## 1canea2025006-source-pdf ## Source details **Canonical URL:** [1canea2025006-source-pdf](https://www.imf.org/-/media/files/publications/cr/2025/english/1canea2025006-source-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/cr/2025/english/1canea2025006-source-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/cr/2025/english/1canea2025006-source-pdf.pdf.json) --- ### Overview and mission - FSAP mission to Canada: October 21-November 4, 2024, and February 10–24, 2025. - Canada: major fossil fuel producer, the 10th largest global emitter; committed to reduce emissions by 40–45 percent by 2030 relative to 2005. - Financial sector exposed to physical (notably floods and wildfires) and transition climate risks. ### Authorities’ climate risk framework and exercises - Authorities use state-of-the-art methodologies and granular public and private data. - OSFI and AMF, with BOC feedback, conducting first Standardized Climate Scenario Exercise (SCSE) covering most Canadian financial institutions; SCSE: - Covers transition risks and physical risk exposure assessment for floods and wildfires. - Expanded coverage from OSFI/BOC 2020 pilot. - Authorities’ top-down work includes: - Flood risk stress testing on residential lending portfolios. - Systemic implications of transition risk (direct and indirect effects). - FSAP recommendation: move away from private data vendors toward data sharing and collaboration with climate experts (e.g., Public Safety Canada, Natural Resources Canada, ECCC). ### FSAP coverage and focus - FSAP independent analyses covered six domestic systemically important banks (D-SIBs) and Québec’s cooperative credit institution (D‑SIFI) — “seven systemic DTIs”. - These seven systemic DTIs hold over 90 percent of overall DTI assets. - Physical analysis focus: wildfire risks to residential properties serving as mortgage collateral using a micro-approach. - Transition analysis focus: potential pressure points from the transition to a low-carbon economy, concentrating on the nonfinancial corporate (NFC) sector. ### Wildfire physical-risk methodology (simplified, exploratory) - Data and sample: - Confidential anonymized loan-level RESL dataset (cut-off September 2023) for seven systemic DTIs; final dataset total: 1,928 CAD Billions; spans all 10 provinces and 3 territories. - Geolocation available at census subdivision (CSD) level (OSFI) and forward sortation area (FSA) level (AMF) rather than address level. - Public fire weather data from ECCC (50 kilometers resolution) and NRCan Wildland-Urban Interface (WUI) data (30 meters, aggregated to 10 km). - Risk indicators and categorization: - Indicators: Fire Weather Index (FWI), Buildup Index (BUI), Fire Season Length (FSL). - FWI/BUI metrics: May–September ninety-fifth percentile; each indicator categorized into five buckets (low to extreme) using climatedata.ca thresholds. - FWI bucket thresholds preserved exactly as: FWI≤5 low; 530 extreme. - Selection of areas at risk: - Areas classified as meeting wildland fuels if containing at least one WUI grid (sensitivity uses ≥25 percent coverage). - Main analysis restricted to areas with extreme FWI (FWI>30); such areas located in British Columbia, Alberta, Saskatchewan and Manitoba. - Damage functions (conditional on a fire): - Main damage function: damage rates increase with FWI; FWI of 30 => damage rate 25 percent; FWI of 60 => 100 percent. - Sensitivities: binary damage function (complete destruction) and sigmoidal damage function (smoother non-linear increase). - Loss linkage and LGD definition: - Gross loss amount (GLA) linked to outstanding balances, accrued interest, and holding costs; mortgage insurance (MI) deducted when applicable. - LGD = max(GLA − NRA, 0) where NRA is post-fire adjusted recoverable property value less vacancy and other adjustments. - Key model parameters (preserved exactly): - Max α: 50 percent - β: 80 percent - House depreciation: 10 percent - Real estate commission: 5 percent - Appraisal fee: 500 CAD - Time to settlement: 18 months - Property tax: 2 percent - Maintenance rate: 1 percent - Property insurance cost: 3 percent - Legal cost: 5,000 CAD - Fixed maintenance cost: 2,000 CAD - Pass-through and insurance assumptions: - Pass-through of wildfire damages to DTIs (α) modeled as non-linear function of damage rate; maximum α in main analysis = 50 percent (sensitivity tests up to 75 percent). - Remaining damages (1−α) split between P&C insurance (β) and homeowners (1−β); β assumed equal to 80 percent. - Assumption: mortgage loan insurance excludes unrepaired damages from fires (aligned with CMHC practices). - Major methodological caveats: - Focuses on fire weather only (ignores changes in fuels, ignitions, adaptation). - Does not model probability of fire occurrence; FWI used as intensity proxy conditional on fire. - Exposure data static and aggregated across DTIs (may understate individual-bank concentration). - Estimates are conditional on defaults; probabilities of default not modeled. - Analysis likely underestimates physical risks and ignores indirect/systemic channels and compounding hazards. ### Wildfire exposure and aggregate results (selected exact figures) - Geographic and portfolio exposures: - Combined 6 D-SIB portfolio broadly diversified; one CSD in Toronto (land area 631 Km2) contains 11.6 percent of total loans across Canada. - Final dataset spans 3,324 CSDs and 997 FSAs for Québec’s D-SIFI. - Migration to very-high and extreme fire-weather: - Under both RCP 4.5 and RCP 8.5, by mid-century 65 percent of loans will be located in areas subject to very high and extreme fire weather, relative to 38 percent in the historical baseline. - More than 25 percent of loans will become exposed to “extreme” BUI under RCP 4.5 and RCP 8.5, relative to 9 percent in the historical baseline. - Fire season length (FSL): by mid-century, 90 percent of loans will be exposed to fire seasons longer than 200 days under both scenarios, relative to 70 percent under the historical baseline. - By 2100 under RCP 8.5, 57 percent of loans exposed to fire seasons longer than 250 days, relative to 21 percent under RCP 4.5. - Geographic expansion of identified at‑risk CSDs: - Number of CSDs identified expands from historical baseline 187 CSDs to RCP 8.5 end of century 616 CSDs. - Share of RESL loans in these areas: historical scenario 2.9 percent; RCP 8.5 end of century 22.6 percent. - High-risk loan amounts by indicator (Table values preserved exactly; percent notation as in source): - FWI >20: Historical baseline 45 77 77 77 79 - BUI > 60: 64 83 83 83 85 - FSL > 200: 86 97 97 98 98 - Total residential real estate damages conditional on fires in identified at-risk areas: - Baseline: 38 CAD billion. - RCP 4.5: damages increase by around 2.5 times relative to baseline. - RCP 8.5 (end of the century): damages increase by around 7.3 times, reaching 278 CAD billions. - Note: estimates are hypothetical values at risk and assume simultaneous impact across at-risk areas. - Aggregate LGD changes for affected loans in Western Canada relative to baseline: - RCP 4.5, mid-century: increase of 1.6 percentage points (PP). - RCP 8.5, end of the century: increase of 6.4 PP. - Loan-level vulnerabilities and insurance: - Median LTV of affected loans under RCP 8.5 at end of the century: 42 percent. - Only 24 percent of loans in Western Canada have mortgage insurance. - Loans with LTV above 80 percent experience significantly higher LGD increases but constitute only a small percentage of total loans in Western Canada. - Provincial heterogeneity example: - Saskatchewan: LGDs of loans located in Saskatchewan increase up to 17.5 PP under RCP 8.5 by the end of the century relative to the historical baseline. - Sensitivity findings: - Increasing pass-through to DTIs leads to substantially larger LGDs; convex pass-through lowers DTIs’ LGDs. - Binary (complete destruction) damage function yields significantly higher DTIs’ losses; sigmoidal function yields lower losses. - WUI mapping choice materially affects results: main mapping (≥1 WUI grid) identifies 97 percent of RESL loans as meeting wildland fuels; stricter mapping (≥25 