## The ENV-FIBA Model for Climate Risk Analysis

## Source details

**Canonical URL:** [The ENV-FIBA Model for Climate Risk Analysis](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025230-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025230-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025230-source-pdf.pdf.json)

---

### Model purpose, scope, and outputs
- ENV-FIBA is an integrated micro-macro simulation-based climate-economic model linking a multi-country, multi-sectoral CGE model (IMF-ENV) with a firm-bank micro simulation layer (FIBA).
- Inputs:
  - Temperature and emission targets (pathways) and policy assumptions (e.g., carbon taxation, fiscal revenue recycling and reinvestment).
- Outputs include:
  - Model-implied trajectories for carbon prices (so that desired emission targets are met).
  - Macroeconomic variable paths at industry and economy level.
  - Firm-level balance sheet and profit-and-loss components, default probabilities, loss given default, credit spreads, and other bank-lending metrics.
  - Bank-level capital impacts and drivers/contributions to bank capitalization across the scenario horizon.

### Key methodological relationships and recommendations
- ENV-FIBA integrates physical climate risk (chronic and acute) and transition risk in a single framework, producing sectoral and bilateral trade effects alongside macroeconomic impacts.
- Three methodological recommendations emphasized:
  - (1) Model physical and transition risk effects in an integrated manner, including both costs (decline of “brown” industries) and benefits (growth of “green” industries and mitigation of physical damages).
  - (2) Embed dynamic bank balance sheet dynamics; ENV-FIBA implements a simple dynamic scheme where debt stock to economic output (flow) ratios for industry segments are constant (debt stocks shrink/grow proportionally when industry output falls/rises).
  - (3) Account for interest income effects alongside default risk, since declining lending to shrinking industries reduces bank interest income while growing green industries increase it.
- Complementary recommendation: improve loss given default (LGD) metrics; ENV-FIBA implements an endogenous LGD module but notes it remains simplistic and requires further work.

### Model architecture and data foundations (IMF-ENV CGE core)
- High-level link: IMF-ENV CGE provides scenario-conditional macro and sectoral paths (including carbon taxes) that feed into FIBA to project firm and bank financial flows over a multi-decade horizon.
- IMF-ENV characteristics:
  - Global recursive CGE model; distinguishes >30 economic activities linked to GHG emissions; includes eight electricity generation sources.
  - Neo-classical structure with market-clearing equilibria; factors mobile across sectors (capital excluded); capital stocks vintaged (putty-clay).
  - International trade: Armington specification.
  - Detailed mitigation policy levers: GHG- and activity-specific carbon taxes, multi-country ETS’s, feebates, feed-in subsidies, regulations, energy efficiency.
- Database and aggregation used in study:
  - GTAP-Power: 141 countries and 77 commodities.
  - Version used: 36 activities, 28 commodity sectors, and 26 country/regions.
  - Electricity generation separated into eight power sources: coal, natural gas, oil (diesel), hydro, nuclear, solar, wind and others.
- Climate damages modeling:
  - Chronic physical risks: temperature/precipitation-driven reductions (e.g., labor productivity declines, yield reductions).
  - Acute physical risks: extreme weather event damages, if country-specific data available.
  - Can model energy demand changes, arable land loss, tourism losses, or aggregate GDP reductions from chronic and acute risks.
- Solution approach:
  - Recursive dynamic (sequence of comparative static equilibria); agents are not forward looking; investment driven by savings (household savings + government budget balance + current account).

### FIBA (Firms and Banks) micro-simulation component
- Firm module structure:
  - Linkage equations mapping sectoral macro variables to firm-level variables (e.g., firm sales move proportionally with sector output).
  - Accounting and dynamic balance-sheet equations projecting earnings after tax, cash, assets, and debt over the scenario horizon.
  - Derivation of firm-level risk metrics: PD, LGD, and credit spreads.
- Emissions uncertainty:
  - Few firms disclose GHG emissions; a Monte Carlo simulation scheme is used to simulate firms’ unobserved emissions.
- Accounting and balance-sheet rules:
  - Post-tax profits/losses added to cash and total assets each period.
  - Negative cash shortfalls added to short-term debt.
  - Value of nonfinancial assets assumed constant.
- Firm PD determinants and estimation:
  - Determinants: LEV = (short-term debt + 0.5 × long-term debt) / total assets; ICR = EBIT / interest expense; EBITR = EBIT / assets; CDR = cash / short-term debt.
  - Panel logit specification (firm-fixed effects) for listed firms:
    - logit(PD_f,t) = α_f + β LEV_f,t + γ ICR_f,t + δ EBITR_f,t + θ CDR_f,t + ε_f,t
  - Estimated pre-pandemic (2005-2019, listed firms) coefficients (reproduced exactly):
    - LEV 3.534*** (.0917)
    - ICR -0.01*** (0.0002)
    - EBITR -0.791*** (.1568)
    - CDR -0.016*** (.0037)
    - _cons -6.031*** (.0327)
    - Observations 26,793
    - Within R2 .173
    - Notes: Standard errors in parentheses; *** p<.01, ** p<.05, * p<.1. ICR, EBITR, and CDR are winsorized at 5 (95) percent, while LEV is winsorized at one.
- LGD modeling:
  - Parsimonious Vašíček-distribution-based formula:
    - LGD_f,t = Φ(Φ^{-1}(PD_f,t) − k_f) / PD_f,t
    - k_f = [Φ^{-1}(\overline{PD_f}) − Φ^{-1}(\overline{PD_f} × \overline{LGD_f})] / √(1 − ρ)
- Credit pricing and feedbacks:
  - Loan-pricing identity:
    - PD_f,t × LGD_f,t + (1 − PD_f,t) i_t^{COF} = (1 − PD_f,t) i_f,t
    - Implied cost of debt: i_f,t = i_t^{COF} + [PD_f,t × LGD_f,t] / (1 − PD_f,t) ≡ i_t^{COF} + σ
  - Firms’ cost of debt reacts to PD with a lag (PD at time t affects cost of debt in next period).
- Aggregation to industry-level (debt-weighted):
  - PD_{n,h} = Σ w_f PD_{f,h}; LGD_{n,h} = Σ w_f LGD_{f,h}; σ_{n,h} = Σ w_f σ_{f,h} (w_f based on firm debt).

### Bank module mechanics, losses, and RWAs
- Bank inputs: outstanding loan portfolios by industry segment; exposures L_{n,b,h} grow at sectoral GVA growth g_{n,h}:
  - L_{n,b,h} = L_{n,b,h−1} (1 + g_{n,h})
- NPL stock evolution:
  - NPL_{n,b,h} = NPL_{n,b,h−1} (1 − WROR − CURER) + PD_{n,b,h} × PL_{n,b,h}
  - PL_{n,b,h} = L_{n,b,h} − NPL_{n,b,h}
  - Climate application assumptions: WROR = 100 percent; CURER = 0 percent (PDs taken closer to ultimate bankruptcy rates).
- Provisioning and loan-loss flows (incurred-loss concept):
  - PROV_{n,b,h}^{NPL} = LGD_{n,b,h} × NPL_{n,b,h}
  - LL_{n,b,h} = PROV_{n,b,h}^{NPL} − PROV_{n,b,h−1}^{NPL} + WROO × LGD_{n,b,h} × NPL_{n,b,h−1}
  - Loan loss flows plus interest income flows (i_{n,h} × PL_{n,b,h}) impact bank capital period by period.
- RWAs:
  - Modeled under either standardized approach (STA) or internal ratings-based (IRB).
  - STA: risk weights held constant; RWAs change via migration and loan growth.
  - IRB: risk weights computed using Basel corporate formula with TTC PDs and downturn LGDs as inputs.
  - Assumption: complete pass-through from point-in-time PDs to regulatory TTC PDs.

