowering cash holdings.
- Green lenders may allocate capital toward firms pursuing environmental strategies, reinforcing behavioral shifts.

### Empirical strategy
- Data sources and sample:
  - Firm-level carbon emissions from Urgentem merged with Compustat balance sheet information.
  - Sample restricted to firms borrowing in the syndicated loan market (sample consistent with Table 3, column 1).
- Outcome variables:
  - natural logarithm of capital expenditure ($ million),
  - natural logarithm of R&D expenses ($ million),
  - natural logarithm of cash and short-term investments ($ million).
- Main regression (firm-quarter level):
  - Y_bt = β0 + β1 Carbon_bt(y) + β2 Z_bt + β3 Carbon_bt(y) × Z_bt + β4 Borrower_bt + φ_b + λ_t + ε_bt
  - Borrower controls: lagged leverage, total assets, return on assets.
  - Borrower commitments (Z_bt): emission reduction target (Target), disclosed emissions (Disclosed), green/sustainability-linked tranche (Green loan).
  - Lender commitment: Post SBTi lead (dummy equals one if borrower previously engaged with a lead arranger that subsequently committed/validated SBTi target).
  - Standard errors clustered at the borrower level.
- Coefficients of interest:
  - β2: direct effect of commitment on outcomes,
  - β3: differential impact of carbon intensity on behavior for committed vs non-committed borrowers.

### Capital expenditure findings (Table 5)
- Baseline:
  - For borrowers without an emission reduction target, carbon intensity does not significantly affect investment decisions.
- Commitment effects:
  - Borrowers that committed to an emission reduction target show higher capital expenditure (Target coefficient 0.08* in column 1).
  - The positive effect of Target on capital expenditure declines with carbon intensity (Target×Carbon = -0.07***).
  - Quantitative example: for a high-emitting borrower (at the 90th percentile), committing to an emission reduction target is associated with a 6% increase in capital expenditure.
    - The 90th percentile of carbon intensity in the firm-level dataset = 0.28.
- Alternative commitment measures:
  - Disclosed: Disclosed coefficient 0.07*; Disclosed×Carbon = -0.10*** (column 2).
  - Green loan: Green loan coefficient 0.20 (not statistically significant); Green loan×Carbon = -2.55*** (column 3).
- Lender commitment:
  - Post SBTi lead coefficient not significant in level (column 4), but interaction Post SBTi lead×Carbon = -0.09* is negative and significant, indicating carbon-intensive borrowers reduce investment if a prior lead arranger becomes green.
- Sample statistics and fit:
  - Observations = 19,574 across capital expenditure regressions.
  - R-squared ranges: 0.907–0.908.
  - Borrower FE, Time FE, and borrower controls included.

### R&D expense findings (Table 6)
- Overall patterns:
  - R&D expenses increase for borrowers with environmental commitments (positive and statistically significant β2 in most specifications).
  - Carbon intensity often positively associated with R&D (Carbon coefficients: 0.03*** in column 1; 0.18** in column 2; 0.02 in column 3; 0.02*** in column 4).
- Interactions:
  - Target×Carbon = -0.26*** indicates the commitment effect on R&D weakens with carbon intensity.
  - Disclosed×Carbon = -0.17* (column 2).
  - Post SBTi lead×Carbon = -0.26*** (column 4).
- Sample statistics and fit:
  - Observations = 22,709.
  - R-squared = 0.980 across specifications.
  - Borrower FE, Time FE, and borrower controls included.

### Cash and short-term investments findings (liquidity hoarding) (Table 7)
- Target effects:
  - Target = -0.10** (column 1) — borrower commitments associated with lower cash holdings.
  - Target×Carbon not significant (0.06).
- Disclosure:
  - Disclosed = -0.05 (not significant) (column 2); Disclosed×Carbon = 0.01.
- Green loan and lender commitments:
  - Green loan = -0.28** (column 3) — green/sustainability-linked tranches associated with lower cash holdings.
  - Post SBTi lead = -0.21*** (column 4) — relationships with lenders who commit to SBTi associated with lower cash holdings.
- Sample statistics and fit:
  - Observations = 45,483.
  - R-squared = 0.880 across specifications.
  - Borrower FE, Time FE, and borrower controls included.
- Interpretation:
  - Pricing benefits associated with borrower commitments increase financial flexibility and reduce incentives to hoard liquidity.