percent WUI) reduces considered RESL to 23 percent. ### Flood risk stress testing (authorities’ work) — key figures - Public Safety Canada (PS) estimated average expected flood-related loss for residential real estate: CAD 2.97 billion per year. - PS estimated this loss to increase by 6.7 percent and 11.9 percent under RCP 4.5 and RCP 8.5 scenarios by the year 2100 relative to baseline. - Authorities’ flood damages had modest impacts on banks’ LGDs for residential real estate secured lending, with borrowers LTV >80 percent more sensitive to floods. ### Transition risk methodology and scenarios - Scenarios: Current policies (CP), Net Zero 2050 (NZ), Delayed Transition (DT); simulations up to 2040. - CP: policies implemented up to end-2023 assumed to stay in place until 2040. - Net Zero 2050: limits warming to 1.5 ̊C; emissions in 2030 achieve 45 percent reduction vs 2005; emissions reach zero in 2050. - Delayed Transition: emissions fixed until 2030, then steep reductions to reach Net Zero by 2050. - Model: IMF-ENV used to derive scenario-conditional macro and sectoral paths up to 2040; integrated micro-macro firm and bank simulation for NFCs and DTIs. - Scenario calibration and key policy parameters (as of mission end February 24, 2025): - National carbon price: started at CAD 65 per ton of GHG emissions in 2023 and intended to increase by CAD 15 per year to CAD 170 per ton CO2e in 2030. - Subsequent policy change noted: as of April 2025, Government of Canada removed the federal fuel charge (Quebec’s fuel carbon price remains). - Other policies: phase-out of coal generation by 2030; net-zero grid by 2050; Clean Fuel Regulations (2023); methane reduction target of at least 75 percent relative to 2012 by 2030; ZEV availability standard for new light-duty vehicles to be 100 percent by 2035 (interim 20 percent by 2026, 60 percent by 2030); “2 billion Trees” program invests CAD 3.2 billion over 10 years by 2031. - Firm and bank data: - Firm micro-sample T0 constructed from Orbis and CreditEdge; final T0 dataset: 2,245 firms (after filters). - Bank module uses confidential DTIs’ NFC loan portfolios as of Q3: 2024; dataset covers NFC loans > CAD 10 million (estimated 60 percent coverage of DTIs’ NFC loans). - Sectoral aggregation: 16 sectors used for micro-simulation. - PD and LGD modelling: - PD logit model coefficients (preserved exactly): - LEV: 3.280*** (.0770) - ICR: -.0028*** (.0003) - EBITR: -.5354*** (.0362) - CDR: -.0144*** (.0022) - cons: 5.288*** (.0266) - Observations: 17,060; Within R2: 0.2079. - LGDs derived using Frye and Jacobs (2012) formulation with asset correlation parameter ρ and TTC/PiT mappings. - Simulation mechanics: - Firms’ sales follow sectoral output; operating expenses follow sectoral intermediate input growth; emissions costs subtracted from EBIT. - Negative cash triggers additional borrowing; leverage increases PDs in a reinforcing cycle. - NPL evolution assumes write-off rate WRO = 100 percent and cure rate CURER = 0 percent. - Results compute annual expected losses and do not follow IFRS 9 methodology. ### Transition risk results (selected exact figures and dynamics) - Sectoral impacts: - Two heaviest emitters: oil & gas and electricity non-renewable (largest emission intensities). - Non-renewable electricity and oil & gas show largest relative PD increases under transition scenarios. - Timing and scenario dynamics: - Net Zero 2050 produces larger adverse impacts earlier (by 2030) relative to baseline; DT not impactful by 2030. - After 2030, DT induces sharper impacts; by 2040 DT impacts exceed Net Zero for most sectors. - DTIs’ credit losses: - DTIs’ credit losses are estimated to increase between 3.8–7.1 percent under the Net Zero 2050 and DT scenarios relative to CP by 2040 (bank- and scenario-dependent). - Cumulative losses over the simulation period are greater under Net Zero 2050, driven by stronger early impacts; DT adverse effects intensify after 2040 (simulation end date limits full DT impact capture). - Credit losses mainly driven by exposures in “other business services”, transport, and construction; oil & gas and electricity non-renewable show largest relative increases by sector. ### Limitations, sensitivities, and major caveats (summary) - Wildfire analysis: - Exploratory with large uncertainties: lacks address-level geolocation, Canada-wide forward-looking hazard likelihood/intensity data, and a Canada-specific damage function. - Focuses on one hazard (wildfire) and one transmission channel (credit losses via collateral); ignores compounding events, indirect/systemic channels, and probabilities of occurrence. - Combined portfolio approach may understate individual-bank concentration risks. - Transition analysis: - Covers only NFC loans > CAD 10 million; excludes smaller firms that may be more vulnerable. - Simulation ends in 2040, limiting assessment of long-term Delayed Transition impacts. - Results reflect only direct credit-risk channel; other channels (market, liquidity, macro feedbacks beyond model scope) not fully captured. - Both analyses likely understate overall climate-linked risks to the Canadian banking and deposit-taking sector. ### Key FSAP policy recommendations (summary, timing legend preserved as in source) - Strengthen data sharing and collaboration with natural hazard and climate experts. — OSFI, BOC, AMF jointly with relevant climate agencies — MT - Establish standardized public climate data and risk frameworks across provinces (including damage functions). — FA, PA — MT - Expand physical risk assessment (geographical areas, hazards damages, scenarios, return periods). — OSFI, BOC, AMF — MT - Remove barriers for exchange of confidential information between federal and provincial authorities (PA). — OSFI, AMF, FA, PA — ST - Enhance data standardization and cooperation between provincial and FA. — OSFI, AMF, FA, PA — MT - Strengthen the supervision of the adequacy of property insurance protection on the collateral held by banks and insurance companies. — OSFI, AMF — MT - Move away from private data vendors towards public or internal data (when possible). — OSFI, AMF — MT - Timing legend: I — Immediate (within 1 year), ST—Short term (within 1–2 years), MT—Medium term (within 3–5 years). *Source: International Monetary Fund — Technical Note prepared by Caterina Lepore, Tumer Kapan, and Hugo Rojas-Romagosa, based on the IMF FSAP mission to Canada (October 21-November 4, 2024, and February 10–24, 2025).* ### EXECUTIVE SUMMARY __________________________________________________________________________ 6 ### EXECUTIVE SUMMARY ### Overview - The financial sector in Canada is exposed to both physical and transition climate risks. - Prominent physical hazards: floods and wildfires. Climate change could increase their intensity. - Canada is a major fossil fuel producer, the 10th largest global emitter, and has committed to reduce emissions by 40–45 percent by 2030 relative to 2005. - The FSAP mission to Canada occurred during October 21-November 4, 2024, and February 10–24, 2025. ### Authorities’ climate risk framework and exercises - Authorities’ climate risk analyses adopt state-of-the-art methodologies and leverage granular data from both public and private sources. - OSFI and AMF, with feedback from BOC, are conducting their first Standardized Climate Scenario Exercise (SCSE), a comprehensive climate risk stress test covering most Canadian financial institutions. The SCSE: - Covers both transition risks and a physical risk exposure assessment for floods and wildfires. - Significantly expanded coverage from the pilot analysis conducted by OSFI and BOC in 2020. - Authorities have also conducted top-down analyses, including: - A flood risk stress testing on residential lending portfolios. - An analysis of systemic implications of transition risk including direct and indirect effects on the Canadian financial system. - FSAP recommendation: move away from private data vendors towards data sharing and additional collaboration with climate experts in Canada (e.g., Public Safety Canada (PS), Natural Resources Canada (NRCan), and ECCC). ### FSAP independent analyses (scope