### Climate risk scenarios and calibration
- Three NGFS-aligned scenarios:
  - Net Zero 2050 (NZ): global warming limited to 1.5°C above pre-industrial; global net-zero CO2 emissions by 2050.
  - Fragmented World (FW): delayed/divergent policy response; currently implemented policies maintained until 2030; thereafter countries with net-zero targets achieve an 80 percent reduction only by 2050.
  - Current Policies (CP): maintains only currently implemented policies; used as business-as-usual baseline for comparisons.
- Calibration notes:
  - IMF WEO macro projections used until the latest available year (2028).
  - NGFS projections used for greenhouse gas emissions and electricity generation by power source.
  - For 2029-2040: assume growth rates of real GDP and total labor supply are the same as of 2028; current account and government balances kept fixed as shares of GDP.
  - Model endogenously estimates carbon tax levels achieving NGFS trajectories.
  - NZ includes Japan-specific assumptions: increased EV penetration and CCUS absorbing 95 million CO2 tons (which is 18 percent toward the total emission reductions for the NZ scenario).
- Chronic physical risk incorporations:
  - CP includes total NGFS-estimated chronic physical risk, representing a reduction of GDP levels by 2040 of 2.7 percent in Japan.
  - FW: chronic physical risk reductions are negligible.
  - NZ: chronic physical risk reduced by around 25 percent of the risks (equivalent to a 0.7 percent higher GDP level in 2040 with respect to the CP scenario).

### Fiscal recycling rules and macro effects
- Fiscal neutrality premise: carbon tax revenues offset by increased government expenditures, reduced taxes elsewhere, or both.
- Two recycling rules:
  - Rule 1: all revenues transferred to households.
  - Rule 2: half allocated to feed-in tariffs for renewables, half transferred to households (emulates Japan’s GX policy).
- Key finding: required carbon tax rates are lower under Rule 2 than Rule 1 because feed-in subsidies accelerate green transition.
- Emissions reductions by 2040 relative to CP:
  - NZ: Total emissions are 60 percent lower by 2040 compared to Current Policies.
  - FW: Total emissions are 30 percent lower by 2040 compared to Current Policies.
- Macroeconomic impacts by 2040:
  - NZ exhibits slightly larger adverse macroeconomic effects by 2040 compared to FW; Rule 2 mitigates some negative repercussions under NZ.
  - Employment impacts remain relatively subdued under NZ.

### Application to Japan — dataset, calibration, and key context
- Japan application scope:
  - CGE coverage: 22 industry segments.
  - FIBA micro component: about 270,000 nonfinancial firms and 22 banks.
- Japan context and policy highlights:
  - Since Fukushima (2011), fossil fuel power generation exceeded 70 percent.
  - Seven emission-intensive industries account for about 80 percent of Japan’s total CO2 emissions.
  - Direct and indirect emission-intensive sectors represent 13 percent of GDP and account for about 22 percent of total bank lending in Japan.
  - Japan interim targets: reduce GHG emissions by 46 percent from 2013 levels by 2030 and net-zero by 2050.
  - Green Transformation Promotion Act (May 2023): upfront investment of JPY 20 trillion over the next decade, funded largely through Japan Climate Transition Bonds; plans to introduce a carbon levy on fossil fuel supplies from FY2028.
  - Emissions trading system in high-emission industries starting FY2026; allowance auctioning to power generation companies phased in from FY2033.
  - Existing carbon tax (Tax for Climate Change Mitigation) set at JPY 289 (approximately $2) per ton of CO2 equivalent.

### Firm and bank calibration details for Japan
- Firm microdata:
  - Moody’s/Orbis dataset period: 2005-2023.
  - Annual panel: between 150,000 to 280,000 firms.
  - Excluded sectors: finance and insurance (NACE Rev. 2, 64-66), public administration (84), activities of households (97-98).
  - Filtering conditions (retained exactly as in source): Cash_and_cash_equivalent (≥0); Operating_revenue_turnover (>0); Total_assets (>0); Operating_P_L_EBIT (≠0); Debt_holding* (>0); P_L_before_tax (≠0); Long_term_debt** (>0); P_L_after_tax (≠0); Costs_of_goods_sold** (≥0); Costs_of_employees** (≥0). Notes: * Debt_holding = Loans + Creditors + Long_term_debt. ** This filtering is only applied to the construction of the “T0” database.
  - “T0” construction: flow variables averaged over up to three years when available; pre-pandemic averages for flow variables and latest data for stock variables; only active firms retained.
  - Missing-value imputations: costs of employees imputed using industry median wage × firm’s number of employees for firms representing 9.8 percent of sample; interest expenses imputed using industry median interest rate × firm’s debt holdings for firms representing 8.2 percent of sample.
- Emission-intensity data:
  - ICE (Scope 1) covers about 360 Japanese firms with emissions in ton CO2/USD million revenue; nonreporting firms use Asia-Pacific NACE Level 4 industry averages.
  - For Japan, 75th and 90th percentiles of residualized emission intensities are about 2 and 3 times greater than the median, respectively.
  - Uncertainty assessed at Q50, Q75, Q90.
- Bank inputs and calibration:
  - Bank industry exposure data as of end-March 2023 from individual banks.
  - 17 portfolio segments constructed for scenario-conditional credit losses and interest income.
  - Average loan exposure to emission-intensive sectors roughly 30 percent (compared to BOJ loan statistic share of 22 percent).
  - Debt interest rate calibration: the debt interest rate for the Japanese corporate sector stood at about one percent in 2022 (used for base lending rate).
  - PDs and LGDs at bank-portfolio level used as anchor points; observed cross-bank pattern: regional banks higher TTC PDs and LGDs than internationally active banks.

### Key Japan simulation findings: firm heterogeneity and uncertainty
- Pre-simulation sectoral heterogeneity examples (sector labels index in source):
  - High leverage examples: fishery (2), electricity (13), iron and steel (8), other manufacturing (18), business services (22).
  - Low leverage examples: gas (14), mining and quarrying (3), chemical products (6).
  - Weak ICR examples: electricity (13), water transport (15), paper products (4), fishery (2).
  - Initial high PDs: non-metallic minerals (7), iron and steel (8), non-ferrous metals (9), fabricated metal products (10), electronic equipment (11), water transport (15).
- Monte Carlo emission-intensity uncertainty:
  - No notable differences between Q50 and Q75 for many outcomes.
  - At Q90, certain emission-intensive sectors (e.g., mining, iron and steel) show large increases in PDs and LGDs, driving sensitivity in bank losses.