### Environmental expenditure and robustness
- Findings for environmental expenditure (Table A.5) broadly consistent with capital expenditure and R&D expense, suggesting borrowers internalize commitments; caveat: limited sample for these regressions.

### Monetary policy interactions (Table 8 and Figure 3)
- Extended model includes monetary policy shock (MPS) interactions with Carbon, borrower commitments, and lender commitment.
- Key coefficients and interpretations:
  - Carbon coefficient (column 1) = 3.64***.
  - MPS×Carbon (β4) = 3.16*** (column 1): carbon premium increases when monetary policy tightens.
  - Magnitude example: a one standard deviation monetary policy shock is associated with a 1.6 bps additional increase in loan spreads for borrowers at the 90th percentile of carbon intensity.
- Borrower commitments under monetary tightening (column 2):
  - MPS×Target = -2.69* (mitigating effect: loan spreads increase by less for borrowers that committed to targets under monetary tightening).
  - MPS×Target×Carbon = 3.61** (mitigating effect diminishes with carbon intensity).
- Disclosure and green loan interactions:
  - MPS×Disclosed = -4.10*** ; MPS×Disclosed×Carbon = 0.62 (not significant) (column 3).
  - MPS×Green Loan = -12.42*** ; MPS×Green Loan×Carbon = 24.63* (column 4).
- Lender heterogeneity:
  - SBTi member×Carbon = 8.55*** (column 5).
  - MPS×SBTi member×Carbon = -12.23*** (column 5): following monetary tightening, the increase in loan spreads for carbon-intensive firms is smaller when borrowing from SBTi members; however, SBTi members demand significantly higher loan spreads when monetary policy tightens, raising borrowing costs for carbon-intensive firms from SBTi members.
- Table 8 details (selected entries):
  - Observations = 29,130.
  - R-squared ranges 0.610–0.613.
  - Controls: PD, loan controls, lender-time FE, Industry FE, Location FE.

### Time variation in the carbon premium (rolling regressions and Table 9, Figure 4)
- Rolling six-year window regressions (first window 2010–2015, advancing by one year):
  - For windows 2010–2015 to 2014–2019, carbon premium positive and statistically significant and increasing.
  - In subsequent windows the carbon premium declines in magnitude and significance.
  - Final window 2018–2023: coefficient turns negative but remains insignificant.
- Explanations tested via interaction terms with event dummies (Table 9):
  - Paris withdrawal (U.S. withdrawal from Paris Agreement in June 2017): Paris withdrawal×Carbon coefficients not significant (1.47; 1.64).
  - Post COVID (loans originated March 2020 to March 2022): Post COVID×Carbon = -4.76*** (column 3) and -5.08*** (column 4) — post–COVID-19 decline in carbon premium.
  - Post IRA (after Inflation Reduction Act in July 2022): Post IRA×Carbon = -8.69* and -9.02* — IRA associated with weakening of the carbon premium.
- Interpretation:
  - The decline in carbon premium likely reflects pandemic-related expansionary policies and COVID-19 support measures improving financing conditions for carbon-intensive industries, and IRA provisions (tax credits, grants, loan guarantees) encouraging low-carbon investment.
- Table 9 details:
  - Carbon coefficients across specifications: 3.04***, 3.74***, 3.33***, 3.51***.
  - Observations = 29,130.
  - R-squared = 0.610.
  - Controls: PD, loan controls, lender-time FE, Industry FE, Location FE.

### Overall conclusions (section 6)
- Robust evidence that lenders charge higher loan spreads to borrowers with higher carbon intensity.
- Borrower commitments (disclosures, emission reduction targets, green/sustainability-linked loans) and lender characteristics (SBTi membership) substantially influence the magnitude of the carbon premium.
  - Borrowers with commitments receive lower risk spreads, but these discounts decline with borrower carbon intensity.
  - SBTi members charge higher spreads to more carbon-intensive borrowers.
- Real effects:
  - Borrower commitment associated with higher capital expenditure and R&D spending, though effects weaken for more carbon-intensive firms.
  - Committed borrowers hold lower liquidity.
  - Relationships with lenders that commit to environmental objectives can reduce investment for carbon-intensive firms.
- Monetary policy:
  - Tightening monetary policy amplifies the carbon premium (consistent with the risk-taking channel).
- Time trends:
  - The carbon premium has diminished in recent years; evidence suggests COVID-19 policy responses and the Inflation Reduction Act contributed to this decline.