and focus) - FSAP performed independent analyses of key physical and transition risks for large deposit takers. - Coverage: six domestic systemically important banks (D-SIBs) and Québec’s cooperative credit institution designated as a domestic important financial institution (D-SIFI) — collectively referred to as the “seven systemic DTIs”. - These seven systemic DTIs hold over 90 percent of overall DTI assets. - Physical risk analysis focus: wildfire risks to residential properties serving as collateral for mortgages using a micro-approach. - Transition risk analysis focus: potential pressure points from the transition to a low-carbon economy, focusing on the nonfinancial corporate (NFC) sector. ### Physical (wildfire) risk key findings - Residential mortgage portfolio is geographically diversified across Canada but exposed to notable wildfire risks. - A significant portion of loans is located in areas with very high or extreme fire-weather; this portion is expected to increase with climate change under both medium and high-emissions scenarios already by mid-century. - An innovative wildfire risk analysis estimated potential losses for the seven systemic DTIs conditional on borrower default for residential real estate lending in areas with extreme fire-weather and where wildland fuels are present. - Analysis accounted for the role of both mortgage and property and casualty (P&C) insurance. - Loss Given Defaults (LGDs) increase significantly more under the high emission scenario than the medium emission scenario relative to historical climate. - LGDs vary substantially by province; loans with high Loan-to-Value (LTV) are more sensitive to wildfire-driven losses, although these represent a small portion of mortgages. - Results are sensitive to: - Pass-through rate of wildfire damages to DTIs: increasing pass-through leads to substantially larger LGDs. - Damage function: assuming complete destruction of houses experiencing a fire yields significantly higher DTI losses. ### Transition risk key findings - Transition risk results are heterogeneous across sectors, with notable impacts for heavy emitters. - NFCs’ credit risk would generally increase modestly but with significant variation across sectors. - Notable impacts observed for the oil and gas sector. - The electricity non-renewable sector experiences the largest relative Probability of Default (PD) increase under the two transition scenarios relative to baseline in 2040. - Under a Delayed Transition (DT), earlier-year costs are lower, but from 2040 on losses exceed those of the Net Zero 2050 scenario. - The adverse impact is larger under Net Zero 2050 than under DT for most of the simulation period; however, by 2040 DTIs’ credit losses under DT begin to exceed the ones under Net Zero. - Credit losses are estimated to increase between 3.8–7.1 percent under the Net Zero 2050 and DT scenarios relative to current policies (CP) by 2040. - Cumulative losses are greater under Net Zero 2050 over the simulation period ending in 2040; the simulation end date is a factor since DT adverse effects intensify after 2040. - Results calculate annual expected losses and do not follow the IFRS 9 methodology. ### Sensitivities, limitations, and caveats - Wildfire risk analysis is exploratory and subject to large uncertainties and data limitations: - May underestimate concentration of individual banks’ exposures, which are not identifiable in FSAP dataset. - Relies on a simplified methodology due to lack of Canada-wide wildfire hazard data and damage function. - Only covers one hazard (wildfire) and does not account for compounding effects, likely underestimating physical risks. - Transition risk assessment limitations: - Covers only NFC loans in excess of 10 Canadian dollars (CAD) million, excluding smaller firms that may be more vulnerable. - Simulation ends in 2040, limiting assessment of longer-term DT impacts. - Both analyses focus on a specific credit risk channel and capture only direct impacts, ignoring indirect and systemic effects, hence potentially underestimating overall climate risks to the Canadian banking and deposit-taking sector. ### Key policy recommendations (summary from Table 1) - Strengthen data sharing and collaboration with natural hazard and climate experts. — OSFI, BOC, AMF jointly with relevant climate agencies — MT - Establish standardized public climate data and risk frameworks across provinces (including damage functions). — FA, PA — MT - Expand physical risk assessment (geographical areas, hazards damages, scenarios, return periods). — OSFI, BOC, AMF — MT - Remove barriers for exchange of confidential information between federal and provincial authorities (PA). — OSFI, AMF, FA, PA — ST - Enhance data standardization and cooperation between provincial and FA. — OSFI, AMF, FA, PA — MT - Strengthen the supervision of the adequacy of property insurance protection on the collateral held by banks and insurance companies. — OSFI, AMF — MT - Move away from private data vendors towards public or internal data (when possible). — OSFI, AMF — MT - Timing legend: I — Immediate (within 1 year), ST—Short term (within 1–2 years), MT—Medium term (within 3–5 years). *Source: International Monetary Fund — Technical Note prepared by Caterina Lepore, Tumer Kapan, and Hugo Rojas-Romagosa, based on the IMF FSAP mission to Canada (October 21-November 4, 2024, and February 10–24, 2025).* ### 5. The authorities have conducted detailed analysis on floods risk. The OSFI and Bank of ### 5. The authorities have conducted detailed analysis on floods risk. The OSFI and Bank of ### Flood risk stress testing conducted by OSFI and Bank of Canada - Top-down flood risk stress testing on banks’ residential lending portfolios performed (Johnston et al. 2023). - Public Safety Canada (PS) provided damages estimations using flood hazard data from private data vendors. - PS estimated the average expected flood-related loss for residential real estate to be CAD 2.97 billion per year. - PS estimated this loss to increase by 6.7 percent and 11.9 percent under RCP 4.5 and RCP 8.5 scenarios by the year 2100 relative to the baseline. - BOC mapped PS damages to granular data on banks’ residential loans using postal code information along with information on mortgage and P&C insurance. ### Findings from the authorities’ flood analysis and stress test - Authorities examined both annual average floods (baseline) and 1-in-100-year flood event (severe) by the year 2100 under RCP 4.5 and RCP 8.5. - Damages from these floods had modest impacts on financial institutions’ LGDs for their residential real estate secured lending (RESL) portfolio. - Borrowers with LTV ratio above 80 percent were found to be more sensitive to floods. - The modest impacts were attributed to homeowner equity and recent rise in house prices in Canada. - The study included significant limitations that understate risks, including limited access to granular location data and the use of a simplified LGD formula. ### Systemic Climate Scenario Exercise (SCSE) - OSFI and AMF are currently running their first SCSE, covering floods and wildfires. - SCSE combines top-down elements (shared scenarios and risk factors) with bottom-up approaches requiring institutions to use their own exposure data, classification methods, and non-climate-related modeling frameworks. - Objectives: raise awareness, build capacity, and execute a standardized quantitative assessment across participating institutions. - SCSE is a comprehensive climate risk stress test covering most Canadian financial institutions. - SCSE includes an exposure assessment for both floods and wildfire risks. - Floods data are sourced from a private data vendor due to lack of public nationwide forward-looking flood data in Canada. - Wildfires data are publicly available from data published and maintained by ECCC (https://climatedata.ca/fire-weather/). - Participating institutions are asked to map hazards’ data to their geocoded exposures (mortgages, home equity line of credit (HELOCs) non-mortgage loans and reverse mortgages secured by residential property and other) for selected geographical