### Banking system impacts and transmission to capital (Japan)
- Aggregate capital ratio impacts by 2040 (relative to CP):
  - Under Net Zero 2050 (NZ): aggregate capital ratio estimated to decrease by about 0.6-0.7 percentage points by 2040 relative to CP (≈ decline of around 0.03-0.04 percentage points per annum compared to CP).
  - Under Fragmented World (FW): aggregate capital ratio declines by 0.3 percentage points by 2040; decline during 2030s estimated at about 0.03 percentage points per year (mitigation starts in 2031).
- Heterogeneity:
  - Regional banks experience the most pronounced capital impacts; internationally active banks modestly impacted.
- Contribution decomposition:
  - Non-emission-intensive sectors (e.g., other services) account for approximately 30 percent of the shift in the banking system capital ratio.
  - About one-third to one-half of the capital ratio shift attributed to changes in RWAs, partly from rising risk weights due to higher PDs and LGDs in emission-intensive sectors.
- Sensitivity to emission-intensity uncertainty:
  - No notable differences between Q50 and Q75 for bank capitalization.
  - At Q90, substantial additional declines in bank capital ratios driven by increased loan losses in direct emission-intensive segments.

### Dynamic balance sheet and pricing effects (quantitative illustrations)
- Industry-level loan growth differentials (average annual until 2040, NZ scenario, Rule 2, comparison S5 vs S1):
  - Chemical sector: average annual decrease of 0.8 percentage points in growth.
  - Electricity and gas sector: average annual increase of 1.2 percentage points in growth.
- Effects on bank income and losses (cumulative flows until 2040, S5 minus S1, JPY million) — selected industries:
  - Industry #6: Chemical (Shrinking in S5 relative to S1)
    - Static: Interest income 9,623; Loan loss -29,468; Sum -19,845
    - Dynamic: Interest income -218,213; Loan loss 21,322; Sum -196,890
  - Industry #14: Electricity and Gas (Growing in S5 relative to S1)
    - Static: Interest income -2,153; Loan loss 14,963; Sum 12,811
    - Dynamic: Interest income 598,422; Loan loss -348,908; Sum 249,514
- Aggregate capital ratio effects:
  - Under a dynamic balance sheet, banking system capital ratio in NZ further declines by 0.1 percentage points compared to a static balance sheet.
- Relative magnitudes:
  - Dynamic balance sheet leads to a modest decline in aggregate capital ratio of approximately 0.05 percentage points.
  - Increase in lending rates for higher-credit-risk industries enables banks to generate higher interest income relative to RWAs by approximately 0.15 percentage points.
  - The interest-income effect (~0.15 percentage points) is about three times larger than the static-to-dynamic balance sheet effect (~0.05 percentage points).

### Main caveats, limitations, and recommended model extensions
- Caveats and limitations:
  - Firm module lacks entry and exit dynamics.
  - Additional heterogeneity layers (e.g., physical capital quality, green-technology knowledge intensity) not fully captured.
  - Bank parameter granularity could be improved (more bank-industry specific risk parameters).
  - NGFS acute physical risk estimates for Japan were deemed unreliable and excluded; future NGFS updates could allow inclusion.
- Recommended developments:
  1. Better integrate financial components, foremost interest rates, in the macro (CGE) model.
  2. Consider integrating stock-flow consistent elements with macro-financial structures to bring in bond and equity markets more explicitly.
  3. Improve the LGD component to link LGDs more directly to physical risk implications (damage functions).

### Model code and data inputs (FIBA Matlab package)
- Matlab package pertains to FIBA, consisting of six scripts and 32 functions; requires inputs for nonfinancial firms, banks, CGE scenario outputs, and model parameters.
- Scripts (purpose, input, output) include:
  - Script0_ReadMicroData.m: Input Firms_T0.csv → Output Data_Stage_0.mat
  - Script1_VarSubset.m: Input Data_Stage_0.mat → Output Data_Stage_1.mat
  - Script2_ReadSimParameters.m: Input Data_Stage_1.mat → Output Model_Parameters.xlsx; Data_for_Sim.mat
  - Script3_SimulateFirms.m: Input Data_for_Sim.mat → Output Data_SimFirms.mat; Sim_Results.xlsx; Plots
  - Script4_ReadBankData.m: Input Global_Solvency_Data.xlsx → Output Data_Banks.mat
  - Script5_BankImpact.m: Input Data_Banks.mat → Output Sim_Results.xlsx; Bank_Impacts.xlsx; Plots
- Model parameter inputs include PD model coefficients (industry level), tax rates (industry), emission intensities (firm), anchor point PiT LGDs (industry).
- Bank data inputs include capital, RWA, NPL ratio, industry exposures, PD TTC, LGD PiT, LGD DT, NFC IRB shares, STA RW performing/nonperforming, gross loan growth by industry under scenarios.

*Source: IMF Working Paper — "The ENV-FIBA Model for Climate Risk Analysis", Working Paper No. WP/2025/230 (sections 2.1, 2.3, 3.2, 4, Annex I–II as excerpted from the source PDF).*

### 2.1 Overview ...........................................................................................................

### 2.1 Overview

### Purpose and scope of the ENV-FIBA model
- ENV-FIBA is an integrated micro-macro simulation-based climate-economic model designed to analyze macro-financial consequences of climate scenarios and related policy counterfactuals.
- The framework links a multi-country, multi-sectoral CGE model (IMF-ENV) with a firm-bank micro simulation layer (FIBA).
- Model inputs include temperature and emission targets (pathways) and policy assumptions (e.g., carbon taxation, fiscal revenue recycling and reinvestment).
- Model outputs include:
  - Model-implied trajectories for carbon prices (so that desired emission targets are met).
  - Macroeconomic variable paths at industry and economy level.
  - Firm-level balance sheet and profit-and-loss components, default probabilities, loss given default, credit spreads, and other bank-lending metrics.
  - Bank-level capital impacts and drivers/contributions to bank capitalization across the scenario horizon.

### Key model relationships and contributions
- ENV-FIBA integrates physical climate risk (chronic and acute) and transition risk in a single framework, allowing costs and benefits of mitigation policies to be jointly assessed.
- The model produces sectoral and bilateral trade effects alongside macroeconomic impacts for all industries.
- For banks, the model projects capitalization metrics such as regulatory capital over risk-weighted assets.
- The model is positioned within the literature as a macro-financial model suite with embedded bank stress-test methodologies.

### Three methodological recommendations emphasized
- (1) Physical and transition risk effects should be modeled in an integrated manner.
  - Transition risk exists because of physical climate risk; transition policies both incur costs (decline of “brown” industries) and generate benefits (flourishing “green” industries and mitigation of physical damages).
  - ENV-FIBA integrates both cost and benefit components including links to physical risk.
- (2) Dynamic bank balance sheet dynamics should be considered.
  - Existing models often assume static bank lending portfolio size and composition; ENV-FIBA recommends and embeds a dynamic balance sheet scheme.
  - The embedded simple solution: debt stock to economic output (flow) ratios for industry segments are constant so that debt stocks shrink/grow proportionally when industry output falls/rises.
- (3) Interest income effects should be accounted for alongside default risk.
  - Dynamic balance sheet dynamics imply important interest income effects: declining lending to vanishing (brown) industries reduces bank interest income over time, while greener industries increase it.
  - Credit risk effects (defaults) may be less sizable in many climate scenarios, which often evolve over 10-20 years or longer.
- Complementary recommendation: improve loss given default (LGD) metrics in climate risk models.
  - ENV-FIBA implements an LGD module allowing LGDs to move endogenously and deteriorate under adverse climate scenarios, but notes this remains simplistic and requires further work.