*Source: 3.3  Impact of Environmental Commitments on Corporate Behavior — wpiea2025250-source-pdf*

### References

### References and Appendices

### Appendix A — Additional Tables: Variable Definitions and Data Sources (Panel summaries)
- Panel A. Tranche characteristics (source: DealScan)
  - All-in-spread-drawn: Amount the borrower pays in basis points over LIBOR for each dollar drawn down, which adds the spread of the loan with any annual or facility fee paid to the bank group.
  - Maturity: Maturity of the loan tranche in years.
  - Amount: Logarithm of the tranche amount ($ million).
  - Covenant: Dummy variable equal to 1 if the loan tranche has covenants.
  - Collateral: Dummy variable equal to 1 if the loan tranche is secured.
  - Leveraged: Dummy variable equal to 1 if the tranche is a leveraged loan.
  - Term loan: Dummy variable equal to 1 if it is a term loan.
  - Loan purpose: Dummies for different loan purpose indicators.
  - Green loan: Dummy variable equal to 1 if it is a green loan or sustainability-linked loan.

- Panel B. Borrower characteristics (sources: Urgentem, NUS-CRI, Compustat, Refinitiv)
  - Carbon: Scope 1 and 2 carbon intensity measured as annual scope 1 and 2 carbon emissions over annual revenues (thousand tCO2e/1$ million revenue). (Source: Urgentem)
  - Target: Dummy variable equal to 1 if the borrower has an emission reduction target. (Source: Urgentem)
  - PD: Probability of default. (Source: NUS-CRI)
  - Leverage: Total debt divided by total assets. (Source: Compustat)
  - Size: Logarithm of total assets ($ million). (Source: Compustat)
  - Profitability: Return on assets measured as net income divided by total assets. (Source: Compustat)
  - CAPEX: Logarithm of quarterly capital expenditure ($ million). Quarterly capital expenditure equals reported “CAPXY” in the first fiscal quarter, and the difference between current and lagged “CAPXY” in subsequent quarters (Kumar and Yerramilli, 2023). (Source: Compustat)
  - R&D expense: Logarithm of expenditure on research and development ($ million). (Source: Compustat)
  - Cash: Logarithm of cash and short-term investments ($ million). (Source: Compustat)
  - Environmental expenditure: Logarithm of annual environmental investment and expenditures ($ million). These include spending on environmental protection or measures to control environmental impacts. (Source: Refinitiv)

- Panel C. Lender characteristics
  - SBTi member: Dummy variable equal to 1 if the lead arranger makes commitment or has climate mitigation target under the SBTi. (Source: SBTi)

### Table A.2 — Summary Statistics of Real Effect Outcomes (Firm–Quarter Level)
- Notes: All variables measured at firm–quarter level. CAPEX, R&D expense, and Cash are reported in $ million, and regressions use their logarithms.
- CAPEX ($ million): Mean 54.81, Std.Dev. 234.84, p25 1.81, Median 7.90, p75 31.60
- Log(CAPEX): Mean 2.35, Std.Dev. 1.70, p25 1.03, Median 2.19, p75 3.48
- R&D expense ($ million): Mean 74.21, Std.Dev. 535.09, p25 0.00, Median 4.45, p75 26.46
- Log(R&D expense): Mean 1.90, Std.Dev. 1.95, p25 0.00, Median 1.69, p75 3.31
- Cash ($ million): Mean 1342.85, Std.Dev. 11426.60, p25 43.02, Median 145.43, p75 453.34
- Log(Cash): Mean 4.97, Std.Dev. 1.89, p25 3.78, Median 4.99, p75 6.12