regions (urban areas for floods and rural areas for wildfires). ### Limitations of current physical risk work and data gaps - Flood risk stress testing was a pilot micro-approach leveraging loan level data and flood maps from PS but constrained by privacy considerations limiting address-level use. - SCSE helped build capacity (e.g., geocoding exposures) but was not a full physical risk assessment (it did not estimate damages from wildfires). - Current physical risk module of SCSE is restricted in scope: includes only an exposures assessment and is confined to selected geographical regions, potentially overlooking important risk areas, particularly for wildfires. - Recommendation to move away from private data vendors towards data sharing and collaboration with climate experts and authorities in Canada when possible, especially as more data becomes available in the medium term. - Challenges remain due to data gaps and lack of a common climate risk assessment framework in Canada. - Public Safety is leading development of an open-source flood hazard data, expected to be complete in a few years. ### Inter-agency collaboration and data standardization issues - Collaboration between provincial authorities and federal authorities (FA) could be enhanced. - SCSE is a good example of collaboration between OSFI and AMF, but impediments remain for sharing information and data standardization. - RESL dataset collects loan level data on residential real estate secured lending with location at different resolutions: AMF provides forward sortation area (FSA) level; OSFI provides postcode level. ### Decision to focus FSAP wildfire analysis on DTIs - Given extensive work already conducted on flood risk by authorities, the FSAP team and Canadian Authorities decided FSAP should focus on wildfire risks. - The FSAP physical risk analysis investigates banks’ financial stability risks from wildfires under current and future climate conditions. - As of the time of the FSAP, this was among the first studies to perform a wildfire risk analysis for the DTI sector. - The analysis is exploratory and focuses on methodology development and enhancing awareness of climate risks, specifically potential risks from wildfires’ impacts on residential properties serving as collateral for DTIs’ mortgages. ### Institutional responsibilities and ongoing federal initiatives on wildfires - Responsibility for wildland fire management rests primarily with the provinces and territories; the federal government supports provinces and territories with science, technology, emergency management (Public Safety Canada), and managing fires on federal lands (Parks Canada, Department of National Defense). - NRCan is the federal government’s primary source of wildland fire technical and scientific expertise and maintains the Canadian Wildland Fire Information System (CWFIS). - The Canadian Forest Service is working on a national wildfire risk assessment in partnership with ECCC, Public Safety Canada and the National Research Council of Canada to quantify wildfire risk; this project is under development and aims to perform a ‘first of its kind’ wildfire impact assessment and probabilistic hazard and risk assessments. ### Nationwide hazard modeling and indicators - There is currently no comprehensive publicly available nationwide hazard modeling in Canada due to size, landscape diversity, and lack of standardization and unique definitions for fire hazard indicators. - A wealth of fire risk indicators and historical fire data are publicly available, but no public data specifically on forward-looking fire likelihood and intensity covering the entire country. - The Canadian Council of Forest Ministers’ Prevention and Mitigation Strategy is committed to developing and making accessible a wildland fire risk assessment framework and risk planning resources; the risk assessment framework is anticipated to be available to the public in the winter of 2025–26. - One commonly used fire indicator is the FWI (Fire Weather Index), a numeric rating of fire intensity composed of six components (three fuel moisture codes and three fire behavior indices) calculated from daily observations of temperature, relative humidity, wind speed, and 24-hour precipitation. ### ECCC projections and data resolution - ECCC has published forward-looking projections for FWI under RCP 2.6, RCP 4.5, and RCP 8.5. - The ECCC fire weather data has a resolution of 50 kilometers (km) on a national scale. - This fire weather information can be combined with other information on fuel and ignition to assess wildfire intensity changes. ### FSAP methodological choices and limitations for wildfire analysis - FSAP adopted a simplified methodology due to data limitations and uncertainties; Annex I discusses an ideal first-best methodology. - Analysis focuses on direct impacts from wildfire; indirect impacts and cascading effects of multiple natural hazards are out of scope. - The simplified methodology focuses on climate change impacts on fire weather and abstracts from other drivers of fire behavior (e.g., changes in fuels and ignitions, socioeconomic changes, exposures and vulnerability). - Analysis relies on DTIs’ data lacking highly granular geolocational information, which is kept static throughout the exercise. - The analysis likely underestimates physical risks because: - It focuses on the combined mortgage portfolio of all seven systemic DTIs, so individual DTIs’ concentrations are not identified. - It focuses on one transmission channel of losses from wildfires to DTIs; several other hazards and channels are not included. - Large uncertainties exist due to lack of forward-looking data on fire likelihood and intensity and absence of a Canada-specific damage function. - The analysis is an exploratory approach given limited available data. ### Data sources, sample, and processing - Confidential anonymized loan-level dataset obtained for seven systemic DTIs’ mortgages collateralized by residential properties: six domestic systemically important banks (D-SIBs) and Québec’s cooperative credit institution designated as a domestic important financial institution (D-SIFI). - RESL data provided by OSFI for the 6 D-SIBs without identifiers; AMF provided RESL data for Québec’s D-SIFI separately. - RESL data cut-off date: September 2023. - Ideal location information would be at address level; available information is at the census subdivision (CSD) level from OSFI and FSA level from AMF. - Dataset lacks information on building structure or building codes; analysis focused on geographical areas rather than individual buildings. - Processing steps: aggregated loans (including HELOCs) on the same property; property values updated to September 2023 using Teranet-National Bank House Price index; filtered out loans with missing appraisal value, missing outstanding balance, non-positive outstanding balance, missing credit limit and zero outstanding balance. - Final dataset for the 6 D-SIBs spans 3,324 CSDs and 997 FSAs for Québec’s D-SIFI for a total of 1,928 CAD Billions. - Data spans all 10 provinces and 3 territories; largest loan amount observed in Ontario, followed by Québec and British Columbia. ### Use of fire weather and Wildland-Urban Interface (WUI) data - Publicly available fire weather data used, focusing on time periods: 2041–2070 (mid-century) and 2071–2100 (end of the century), and scenarios RCP 4.5 and 8.5, plus historical baseline (1971–2000). - NRCan provided Wildland-Urban Interface (WUI) data (Johnston and Flannigan (2018)) updated in 2019; WUI resolution is 30 meters nationally and aggregated to 10 km for tractability. - WUI represents areas where fuels are close to human-built urban structures (e.g., homes, public buildings). - Generally, wildfires occur more often and are more severe when there is: i) hot, dry, and windy weather, ii) flammable vegetation, and iii) more ignition sources. The analysis accounts for the first two factors and ignores ignition sources. - Due to lack of data on vulnerability, a reduced form approach was adopted. ### Matching exposures to wildfire risk data and risk indicators - DTIs’ exposures were contrasted with wildfire weather data using nearest neighbor matching for smaller CSDs and FSAs not fully containing a 50km x 50km grid, and averaging wildfire risk data of grids contained in larger CSDs and FSAs. - Risk indicators: FWI, Buildup Index (BUI) and Fire Season Length (FSL) under historical and future climate scenarios (RCP 4.5 and RCP 8.5). - For FWI and BUI, metric used: May to September ninety-fifth percentile (indicator of severity of high fire weather conditions). - Each indicator categorized into five buckets (low to extreme) based on climatedata.ca thresholds: - FWI: FWI≤5 low; 530 extreme. - BUI: BUI≤20 low; 20< BUI ≤40 moderate; 40< BUI ≤60 high; 60< BUI ≤90 very high; BUI >90 extreme. - FSL: FSL≤100 low; 100300 extreme. - Banks’ mortgages were categorized into five buckets based on quantiles of their overall distribution; combined with five fire-risk buckets to create 25 buckets showing concentration of mortgages by loan amount and fire risk indicator. - This categorization assessed whether mortgages are concentrated in areas with weather more conducive to fires and how concentration may change under future climate. ### Damage estimation and linkage to DTIs’ losses - FSAP estimated potential damages from wildfires in selected at-risk areas and potential losses to DTIs conditional on defaults. - Following Adrian et al. (2022), a micro approach was applied: estimate wildfire damages in at-risk areas and link them to DTIs’ loan-level exposures. *Source: 1canea2025006-source-pdf - 5. The authorities have conducted detailed analysis on floods risk. The OSFI and Bank of* ### 24. The analysis selected geographical areas more likely to experience a fire. As data on fire ### 1canea2025006-source-pdf - 24. The analysis selected geographical areas more likely to experience a fire. As data on fire ### Methodology for selecting areas at fire risk - Geographical areas were selected where residential buildings collateralizing banks’ mortgages meet with, or are intersperse within, wildland fuels using Wildland-Urban Interface (WUI) data. - Each CSD and FSA was categorized as meeting wildland fuels if it contained at least one WUI grid. - Sensitivity analysis: definition changed to CSD and FSA covered by at least 25 percent of WUI. - Analysis restricted to areas with extreme Fire Weather Index (FWI), i.e. higher than 30. All such areas were located in Western Canada (British Columbia, Alberta, Saskatchewan and Manitoba). - The team adopted an FWI threshold of 30 to be conservative, noting other studies suggest spread-event thresholds around FWI of 19 but the team used a higher threshold. ### Damage functions and damage estimation (conditional on a fire) - Damages are conditional on a fire occurring in the identified areas and assume all structures within these areas would encounter a fire and be (at least partially) damaged. - Degree of impacts linked solely to fire intensity proxied by FWI. - Main analysis damage function: damage rates increase with FWI; FWI of 30 implies a damage rate of 25 percent and FWI of 60 implies 100 percent. - Sensitivity analyses: - Binary damage function: assumes all real estate subject to FWI above 30 would completely be destroyed. - Sigmoidal damage function: assumes a smoother but non-linear increase of damage rates. ### Insurance, borrower losses, and pass-through to Deposit-Taking Institutions (DTIs) - Wildfire damage allocation: - Property & Casualty (P&C) insurance covers a portion of house damages. - Mortgage loan insurance (MI) covers shortfalls of the outstanding loan amount and interest after foreclosure, accounting for the home value excluding damages from fire. - Assumption: mortgage loan insurance excludes unrepaired damages from fires (aligned with CMHC practices). - RESL data includes information on whether the loan is mortgage insured; no data on P&C insurance was available. - FSAP recommendation: authorities should strengthen supervision of the adequacy of property insurance protection on collateral held by banks and insurance companies. - Fire insurance limitations: - Fire insurance does not directly cover mortgage payments or outstanding mortgage amounts after a loss. - Insurance settlements can take extended periods when many people are affected simultaneously. - Additional borrower distress channels: adverse local economic impacts, transient/localized house price shocks, non-recourse mortgages in Alberta and Saskatchewan, and insurer market withdrawal or premium increases. ### Pass-through of wildfire damages to DTIs (α) and P&C insurance share (β) - Portion (α) of wildfire damages passed on to DTIs modeled as a non-linear function of the wildfire damage rate: - Small damages largely absorbed by homeowners and P&C insurance. - For larger damages, higher pass-through to DTIs. - In the case of complete destruction of the residential real estate, 50 percent of the loss is passed on to DTIs in the main analysis. - Sensitivity analyses on pass-through: - Two variants increasing maximum pass-through rate to 75 percent: one with same functional shape and one assuming a convex shape. - Remaining portion of damages, (1-α), split between P&C insurance (β) and homeowners (1-β). - Given lack of P&C insurance data, β assumed equal to 80 percent. ### Definition of LGD and recoverable amount - DTIs’ LGDs defined as gross loss amount (GLA) in excess of net recoverable amount (NRA): - LGD = max(GLA − NRA, 0). - Net recoverable amount (NRA) defined as the maximum of a post-fire adjusted property value minus post-fire vacancy (expressed in the source text as a formal formula) and 0. - Gross loss amount depends on mortgage insurance and includes outstanding balances, accrued interest until settlement, and costs of holding the defaulted property if these exceed the recoverable sale amount excluding damages from wildfire. - If loan covered by mortgage insurance, GLA expression includes MI subtraction; otherwise GLA is computed without MI. ### Key wildfire risk model parameters (Table 2) - Max α: 50 percent - β: 80 percent - House depreciation: 10 percent - Real estate commission: 5 percent - Appraisal fee: 500 CAD - Time to settlement: 18 months - Property tax: 2 percent - Maintenance rate: 1 percent - Property insurance cost: 3 percent - Legal cost: 5,000 CAD - Fixed maintenance cost: 2,000 CAD ### Caveats and limitations of the analysis - Focuses only on one driver of wildfires: fire weather (FWI). Does not account for changes in fuels or ignitions. - Does not account for likelihood/probability of fires; adopts FWI as a proxy for intensity conditional on fire. - Geospatial granularity limited: mapping of WUI to CSDs subject to uncertainties; WUI dataset latest update uses 2019 structures. - Exposure data is static (does not change over time). - Damages computed under reduced-form approach; higher FWI is assumed to imply higher damages, though in reality many other factors influence actual damages (region, fuels, fire regimes, land management, mitigation/suppression capabilities, past fire experience). - Analysis estimates losses conditional on defaults and does not model probabilities of default. - Combined portfolio used for exposure analysis may understate individual-bank concentration risks due to confidentiality constraints. - Analysis captures direct credit-risk channel only; ignores indirect and systemic effects and other hazards or compounding hazard interactions (e.g., wildfire-drought cycles). - Overall, analysis likely underestimates potential losses from physical risks. ### Results — exposure analysis (aggregate and distributional findings) - Portfolio diversification: - Combined portfolio of the 6 D-SIBs is broadly diversified; one CSD in Toronto (land area 631 Km2) contains 11.6 percent of total loans across Canada. - Other top ten concentrated CSDs: Calgary, Brampton, Ottawa, Mississauga, Vancouver, Montreal, Surrey, Edmonton and Hamilton. - High-risk loan amounts by indicator (Table 3, in percent; values preserved exactly): - FWI >20: Historical baseline 45 77 77 77 79 - BUI > 60: 64 83 83 83 85 - FSL > 200: 86 97 97 98 98 - (Sources: OSFI