### Model architecture and data foundations
- High-level framework: IMF-ENV CGE model provides scenario-conditional macro and sectoral paths (including carbon taxes) that feed into the FIBA micro-simulation layer to project firm and bank financial flows and balance-sheet dynamics over a multi-decade horizon.
- IMF-ENV model characteristics:
  - Global recursive CGE model operated by IMF Research Department.
  - Distinguishes between more than 30 economic activities linked to GHG emissions and includes eight electricity generation sources.
  - Neo-classical structure: utility-maximizing households, profit-maximizing firms, circular flow rationale, market-clearing equilibrium for factors, goods, and services.
  - Factors mobile across sectors (capital excluded); capital stocks have vintages (putty-clay specification).
  - International trade modeled with Armington specification.
  - Capable of detailed mitigation policies: GHG- and activity-specific carbon taxes, multi-country ETS’s, feebates, feed-in subsidies, regulations, energy efficiency measures.
  - Accounts for both transition risk (economic costs of mitigation) and physical risk (economic damages averted by mitigation).
- Climate damages in IMF-ENV:
  - Chronic physical risks: temperature/precipitation-driven reductions (e.g., labor productivity declines in agriculture/construction, yield reductions for crops).
  - Acute physical risks: extreme weather event damages, if country-specific data available.
  - Can model changes in energy demand (heating/air conditioning), arable land loss (sea level rise), tourism losses, or aggregate GDP reductions from chronic and acute risks.
- Model solution and database:
  - Recursive dynamic (sequence of comparative static equilibria); agents are not forward looking; investment driven by savings (household savings + government budget balance + current account).
  - Central input: GTAP-Power database with country-specific input-output tables.
  - Database coverage and model aggregation as used in this study:
    - GTAP-Power: 141 countries and 77 commodities.
    - Version used: 36 activities, 28 commodity sectors, and 26 country/regions.
    - Electricity generation separated into eight power sources: coal, natural gas, oil (diesel), hydro, nuclear, solar, wind and others.

### Application to Japan (overview from this section)
- The paper presents an application of ENV-FIBA to Japan used in the IMF (2024) FSAP context.
- Japan application scope and sample:
  - 22 industries.
  - 270,000 Japanese nonfinancial firms.
  - 22 Japanese banks.
- Primary empirical questions for Japan:
  - What will be the impact of climate mitigation policies (carbon taxation etc.) on firms and their banks in Japan by 2040?
  - How heterogeneous will these impacts be across firms, industry segments, and banks?

*Source: IMF Working Papers — The ENV-FIBA Model for Climate Risk Analysis, Section 2.1 Overview.*

### 2.3 Firms and Banks (FIBA) Model Component

### 2.3 Firms and Banks (FIBA) Model Component

### Model structure and firm-level micro-simulation
- The firm-level component (the FI in FIBA) is a micro simulation anchored in firm-level data to derive paths for firm-level risk metrics conditional on climate scenarios and CGE model outputs.
- The firm module is divided into three parts:
  - Linkage equations that map macro model sectoral variables to firm-level variables (structural links). Example: a firm’s sales revenues move proportionally with output growth for the sector it belongs to.
  - Basic accounting equations and dynamic balance sheet equations that project earnings after tax, cash, assets, and debt over the scenario horizon.
  - Derivation of firm-level risk metrics: probability of default (PD), loss-given default (LGD), and credit spreads, to feed into bank impacts.
- Emissions uncertainty:
  - Few firms disclose GHG emissions; substantial uncertainty exists at firm level.
  - Capelle et al. (2023) find within-industry variation in emission intensities comparable to heterogeneity in total factor and labor productivity.
  - A Monte Carlo simulation scheme is used to simulate firms’ unobserved emissions.

### Accounting, balance-sheet dynamics, and treatment of cash shortfalls
- Firm periodic cash flow (earnings after tax) is the sum of projected income (+) and expenses (−) along the scenario horizon; direct costs associated with a firm’s direct emissions are included in earnings.
- Post-tax profits or losses are added to firms’ cash and cash equivalents and total assets, with balance sheets simulated period by period.
- If a firm’s cash holdings become negative, the shortfall is added to the firm’s short-term debt.
- The value of nonfinancial assets is assumed constant by assumption.

### Firm risk metrics: determinants and estimation
- Determinants considered for firms’ PDs (based on stock and flow–oriented Merton rationale) include:
  - LEV: leverage ratio = (short-term debt + 0.5 × long-term debt) / total assets.
  - ICR: interest coverage ratio = EBIT / interest expense.
  - EBITR: EBIT to assets ratio.
  - CDR: cash to short-term debt ratio.
- Panel econometric specification (firm-fixed effects, logit transform) for listed subset (source: Moody’s KMV):
  - logit(PD_f,t) = α_f + β LEV_f,t + γ ICR_f,t + δ EBITR_f,t + θ CDR_f,t + ε_f,t
  - Estimated coefficients may differ across industry segments.
  - All right-hand-side variables are structural and linked to CGE outputs.

### LGD modeling approach
- LGDs are modeled parsimoniously following Frye and Jacobs (2012), assuming structural correlation between default rates and loss rates and common-parameter distributions in the asymptotic portfolio.
- Under a Vašíček distribution for losses, LGD is given as:
  - LGD_f,t = Φ(Φ^{-1}(PD_f,t) − k_f) / PD_f,t
  - where k_f = [Φ^{-1}(\overline{PD_f}) − Φ^{-1}(\overline{PD_f} × \overline{LGD_f})] / √(1 − ρ)
  - Φ denotes the standard normal cumulative distribution function; k_f is driven by through-the-cycle (TTC) PDs and LGDs and the asset correlation parameter ρ.

### Firm-level credit pricing and feedback to funding costs
- Loan-pricing identity relates expected loss, expected interest expense, and expected interest income:
  - PD_f,t × LGD_f,t (Expected Loss) + (1 − PD_f,t) i_t^{COF} (Expected Interest Expense) = (1 − PD_f,t) i_f,t (Expected Interest Income)
  - Implied cost of debt:
    - i_f,t = i_t^{COF} + [PD_f,t × LGD_f,t] / (1 − PD_f,t) ≡ i_t^{COF} + σ
  - i_f,t is the firm cost of debt; i_t^{COF} is banks’ cost of funding; σ is the risk premium.
- Two-way contemporaneous dependence between firms’ cost of debt and PDs is simplified: firms’ cost of debt reacts to PD with a lag (PD at time t affects cost of debt in next period).

### Aggregation to industry-level firm risk metrics
- After computing PDs, LGDs, and credit risk spreads for firms, industry-specific debt-weighted averages are computed:
  - PD_{n,h} = Σ w_f PD_{f,h} (for f ∈ n)
  - LGD_{n,h} = Σ w_f LGD_{f,h} (for f ∈ n)
  - σ_{n,h} = Σ w_f σ_{f,h} (for f ∈ n)
  - Weights w_f are based on individual firms’ level of debt.