### Table A.3 — Carbon Risk and Loan Pricing: Robustness Checks (Dependent variable: all-in-spread-drawn)
- Explanatory variables include: Carbon (S1) (scope 1), Carbon (scope 1 and 2), and PD (Probability of default). Loan controls include tranche size, maturity, collateral, covenant status, leveraged status, and loan purpose. Standard errors clustered at lender level; ***, **, * denote significance at 1%, 5%, 10% levels.
- Column (1): Carbon (S1) coefficient 3.85*** [0.997]
- Column (2): Carbon coefficient 3.50*** [0.982]; Observations 25,082; R-squared 0.610; Loan Controls Yes; Lender-Time FE Yes; Industry FE Yes; Location FE Yes
- Column (3): Carbon coefficient 3.03*** [0.893]; PD coefficient 478.92*** [61.419]; Observations 29,086; R-squared 0.618; Loan Controls Yes; Lender-Time FE Yes; Industry FE Yes; Location FE Yes
- PD (reported across columns): 481.41*** [61.818] (column 1), 478.34*** [62.150] (column 2), 478.92*** [61.419] (column 3)
- Observations and fit: Observations 29,130 (col 1), 25,082 (col 2), 29,086 (col 3); R-squared 0.610, 0.610, 0.618 respectively.

### Table A.4 — Carbon Risk and Loan Pricing: Additional Robustness Checks (Dependent variable: all-in-spread-drawn)
- Explanatory variables: Carbon (scope 1 and 2) and PD. Various clustering and fixed-effect specifications. Standard errors clustered at lender level and reported in brackets. ***, **, * denote significance at 1%, 5%, 10% levels.
- Column (1): Carbon 3.30*** [1.035]; PD 481.19*** [63.263]; Observations 29,130; R-squared 0.610; Loan controls Yes; Lender-time FE Yes; Industry FE Yes; Location FE Yes
- Column (2): Carbon 3.30*** [1.068]; PD 481.19*** [59.713]; Observations 29,130; R-squared 0.610; Loan controls Yes; Lender-time FE Yes; Industry FE Yes; Location FE Yes
- Column (3): Carbon 3.01*** [1.012]; PD 568.16*** [71.421]; Observations 29,071; R-squared 0.718; Loan controls Yes; Lender-time FE Yes; Industry-time FE Yes; Location FE Yes
- Column (4): Carbon 3.97*** [0.871]; PD 525.56*** [56.141]; Observations 30,227; R-squared 0.537; Loan controls Yes; Lender FE Yes; Time FE Yes; Industry FE Yes; Location FE Yes

### Table A.5 — Impact of Environmental Commitments on Environmental Expenditure (Dependent variable: log annual environmental expenditure $ million)
- Explanatory variables: Carbon (scope 1 and 2), Target (emission reduction target dummy), Disclosed (emission data disclosed dummy), Green loan (green or sustainability-linked loan dummy), Post SBTi lead (borrower previously engaged with lead arranger that subsequently committed/validated a target under the SBTi). All specifications include borrower controls (leverage, total assets, return on assets, each lagged by one quarter). Standard errors clustered at borrower level. ***, **, * denote significance at 1%, 5%, 10% levels.
- Column (1): Carbon 0.28* [0.170]; Observations 448; R-squared 0.9000; Borrower controls Yes; Borrower FE Yes; Year FE Yes
- Column (2): Carbon 0.27 [0.168]; Target 0.32* [0.186]; Target×Carbon -0.24** [0.105]; Observations 448; R-squared 0.8970; Borrower controls Yes; Borrower FE Yes; Year FE Yes
- Column (3): Carbon 0.10 [0.091]; Disclosed 0.11 [0.182]; Disclosed×Carbon -0.17* [0.102]; Observations 448; R-squared 0.9080; Borrower controls Yes; Borrower FE Yes; Year FE Yes
- Column (4): Carbon 0.17 [0.103]; Green loan 0.94** [0.387]; Green loan×Carbon -1.03*** [0.126]; Post SBTi lead 0.07 [0.243]; Post SBTi lead×Carbon -0.39*** [0.127]; Observations 448; R-squared 0.905; Borrower controls Yes; Borrower FE Yes; Year FE Yes

*Carbon Risk in Loan Pricing: Commitment Channels and Real Effects — Working Paper No. WP/2025/250*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025250-source-pdf.pdf_