RESL data and CanLEAD— FWI dataset from ECCC, IMF Staff computations.) - Migration of loans to very-high and extreme fire-weather: - Under both RCP 4.5 and RCP 8.5, by mid-century 65 percent of loans will be located in areas subject to very high and extreme fire weather, relative to 38 percent in the historical baseline. - Majority of loans already exposed to “very high” BUI. - More than 25 percent of loans will become exposed to “extreme” BUI under RCP 4.5 and RCP 8.5, relative to 9 percent in the historical baseline. - Fire season length (FSL): by mid-century, 90 percent of loans will be exposed to fire seasons longer than 200 days under both scenarios, relative to 70 percent under the historical baseline. - By 2100 under RCP 8.5, 57 percent of loans exposed to fire seasons longer than 250 days, relative to 21 percent under RCP 4.5. - Geographic concentration of areas "more likely to experience a fire": - Identified provinces: British Columbia, Alberta, Saskatchewan and Manitoba. - Number of CSDs identified expands from historical baseline (187 CSDs) to RCP 8.5 end of century (616 CSDs). - Share of RESL loans in these areas: historical scenario 2.9 percent; RCP 8.5 end of century 22.6 percent. ### Results — risk analysis (summary) - The analysis links wildfire damages to DTIs’ credit losses via the pass-through function α and the assumed damage functions tied to FWI. - Empirical evidence cited: 2016 Fort McMurray wildfire associated with a doubling of mortgage arrears in severely damaged areas, dissipating between 12 and 18 months after the event; effect did not depend on existence of mortgage insurance. - Main analysis assumes maximum pass-through to DTIs of 50 percent in case of complete destruction; sensitivity tests include 75 percent maximum pass-through and convex pass-through shapes. - Given assumptions and caveats, the FSAP notes the analysis is a useful starting point but likely underestimates full potential losses. *Source: IMF Staff (content unit 1canea2025006-source-pdf - excerpt).* ### 38. Conditional on fires erupting in these areas, total damages to residential real estate ### 1canea2025006-source-pdf - 38. Conditional on fires erupting in these areas, total damages to residential real estate ### Wildfire physical-risk findings and estimates - Total damages to residential real estate, conditional on fires erupting in identified at‑risk areas: - Baseline: 38 CAD billion. - RCP 4.5: damages increase by around 2.5 times relative to baseline. - RCP 8.5 (end of the century): damages increase by around 7.3 times, reaching 278 CAD billions. - Caveats on interpretation: - Estimates are hypothetical values at risk, not total expected losses. - Assumes all at‑risk areas are affected simultaneously, which may over-estimate losses. - Comparison note: Wang et al. (2020) projects increases in burn area in Canada up to 4.3 times relative to the baseline for the same period and scenario. ### Loan- and bank-sector impacts (DTIs and LGDs) - Aggregate LGD (loss given default) increases for affected loans in Western Canada relative to baseline: - RCP 4.5, mid-century: increase of 1.6 percentage points (PP). - RCP 8.5, end of the century: increase of 6.4 PP. - Loan-level vulnerabilities: - Median LTV of affected loans under RCP 8.5 at end of the century: 42 percent. - Loans with LTV above 80 percent experience significantly higher increases in LGDs than lower‑LTV loans. - Loans with LTV above 80 percent constitute only a small percentage of total loans in Western Canada. - Provincial heterogeneity: - Saskatchewan: LGDs of loans located in Saskatchewan increase up to 17.5 PP under RCP 8.5 by the end of the century relative to the historical baseline, reflecting higher fire weather exposure. ### Insurance-sector exposures and coverage assumptions - Property & Casualty (P&C) insurance: - FSAP assumption: P&C insurance covers 80 percent of the damages not passed on to DTIs. - If P&C insurance absorbs a smaller portion of wildfire damages, more losses will be passed on to DTIs. - Given this assumed coverage, exposures for the P&C insurance sector are large. - Mortgage insurance: - Only 24 percent of loans in Western Canada have mortgage insurance. - Mortgage insurance covers only shortfalls of the outstanding loan amount and interest after foreclosure, accounting for home value excluding fire damages; it pays out only if the loan is underwater excluding fire damages. - Because LTVs are generally low, these shortfalls are rare and small overall. ### Sensitivity analyses and key modelling dependencies - Pass-through rate sensitivity: - Increasing the pass-through rate to DTIs leads to substantially larger LGDs. - A more convex pass-through function (high pass-through only for higher damage rates) lowers DTIs’ LGDs. - Inverse relationship: if DTIs’ losses increase (decrease), P&C insurance exposures decrease (increase). - Implication: insurer market exit or premium increases could materially change loss allocation between insurers and DTIs. - Damage-function sensitivity: - Assuming complete destruction of houses (binary damage function) yields significantly higher DTIs’ losses. - Assuming damages increase significantly only for extremely high FWI (sigmoidal function) yields lower DTIs’ losses. - Results are highly sensitive to the chosen damage function. - Wildland-Urban Interface (WUI) mapping sensitivity: - Main analysis definition (CSD and FSA categorized as meeting wildland fuels if containing at least one WUI grid) identifies 97 percent of RESL loans as meeting wildland fuels. - Stricter definition (CSD and FSA covered by at least 25 percent of WUI) reduces considered RESL to 23 percent of total. - Under the stricter mapping, banks’ LGDs and overall wildfire damages are smaller. - Highlight: granular geospatial data materially affects results; RESL geographical data available to FSAP was relatively aggregated, limiting mapping precision. ### Transition risks, policies, and mitigation measures - Canada’s commitments and targets: - Net-zero emissions by 2050. - Emissions reduction targets: 40–45 percent by 2030 and 45–50 percent by 2035 relative to 2005 levels. - Carbon pricing policy: - National price on carbon pollution for explicit price-based systems started at CAD 65 per ton of GHG emissions in 2023 and was intended to increase by CAD 15 per year to CAD 170 per ton CO2e in 2030. - Note: policies summarized are up to date as of the end of the main IMF mission on February 24th, 2025. - Subsequent policy change: as of April 2025, the Government of Canada removed the federal fuel charge. The Quebec province own fuel carbon price is still implemented. - Other mitigation policies and targets: - Phase-out of coal generation by 2030 and requirement of a net-zero grid by 2050. - Clean Fuel Regulations (2023) requiring reductions in carbon intensity of gasoline and diesel supplies over time. - Commitment to reduce methane emissions from the oil and gas sector by at least 75 percent relative to 2012 levels by 2030. - Electric Vehicle Availability Standard: require 100 percent of new light-duty vehicles offered for sale in Canada to be ZEVs by 2035, with interim sales targets of 20 percent by 2026 and 60 percent by 2030. - “2 billion Trees” program: invests CAD 3.2 billion over 10 years by 2031 to support tree planting to help achieve net-zero by 2050. ### Authorities’ transition-risk analysis and FSAP recommendations - Authorities’ exercises: - OSFI and BOC pilot transition risk analysis (started 2020) covered six federally regulated financial institutions and focused for banks on credit risk for wholesale loans in the 10 most emission-intensive sectors. - BOC extended analysis to cover credit, market, and liquidity risks for a broader set of financial institutions. - SCSE expanded coverage to most Canadian financial institutions, analyzing credit and market risk of global commercial exposures and including an exposure assessment of Canadian real estate-related assets. - Methodology and data recommendations: - SCSE used macro-financial factors for three transition scenarios and 25 industry sectors developed by OSFI and AMF using BOC scenarios and Moody’s Analytics data; participating institutions used these factors to estimate credit and market risk. - FSAP recommendation: move away from prescriptive, standardized exercises over-relying on private data vendors; instead, allow more use of financial institutions’ internal data and methodologies and develop internal datasets (for example, to estimate PDs and LGDs under transition scenarios) for top-down studies and analyses. *Source: IMF Staff computation and related analysis in the referenced chapter.