### Bank module: data, dynamics, and losses
- The bank module (BA in FIBA) requires banks’ outstanding loan portfolios by industry segment, ideally aligned with CGE industry definitions and including separation of emission-intensive sectors.
- Bank sectoral loan exposures L_{n,b,h} grow at the same rate g_{n,h} as scenario-conditional sectoral GVA:
  - L_{n,b,h} = L_{n,b,h−1} (1 + g_{n,h})
  - This reflects medium- to long-run deleveraging in declining industries and leveraging in expanding industries, affecting interest income.
- Nonperforming loan stock evolution for bank b’s exposure to sector n at horizon h:
  - NPL_{n,b,h} = NPL_{n,b,h−1} (1 − WROR − CURER) + PD_{n,b,h} × PL_{n,b,h}
  - Performing exposures PL_{n,b,h} = L_{n,b,h} − NPL_{n,b,h}
  - For the climate risk application: WROR = 100 percent; CURER = 0 percent (PDs interpreted closer to ultimate bankruptcy rates).
- Provisioning and loan-loss flows (incurred-loss concept; cumulative over scenario horizon):
  - PROV_{n,b,h}^{NPL} = LGD_{n,b,h} × NPL_{n,b,h}
  - LL_{n,b,h} = PROV_{n,b,h}^{NPL} − PROV_{n,b,h−1}^{NPL} + WROO × LGD_{n,b,h} × NPL_{n,b,h−1}
  - Loan loss flows plus interest income flows (i_{n,h} × PL_{n,b,h}) impact bank capital period by period.
- At simulation start, observed PDs and LGDs at bank-portfolio level are used as anchor points; over the simulation PDs and LGDs are aligned with debt-weighted average firm trajectories.
- Risk-weighted assets (RWAs):
  - Modeled under either the standardized approach (STA) or internal ratings-based (IRB) approach.
  - STA: risk weights held constant; absolute RWAs change via migration effects and loan growth.
  - IRB: risk weights computed using Basel corporate formula with TTC PDs and downturn LGDs as inputs.
  - Assumption: complete pass-through from point-in-time PDs to regulatory TTC PDs is applied.

### Design of climate scenarios and model inputs (overview)
- IMF-ENV compares a business-as-usual (baseline) scenario without future climate policies with a policy counterfactual scenario that includes climate and energy policies.
- Baseline construction uses GTAP-Power database (base year 2017), historical macro and climate-related data to project up to the most recent year (e.g., 2023), and IMF WEO projections to extend to end of simulation horizon (2040).
- External inputs include estimations on future emission paths, global temperature effects, climate damages, electricity demand and supply, and exogenous technology changes (e.g., CCUS, EVs, hydrogen).
- NGFS scenario outcomes are used as a reference; calibration aligns baseline and policy scenarios to country- or region-specific total GHG emission paths and, when available, projections on electricity demand, supply, and technology deployment.

### Application to Japan: context and calibration highlights
- Relevance:
  - Since Fukushima (2011), Japan has mainly relied on fossil fuels for power generation; fossil fuel power generation exceeded 70 percent.
  - Seven emission-intensive industries account for about 80 percent of Japan’s total CO2 emissions.
  - Direct and indirect emission-intensive sectors represent 13 percent of GDP and account for about 22 percent of total bank lending in Japan.
- Policy context:
  - Japan set an interim target to reduce GHG emissions by 46 percent from 2013 levels by 2030 and to achieve net-zero GHG emissions by 2050.
  - Green Transformation Promotion Act (May 2023) outlines upfront investment of JPY 20 trillion over the next decade, funded largely through Japan Climate Transition Bonds, and plans to introduce a carbon levy on fossil fuel supplies from FY2028.
  - Japan is set to implement an emissions trading system in high-emission industries starting from FY2026; allowance auctioning to power generation companies planned to be phased in from FY2033.
  - Existing carbon tax (Tax for Climate Change Mitigation) set at JPY 289 (approximately $2) per ton of CO2 equivalent.
- Japan-specific simulation:
  - IMF-ENV derives scenario-conditional macro and sectoral paths (including carbon taxes) up to 2040.
  - Micro-simulation layer: about 270,000 nonfinancial firms from Japan, focusing on emission-intensive sectors.
  - Bank module comprises 22 banks.

*IMF Working Paper — excerpt: 2.3 Firms and Banks (FIBA) Model Component*

### 3.2 Climate Risk Scenarios

### 3.2 Climate Risk Scenarios

### Scenario definitions and baseline
- Three focal NGFS-aligned scenarios:
  - Net Zero 2050 (NZ): global warming limited to 1.5°C above pre-industrial levels; global net-zero CO2 emissions by 2050.
  - Fragmented World (FW): delayed and divergent climate policy response; currently implemented policies maintained until 2030 (delayed transition); thereafter, countries with net-zero targets achieve an 80 percent reduction only by 2050, while others continue with current policies (divergent transition).
  - Current Policies (CP): maintains only currently implemented policies; serves as the reference (baseline) scenario.
- CP is used as the business-as-usual baseline for comparisons.

### Modeling assumptions and calibration
- Macroeconomic and emissions inputs:
  - IMF WEO macroeconomic projections used until the latest available year (2028).
  - NGFS projections used for overall greenhouse gas emissions and electricity generation by power source.
  - For 2029-2040: assume growth rates of real GDP and total labor supply are the same as of 2028; current account and government balances kept fixed as shares of GDP.
- Scenario-specific model features:
  - The model endogenously estimates carbon tax levels that achieve targeted greenhouse gas trajectories for each NGFS scenario.
  - NZ scenario includes Japan-specific green technology progress assumptions:
    - Increase in electric vehicle penetration.
    - Carbon capture technology (CCUS) absorbs 95 million CO2 tons, which is 18 percent toward the total emission reductions for the NZ scenario.
- Chronic physical risk incorporation:
  - GDP losses from chronic physical risk (NGFS estimates) are incorporated via adjustments in Total Factor Productivity (TFP) growth.
  - CP scenario includes the total NGFS-estimated chronic physical risk, which represents a reduction of GDP levels by 2040 of 2.7 percent in Japan.
  - Alternative scenarios adjust this risk:
    - FW: chronic physical risk reductions are negligible.
    - NZ: chronic physical risk reduced by around 25 percent of the risks (equivalent to a 0.7 percent higher GDP level in 2040 with respect to the CP scenario).

### Carbon tax revenue recycling (fiscal neutrality) — Rule definitions and effects
- Fiscal neutrality premise: carbon tax revenues must be offset by increased government expenditures, reduced taxes elsewhere, or both.
- Two revenue recycling rules in the CGE model:
  - Rule 1: all generated revenues are transferred to households.
  - Rule 2: half of revenues allocated to feed-in tariffs for the renewable energy sector, half transferred to households (designed to emulate the Government of Japan’s Green Transformation (GX) policy).
- Key finding:
  - Required carbon tax rates are lower under Rule 2 than Rule 1 because feed-in subsidies accelerate green transition of the energy sector.

### Macroeconomic impacts by 2040
- Comparative impacts:
  - The NZ scenario exhibits slightly larger adverse macroeconomic effects by 2040 compared to the FW scenario.
  - Rule 2 mitigates some negative repercussions under NZ.
  - Employment impacts remain relatively subdued under NZ.
- Emissions reductions relative to CP by 2040:
  - NZ: Total emissions are 60 percent lower by 2040 compared to Current Policies.
  - FW: Total emissions are 30 percent lower by 2040 compared to Current Policies.
  - Mitigation efforts in FW are designed to start from 2031.