* ### Introduction ### Introduction ### Transition risk analysis overview - The analysis adopted an integrated micro-macro approach following Gross et al. (forthcoming) and its application in the 2024 Japan FSAP. - The IMF-ENV model was used to derive scenario-conditional paths for macroeconomic and sectoral variables up to 2040 and a micro simulation connected those variables to balance sheets of NFCs and banks (Figure 21). - The focus of the analysis was to estimate the potential impact of the transition for the 7 systemic DTIs in terms of credit risk from NFCs. ### Scenarios considered - Three scenarios were considered: CP (the baseline scenario), orderly transition (“Net Zero 2050”) and disorderly transition (“Delayed Transition”). - The transition scenarios align with those of the Network for Greening the Financial System (NGFS) Phase IV. - Scenario descriptions: - Current policies (CP): only currently implemented policies up to end-2023 are taken into account and are assumed stay in place until the end of the simulation period (2040). The scenario simulations do not extend beyond 2040 due to modelling constraints. - Net Zero 2050 (NZ): limits global warming to 1.5 ̊C through stringent climate policies and innovation, reaching global net zero CO2 emissions around 2050. - Delayed transition (DT): assumes annual emissions do not decrease until 2030. Strong policies are needed from 2031 onwards to reach Net Zero in 2050 on an accelerated timeline. - In the Net Zero 2050 scenario, emissions in 2030 achieve a 45 percent reduction with respect to the historical 2005 emissions, in line with Canada’s NDC target for 2030. From 2031 onwards emissions decrease to reach zero in 2050. - In the DT scenario CP emissions are kept fixed until 2030, afterwards emissions decline to reach the same 2040 target emission level of the Net Zero 2050 scenario. - In all scenarios, it is assumed that the revenue generated by the carbon price is transferred back entirely to households. ### IMF-ENV model calibration and features - Scenarios were calibrated using the IMF’s global dynamic computable general equilibrium (CGE) model called IMF-ENV, based on ENVISAGE (Vander Mensburgghe, 2024) and the OECD’s ENV-Linkages model (Chateau et al. 2014). - IMF-ENV features a detailed description of energy production and consumptions, and links between economic dynamics and GHG emissions (Chateau et al. 2025). - IMF-ENV outputs include both country level and sectoral variables and explicitly incorporates 36 different sectors; for the FSAP analysis these sectors were mapped to the sectoral classification specified in Figure 23. ### Macro impacts and carbon pricing - Macro impacts by 2040 are somewhat greater under the DT scenario compared to the Net Zero 2050 scenario. - Under the DT scenario, carbon prices would need to accelerate sharply and surpass those in the Net Zero scenario as we approach mid-century. - The scenario figures reported include: Carbon price (CAD), Total GHG emissions (Metric Tons of Carbon Dioxide Equivalent), GDP losses (Percent level deviation from CP in 2030 and 2040), and Unemployment (Percent level deviation from CP in 2030 and 2040). ### Sectoral impacts and sector classification - The FSAP considered 16 different economic sectors (Table 4) for the micro simulation based on scope and data availability. - Sectoral variables include inter-industry relationships, price adjustments resulting from carbon taxes as well as from other policies, and global demand and supply drivers from international trade. - Under NZ and DT scenarios, output would decrease for most sectors relative to the baseline with variations: - The non-renewable electricity generation sector output declines substantially relative to the baseline as electricity production transitions to renewable sources. - The oil and gas sector is particularly affected. - Renewable energy output increases significantly. - Intermediate input exhibits similar patterns. - Total emissions decrease significantly for every sector except for electronic equipment and renewable energy, which experience output growth relative to the baseline pushing their total emissions slightly higher. - Note: The renewable energy sector (n. 13) includes both renewable electricity generation (nuclear, hydro, wind, solar, and others) and electricity transmission and distribution. ### Firm module: data and construction of T0 - Micro data sources for Canadian NFCs: - Profit and loss (P&L) and balance sheet data from Moody’s Orbis database. - PD data from CreditEdge Moody’s. - Emissions data from Intercontinental Exchange (ICE). - Data filtering and coverage: - Initial Orbis dataset has close to one million entities domiciled in Canada across all years; only less than one percent report balance sheet and P&L data necessary for the model simulation. - After applying filters, Orbis dataset between 2000-2022 is around 5,000 firms. - Credit Edge contains around 4,000 firms across 2000-2024; intersection between Orbis and Credit Edge used for PD modelling is around 2,150 firms. - Industry-median emission intensities data (at NACE4 level) for the North America region from ICE were adopted. - “Time 0” (T0) database construction: - Only data from recent years (2018–23) were kept to construct a “steady state” dataset. - Flow variables were constructed as 3-year averages; for stock variables the latest available year’s observations were adopted. - Observations from sectors with very few firms and corresponding to an insignificant share of banks’ loan portfolios were dropped. - Final T0 dataset consists of 2,245 firms. ### Firms’ financial health at T0 - Four variables used to capture firms’ solvency and liquidity: - LEV: leverage defined as debt over total assets. - EBITR: earnings before interest and taxes (EBIT) over total assets. - ICR: interest coverage ratio defined as EBIT over interest expense. - CDR: cash debt ratio defined as cash over short-term debt. - Most sectors in the T0 sample have low or negative profitability and negative ICR (insufficient earnings to cover interest payments). - The oil and gas sector is one of the sectors with low leverage and positive profitability at T0. - Caveat: The Orbis sample consists almost entirely of listed firms and may not be fully representative of DTIs’ loan portfolios, which likely contain a significant share of smaller firms with smaller financial buffers; risk analysis based on the Orbis sample may underestimate adverse impact on DTIs’ actual loan portfolios. ### T0 sample sectoral counts and assets (Table 5) - Total firms (full sample): 4,967 (100.0%) - T0 sample firms: 2,245 (100.0%) - Total Assets USDmm (T0 sample): 2,939,631 (100.0%) - Selected sector entries (Number of Firms (full sample) / Number of Firms (T0 sample) / Total Assets USDmm (T0 sample) / Share): - Agriculture & forestry: 390 / 8% ? (table formatting retained as in source) / 251 / 1% / 15,972 / 0.5% - Oil & gas: 447 / 9.0% / 149 / 6.6% / 317,198 / 10.8% - Other mining: 1,010 / 20.3% / 327 / 14.6% / 169,529 / 5.8% - Metals (fer&non-fer) manuf: 1,110 / 22.3% / 476 / 21.2% / 268,924 / 9.1% - Elect Gen-RE & Elect T&D: 50 / 1.0% / 23 / 1.0% / 220,686 / 7.5% - Other Busi services: 1,024 / 20.6% / 529 / 23.6% / 1,194,891 / 40.6% - (Table preserves the sectoral structure and exact numeric totals presented in the source.) ### Firms’ PD and LGD modelling - PD model specification: stock and flow-oriented Merton model rationale implemented as a sector-level fixed effects panel econometric equation; macro-financial variables were not included on the right-hand side because they feed