### Sectoral outcomes and transmission channels
- Sectoral GVA and output responses depend on:
  - Direct emission intensity.
  - Inter-industry linkages and price adjustments from carbon taxes.
- Representative sectoral outcomes relative to CP:
  - Large declines in GVA for emission-intensive sectors: natural gas, petroleum and coal, chemical products, iron and steel, and air transport.
  - Electricity sector overall thrives, with divergent outcomes across sub-sectors:
    - Coal and gas power see reduced GVA.
    - Renewable energy sources experience growth.
  - In some sectors (e.g., land transport and business services), output (real and price effects) increases relative to CP, but GVAs decline because intermediate input rises more strongly than output.
- Labor costs show sectoral variation similar to GVA but with less pronounced changes.

### Firm-level risk, heterogeneity, and uncertainty
- Pre-simulation heterogeneity in firm metrics (examples by sector):
  - High leverage: fishery (2), electricity (13), iron and steel (8), other manufacturing (18), business services (22).
  - Low leverage: gas (14), mining and quarrying (3), chemical products (6).
  - Weak ICR (interest coverage): electricity (13), water transport (15), paper products (4), fishery (2).
  - Weak EBIT to total asset ratio (EBITR): agriculture (1), water transport (15), paper products (4), collective services (21).
  - Weak cash-debt positions: iron and steel (8), non-ferrous metals (9), paper products (4).
- Initial PD patterns:
  - High initial PDs among emission-intensive sectors: non-metallic minerals (7), iron and steel (8), non-ferrous metals (9), fabricated metal products (10), electronic equipment (11), water transport (15).
  - Elevated PDs in some non-emission-intensive sectors: water supply (20), collective services (21), business services (22).
- Emission-intensity uncertainty:
  - A Monte Carlo scheme around firms’ emission intensity (for firms lacking data) produces distributions of firm risk metrics.
  - No notable differences between the 50th (Q50) and 75th (Q75) percentiles.
  - At the 90th percentile (Q90), certain emission-intensive sectors (e.g., mining, iron and steel) show large increases in PDs and LGDs, highlighting sensitivity to emission-intensity uncertainty.

### Banking system impacts and dynamics
- Aggregate impacts by 2040:
  - Under Net Zero 2050:
    - Banking system aggregate capital ratio estimated to decrease by about 0.6-0.7 percentage points by 2040 relative to CP.
    - This translates to a decline of around 0.03-0.04 percentage points per annum compared to CP.
  - Under Fragmented World:
    - Aggregate capital ratio declines by 0.3 percentage points by 2040.
    - Given mitigation starts in 2031, the decline in capital ratios during the 2030s is estimated at about 0.03 percentage points per year (comparable to NZ annualized).
- Heterogeneity across banks:
  - Regional banks in the sample experience the most pronounced capital impacts.
  - Internationally active banks are modestly impacted because of more favorable borrower risk profiles.
- Contribution decomposition and drivers:
  - Non-emission-intensive sectors (e.g., other services) account for approximately 30 percent of the shift in the banking system capital ratio.
  - About one-third to one-half of the capital ratio shift can be attributed to changes in risk-weighted assets (RWAs), partly due to rising risk weights from higher PDs and LGDs in emission-intensive sectors.
- Sensitivity to firm emission-intensity uncertainty:
  - No notable differences between Q50 and Q75 for bank capitalization.
  - At Q90, substantial additional declines in bank capital ratios driven by increased loan losses in direct emission-intensive segments.

### Dynamic balance sheet effects and pricing in credit risk
- Dynamic versus static balance sheet (NZ scenario, Rule 2, Q50 comparison S5 vs S1):
  - Industry-level loan growth differentials (average annual until 2040):
    - Chemical sector: average annual decrease of 0.8 percentage points in growth.
    - Electricity and gas sector: average annual increase of 1.2 percentage points in growth.
  - Effects on bank income and losses (cumulative flows until 2040, S5 minus S1, JPY million):
    - Industry #6: Chemical (Shrinking in S5 relative to S1)
      - Static: Interest income 9,623; Loan loss -29,468; Sum -19,845
      - Dynamic: Interest income -218,213; Loan loss 21,322; Sum -196,890
    - Industry #14: Electricity and Gas (Growing in S5 relative to S1)
      - Static: Interest income -2,153; Loan loss 14,963; Sum 12,811
      - Dynamic: Interest income 598,422; Loan loss -348,908; Sum 249,514
- Aggregate capital ratio effects:
  - Under a dynamic balance sheet, the banking system’s capital ratio in NZ further declines by 0.1 percentage points compared to a static balance sheet (zero gross loan growth).
- Relative magnitudes: pricing in credit risk vs dynamic balance sheet
  - Dynamic balance sheet leads to a modest decline in the aggregate capital ratio of approximately 0.05 percentage points.
  - Increase in lending rates for industries with higher credit risk enables banks to generate higher interest income relative to RWAs by approximately 0.15 percentage points.
  - The interest-income effect (~0.15 percentage points) is about three times larger than the effect of moving from static to dynamic balance sheets (~0.05 percentage points).

### Caveats and model limitations
- Firm module limitations:
  - Lacks firm entry and exit dynamics.
  - Additional layers of heterogeneity (e.g., quality of physical capital, knowledge intensity in green technology) are not fully captured and may affect outcomes.
- Bank parameter granularity:
  - More granular bank-industry specific risk parameters would improve refinement of results.
- Acute physical risks:
  - NGFS acute physical risk estimates were deemed unreliable for Japan and were not included; future NGFS updates could allow inclusion.

*Source: IMF staff calculations using NGFS scenarios and IMF WEO projections as described in the chapter.*

### 4. Conclusions

### 4. Conclusions

### Model design and distinguishing features
- The ENV-FIBA model is an integrated micro-macro simulation model with:
  - a CGE macro model core, and
  - a micro simulation module for a sample of nonfinancial firms and banks.
- Focal distinguishing features relative to related models:
  1. the account for chronic physical risk feedback in the otherwise transition-risk focused model;
  2. the use of a dynamic balance sheet mode for firm debt and hence bank balance sheets;
  3. the emphasis on interest income effects stemming from shrinking vs. growing industries.
- Three additional noteworthy features built into the model:
  - an endogenous LGD model scheme (albeit still simple);
  - a stochastic drawing mechanism to account for uncertainty in individual firms’ emission intensities;
  - incorporation of all industry segments (not just emission-intensive ones) to include industries that may gain from transition policies.

### Application to Japan — setup and scenarios
- CGE model coverage: 22 industry segments of the Japanese economy.
- FIBA micro component: 270,000 individual firms and 22 individual banks.
- Three scenarios considered: NGFS’s current policies scenario; the fragmented world scenario; the net zero 2050 scenario.

### Key findings on physical risk and transition policy effects
- Accounting for chronic physical risk feedback allows transition policies to:
  - mitigate physical risk materialization, and
  - produce positive economic effects that can offset economic losses induced by transition policies when viewed in isolation.
- Example quantitative effect: the “current policy” scenario for Japan accounts for the macro impact of estimated chronic physical risk from the NGFS, which represents a reduction of GDP levels of 2.7 percent by 2040 for Japan.
- Under alternative scenarios (e.g., net zero 2050), physical risk impacts are less detrimental, revealing net aggregate benefits of transition policies.
- The physical risk benefits of the net zero 2050 scenario become increasingly visible after 2040.
- Aggregate reallocation: emission-intensive industries materially shrink while other industries benefit.

### Dynamic firm debt, bank profitability, and interest income effects
- Considering dynamic evolution of firm debt is important to capture industry structure shifts and effects on bank profitability.
- Detailed industry comparisons (Japan, net zero 2050 vs. current policies scenario):
  - chemical products sector: average annual decrease of 0.8 percentage points in growth per annum until 2040;
  - electricity and gas sector: increase of 1.2 percentage points of growth per annum until 2040.
- Resulting bank-level effects:
  - chemical sector shrinkage → decline in bank interest income; loan losses become less negative even though PDs and LGDs deteriorate relative to current policies scenario;
  - electricity and gas sector growth → interest income increases; loan losses become more negative despite reductions in credit risk parameters relative to the current policy scenario.
- Loan growth impact on bank capital:
  - more positive loan growth exerts downward pressure on capital ratios (via risk weighted assets);
  - less positive or negative loan growth exerts upward pressure on capital ratios from this perspective.
- Conclusion: dynamic balance sheets, even implemented simply, are instrumental to capture quantitatively important interest income effects of shrinking vs. growing industries.

### Model inputs, calibration, and microdata highlights (Japan)
- Firm microdata (Moody’s/Orbis):
  - dataset period: 2005-2023;
  - annual panel: between 150,000 to 280,000 firms;
  - excluded sectors: finance and insurance (NACE Rev. 2, 64-66), public administration (84), activities of households (97-98).
- Consolidation coding and selection: preference order C2-C1-U1; U2 not used when consolidated results are available.
- Filtering conditions (examples retained exactly as in source):
  - Cash_and_cash_equivalent (≥0)
  - Operating_revenue_turnover (>0)
  - Total_assets (>0)
  - Operating_P_L_EBIT (≠0)
  - Debt_holding* (>0)
  - P_L_before_tax (≠0)
  - Long_term_debt** (>0)
  - P_L_after_tax (≠0)
  - Costs_of_goods_sold** (≥0)
  - Costs_of_employees** (≥0)
  - Notes: * Debt_holding = Loans + Creditors + Long_term_debt. ** This filtering is only applied to the construction of the “T0” database.
- “T0” database construction:
  - flow variables averaged over up to three years when available;
  - pre-pandemic averages for flow variables and latest data for stock variables used to mitigate pandemic distortions;
  - only active firms retained (defaults, merges, or resolved entities removed).
- Missing-value imputations:
  - costs of employees imputed using industry median wage at NACE Level 2 × firm’s number of employees for firms representing 9.8 percent of sample;
  - interest expenses imputed using industry median interest rate × firm’s debt holdings for firms representing 8.2 percent of sample.
- Emission-intensive sector shares in T0 dataset (Japan):
  - 5.6 percent of the total number of firms;
  - 42 percent of total assets;
  - 26 percent of total sales within the firm sample;
  - according to the Japan Industrial Productivity database, these sectors comprise 13.4 percent of total GVA.

### PD, emissions, and uncertainty treatment
- PD data source: Moody’s KMV, covering approximately 1,900 listed Japanese firms from 2005 to 2020; daily PDs converted to yearly averages.
- CO2 emissions (Scope 1) data from Intercontinental Exchange (ICE):
  - ICE dataset includes about 360 Japanese firms with firm-level reported emission intensities measured in ton CO2/USD million revenue;
  - firms lacking reported emissions use industry-average values for NACE Level 4 industries in the Asia-Pacific region from ICE.
- Heterogeneity in emission intensities:
  - substantial within-sector variation observed (e.g., mining (3) and non-metallic minerals (7));
  - the kernel density distribution of firm emission intensity residuals from Capelle et al. (2023) is used to conduct a Monte Carlo simulation;
  - for Japan, the 75th and 90th percentiles of “residualized” emission intensities are about 2 and 3 times greater than the median, respectively.
- Uncertainty assessment: impacts of emission-intensity uncertainty on firm risk metrics assessed at the 50th, 75th, and 90th percentiles.

### Firm P&L, PD model, and key econometric results
- Firm revenue and cost projections:
  - sales revenues move at the same rate as sectoral output projections from the IMF ENV;
  - costs of employees linked to growth of sectoral labor income computed from sectoral employment × equilibrium wage from IMF ENV;
  - operating expenses net of employee costs linked to sectoral intermediate input (output minus GVA);
  - firm CO2 emissions (emission intensity × sales revenues) move with sectoral emission projections from IMF ENV;
  - income tax expense computed based on net income before tax using micro-data-informed industry-specific rates; taxes applied only when earnings before tax are positive; tax credits not considered.
- PD model specification (firm-fixed effects panel with logit-transformed PDs):
  - logit(PD_ft) = α_f + β LEV_ft + γ ICR_ft + δ EBITR_ft + θ CDR_ft + ε_ft
- PD estimation (pre-pandemic 2005-2019, listed firms):
  - coefficient estimates statistically significant with expected signs;
  - normalized coefficient for leverage (normalized by its standard deviation) is about twice as large as normalized coefficients for all other regressors combined.
- Estimation results (Annex I. Table 6) reproduced exactly:
  - (1) logit (PD)
    - LEV 3.534*** (.0917)
    - ICR -0.01*** (0.0002)
    - EBITR -0.791*** (.1568)
    - CDR -0.016*** (.0037)
    - _cons -6.031*** (.0327)
    - Observations 26,793
    - Within R2 .173
  - Notes preserved: Standard errors in parentheses; *** p<.01, ** p<.05, * p<.1. ICR, EBITR, and CDR are winsorized at 5 (95) percent, while LEV is winsorized at one.
- For firms outside the PD estimation sample (primarily unlisted), industry averages of estimated firm fixed effects are used; historical PDs and through-the-cycle (TTC) PDs computed based on estimated panel model and firm risk metrics.

### Bank module inputs, portfolio structure, and calibration highlights
- Bank industry exposure data as of end-March 2023 sourced from individual banks.
- Focus on emission-intensive sectors and sectors reliant on emission-intensive inputs for 22 banks, using Bank of Japan classification of Loans and Bills Discounted by Sector.
- 17 distinct portfolio segments constructed for computing scenario-conditional credit losses and interest income.
- Aggregation approach: PDs, LGDs, and credit spreads reaggregated to match 17 loan portfolios using each firm’s debt as weighting factor.
- Loan exposure to emission-intensive sectors:
  - composition varies significantly across banks;
  - internationally active banks have higher exposure to emission-intensive sectors compared to other bank types;
  - average loan exposure to these sectors is roughly 30 percent (compared to BOJ loan statistic share of 22 percent).
- PDs and LGDs at bank level for NFC portfolio used as “anchor points” to account for structural cross-bank differences in risk profile.
- Observed cross-bank pattern: regional banks tend to exhibit higher risk parameters (TTC PDs and LGDs) relative to other bank types.
- Debt interest rate calibration: the debt interest rate for the Japanese corporate sector, which stood at about one percent in 2022, is used for the “base” lending rate.
- General loan growth alignment: gross loan growth aligned with a rolling multi-year average of scenario-based sectoral GVA growth for each industry.
- Alternative calibration approach using PDs and LGDs at bank-portfolio level was partially tested and yielded quantitatively similar outcomes for impacts on banks’ capitalization.

### Model development avenues (recommended extensions)
- Three example further developments:
  1. better integrate financial components, foremost interest rates, in the macro (CGE) model;
  2. consider integrating stock-flow consistent model elements with macro-financial structures to bring in bond markets and equity markets and capture their role more explicitly;
  3. improve the LGD model component for banks and their lending portfolios to more directly link LGDs to physical risk implications (as output of damage functions).

*Source: IMF Working Paper — "The ENV-FIBA Model for Climate Risk Analysis", Section 4. Conclusions (source PDF).*

### Annex I. Figure 4. Japan: Banks’ Loan Portfolios by Sector and Risk Profile by Type of Banks

### Annex I. Figure 4. Japan: Banks’ Loan Portfolios by Sector and Risk Profile by Type of Banks

### Key findings from the figure and panels
- Internationally active banks have higher exposures to emission-intensive sectors, compared to other types of banks.
- NPL ratios for NFC portfolios of regional banks in the sample are higher than those of internationally active banks.
- Through-the-cycle (TTC) PDs are, on average, higher and more diverse for regional banks compared to internationally active banks.
- Downturn LGDs for NFC portfolios are higher for regional banks relative to internationally active banks.
- Point-in-time (PiT) LGDs for NFC portfolios are higher for regional banks relative to internationally active banks.

### Notes on data visualization and timing
- In the top panels, the loan exposure data is as of March 2023.
- In the box plots:
  - lines in the middle of the box are medians,
  - box edges are the 25th/75th percentiles,
  - the ends of the whiskers mark the 1st/99th percentiles.

### LGD estimation and adjustment methodology (summary from notes)
- To estimate LGDs at the bank-portfolio level at the outset:
  - provision coverage ratios (PCR) for nonperforming loans are first computed within three industry clusters based on aggregated bank data:
    - agriculture and fishery,
    - manufacturing,
    - the rest of the nonfinancial corporate industries,
    - as well as for the overall NFC total.
  - Differences between the PCR of each industry cluster and the PCR of the overall NFC portfolio are used to adjust both point-in-time LGDs and downturn LGDs of individual banks at the outset.
  - Rationale: LGDs are not available at the detailed industry level at the outset, while NPL provision coverage ratios are.

### Lending rate composition (note)
- The lending rate is composed of:
  - a “base lending rate” and
  - the “credit spread” component on top.

### Sources and calculations
- Sources: FSA; and IMF staff calculations.

---

### Annex II. Model Codes (FIBA model component)

### Overview of the Matlab package
- The Matlab model code package—pertaining to the FIBA model component—accompanies this paper.
- It consists of:
  - six scripts and
  - 32 accompanying functions.
- The script package requires data inputs for:
  - nonfinancial firms,
  - banks,
  - scenario inputs that are the outputs from the CGE model component, and
  - other model input parameters.
- The model codes are available from the authors on request.

### Script list (purpose, input, output)
- Script0_ReadMicroData.m
  - Purpose: Process raw micro data from CSV
  - Input: Firms_T0.csv
  - Output: Data_Stage_0.mat
- Script1_VarSubset.m
  - Purpose: Extract subset of relevant model variables
  - Input: Data_Stage_0.mat
  - Output: Data_Stage_1.mat
- Script2_ReadSimParameters.m
  - Purpose: Read model parameters
  - Input: Data_Stage_1.mat
  - Output: Model_Parameters.xlsx; Data_for_Sim.mat
- Script3_SimulateFirms.m
  - Purpose: Firm P&L and Balance Sheet Simulator
  - Input: Data_for_Sim.mat
  - Output: Data_SimFirms.mat; Sim_Results.xlsx; Plots
- Script4_ReadBankData.m
  - Purpose: Read and process banks' T0 data
  - Input: Global_Solvency_Data.xlsx
  - Output: Data_Banks.mat
- Script5_BankImpact.m
  - Purpose: Estimate impact on banks
  - Input: Data_Banks.mat
  - Output: Sim_Results.xlsx; Bank_Impacts.xlsx; Plots

### Nonfinancial firm micro data input
- The nonfinancial firm data as input to Script 0 takes the form depicted in the referenced input figure.
- The content of this data sheet was summarized in Annex 1 (see text and Table 2 therein).

### Model parameters required by Script 2 (Table 2)
- #1 PD model coefficients
  - Level: Industry
  - Comments: Slope coefficients pertaining to logit-PD model component, by industry. They can be homogeneous or heterogeneous across industries. Intercepts (nonfinancial firm fixed effects) are not included in this input file, but in the firm-level input depicted in the nonfinancial firm micro data input.
- #2 Tax rates
  - Level: Industry
  - Comments: Income tax rates, by industry
- #3 Emission intensities
  - Level: Firm
  - Comments: Emission intensity for individual firms (firm micro data)
- #4 Anchor point PiT LGDs
  - Level: Industry
  - Comments: Firm industry-level LGDs, as of most recent. These are needed in the nonfinancial firm micro simulation (Script 3), since firm-level LGDs are not initially observed.

### Bank data inputs required by Script 4 (Table 3)
- #1 Capital
  - Level: Banks
  - Comments: Banks’ regulatory capital as of T0, in local currency
- #2 RWA
  - Level: Banks
  - Comments: Banks’ total risk-weighted assets, in local currency
- #3 NPL ratio
  - Level: Banks
  - Comments: Banks’ nonfinancial corporate portfolio-related NPL ratio, at bank-level
- #4 Industry exposures
  - Level: Banks & industries
  - Comments: Banks’ current credit exposures to the various industries, gross (not net of provisions), performing and nonperforming combined
- #5 PD TTC
  - Level: Banks (optionally industries)
  - Comments: Banks’ recent historical average (through the cycle, TTC) default rates, pertaining to their nonfinancial corporate portfolio in total or by industry; for those banks with portfolios under IRB
- #6 LGD PiT
  - Level: Banks
  - Comments: Banks’ current PiT LGDs, pertaining to their nonfinancial corporate portfolio in total or by industry
- #7 LGD DT
  - Level: Banks
  - Comments: Banks’ current downturn (DT-) LGDs, pertaining to their nonfinancial corporate portfolio in total or by industry; for those banks with portfolios under IRB
- #8 NFC IRB shares
  - Level: Banks
  - Comments: Banks’ IRB-in-total portfolio shares; either for banks’ NFC portfolios in total or for each industry
- #9 STA RW performing
  - Level: Banks
  - Comments: Effective risk weights for banks (portfolios)’ performing portion under the standardized (STA) approach; either for NFC portfolios in total or for each industry
- #10 STA RW nonperforming
  - Level: Banks
  - Comments: Effective risk weights for banks (portfolios)’ nonperforming portion under the standardized (STA) approach; either for NFC portfolios in total or for each industry
- #11 Gross loan growth by industry, under the climate scenarios
  - Level: Industry & forward in time
  - Comments: Gross loan growth, by industry, along the scenario horizon; by industry; informed by output of CGE model component

*Source: IMF Working Paper — The ENV-FIBA Model for Climate Risk Analysis (Working Paper No. WP/2025/230).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025230-source-pdf.pdf_