into other regressors. - PD model results (Table 6, Logit (PD) RHS coefficients and statistics): - LEV: 3.280*** (.0770) - ICR: -.0028*** (.0003) - EBITR: -.5354*** (.0362) - CDR: -.0144*** (.0022) - cons: 5.288*** (.0266) - Observations: 17,060 - Within R2: 0.2079 - Notes: Standard errors in parentheses. *** p<.01. ICR, EBITR and CDR are winsorized at 5 and 95 percent, LEV is winsorized at 1. - Interpretation: leverage plays a prominent role; increases in leverage imply higher PDs, while ICR, EBITR and CDR have negative impacts on PDs. - PiT PDs were derived from Credit Edge data for listed firms; TTC PDs were computed as long-term averages of PiT PDs. - LGD modelling: - LGDs were derived from PDs using the Frye and Jacobs (2012) model with the formula presented in the source, where Φ denotes the standard normal cumulative distribution function, kf is driven by TTC PDs and LGDs (PDf,TTC and LGDf,TTC), and asset correlation parameter ρ is used. - At T0 PiT LGDs were set as the industry-level LGDs using DTIs’ data (weighted average across all DTIs using their NPL stocks as weights). The formula was used to derive TTC LGDs from TTC PDs and PiT PDs and LGDs. ### Firms’ P&L simulation channels - Three key channels link sectoral variables to firm-level variables: - Firms’ sales revenues evolve in line with sectoral output growth. - Firms’ operating expenses follow sectoral intermediate input growth. (It was not possible to divide operating expenses between inputs cost and labor cost due to data gaps in Orbis.) - Firms’ emissions evolve in line with sectoral emissions path; the direct cost of firms’ emissions is subtracted from their EBIT. - Mechanics of distress amplification: - As profitability erodes, firms may borrow more to continue operations, increasing leverage. - Post-tax profits are added to firms’ cash balances and total assets. - When a firm’s cash turns negative during the simulation, firms take on more debt to cover the shortfall. - Increasing leverage drives up firms’ PDs and creates a negative cycle with no options to deleverage in the model. ### Bank module: data, coverage, and aggregation - Bank module uses confidential DTIs’ outstanding NFCs loan portfolios across industry segments. - OSFI and AMF provided loan portfolio data for the 7 systemic DTIs as of Q3: 2024. - Dataset constructed from borrower-level data (Rapid 2) aggregated to the 16 sectors used in the exercise; mapping between 16 sectors and NAICS codes was provided to authorities. - Dataset includes exposures to NFCs as well as PDs and LGDs: - PiT PDs used for economic capital. - TTC PDs used for regulatory capital. - PiT LGDs computed as allowance for expected credit losses for stage 3 loans over borrower total outstanding. - Downturn LGDs used as the LGD for regulatory purposes. - Data coverage limitations: - Data only covers NFCs loans in excess of CAD 10 million, leaving out smaller exposures. - FSAP team estimated coverage of this dataset is 60 percent of the total DTI sector NFC loans. ### DTI portfolio composition and concentration - Aggregate DTI sector exposures by sector: - Largest share (55 percent) of exposures of the aggregate DTI sector is towards the “other business services” sector. - Construction: 14 percent. - Heavy emitting sectors (non-renewable electricity generation, oil and gas, transport, other mining) represent approximately 13 percent of the total loan portfolio. - Sample caveats: - Coverage is only 60 percent of DTIs’ portfolios and biased towards larger exposures. - Authorities find larger concentrations in sectors exposed to transition risks when looking at a broader set of DTIs included in the SCSE; for example, oil and gas alone represents 6 percent of their exposures. ### DTIs’ credit loss simulation approach - Firms’ PDs and LGDs were aggregated at the 16 sector-level, weighting each firm by its total assets. - Sectoral PDs and LGDs were then used to simulate forward in time PDs and LGDs for the seven DTIs, anchored at T0 in reported data. - The evolution of the non-performing loan stock and the computation of provisions (loan losses) follow the simplified equations presented in the source: - The NPL evolution equation uses write-off rate (WRO) and cure rate (CURER) assumptions where WRO is set to 100 percent and CURER to zero percent. - Provisions (PROV) and loan loss (LL) flows are computed as specified, and results calculate annual expected losses and do not follow the IFRS 9 methodology. ### Data and methodological limitations - Key caveats highlighted: - DTIs’ NFC loan portfolio data is restricted to large loans, leaving out smaller and potentially more vulnerable firms. - Firm-level data may not be fully representative of DTIs’ NFC loan portfolios at a granular level. - Substantial uncertainties exist around firms’ emission intensity levels. - Firms’ simulation does not capture all transmission channels of transition risks to individual firms. - Banks’ simulation focuses only on credit risk from NFCs and ignores other risk channels. - An analysis using a macro-approach (Adrian et al. 2022) on the full DTIs’ balance sheet would likely lead to larger impacts. - The scenario analysis ends in 2040 when adverse impact of a DT starts to become visible; extending beyond 2040 would likely present a less sanguine picture about the adverse impact of a DT. ### Results summary - Several factors drive results: transition scenarios at sectoral level, firms’ emissions intensities, firms’ initial risk characteristics, DTIs’ exposures to industries, and DTIs’ initial loan portfolio risk profile. - Under the transition scenarios, sectoral PDs and LGDs would generally increase modestly but with significant variation across sectors. - Notable adverse impacts are observed for the oil and gas and non-renewable electricity generation sectors; these sectors experience the largest absolute and relative increases in PDs under the two transition scenarios relative to baseline. *Source: Introduction section of the Canada FSAP transition risk analysis (IMF staff).* ### 2040. These two sectors are indeed the sectors with the largest emission intensities by far (Figur ### Canada: Transition Risk Results and Wildfire Risk Methodology ### Sector emission intensities and notable sectors - The two heaviest emitters are "oil & gas" and "electricity non-renewable", described as the sectors with the largest emission intensities by far. - Among other sectors, construction and mining also stand out for higher emission intensities. ### Scenario timing and impacts on Probability of Default (PD) and Loss Given Default (LGD) - The adverse impact is larger under Net-Zero 2050 than under DT for most of the simulation period because, unlike DT, under Net-Zero 2050 stringent climate policies are implemented from the beginning of the simulation period. - Sectors’ PDs increase under the Net-Zero 2050 scenario relative to the baseline in 2030. DT is not yet impactful in 2030. - When the DT initiates after 2030, PDs start increasing; by 2040 the DT impacts overcome the ones under Net-Zero 2050 for most sectors. - LGDs follow similar dynamics to PDs. - The relative changes in PDs and LGDs at the bank-level are broadly in line with the aggregate sector-level relative PD and LGD changes reported in the Bank of Canada transition risk analysis study (Bruneau et al (2023)). Key timeline references: - 2030: Net-Zero 2050 increases PDs relative to baseline; DT not yet impactful. - 2040: DT impacts overcome Net-Zero 2050 impacts for most sectors; simulation period ends in 2040. ### Deposit-Taking Institutions’ (DTIs) credit losses and drivers - DTIs’ credit losses under DT begin to exceed the ones under Net Zero in 2040. - DTIs' credit losses are estimated to increase between 3.8 and 7.1 percent under the Net Zero 2050 and DT scenarios relative to CP by 2040, depending on the bank and scenarios. - The cumulative impact is still greater under the Net Zero 2050 scenario over the simulation period; however, this is driven by the fact that the simulation period ends in 2040, just as the adverse effects of the DT—with its accelerated transition actions—begin to exceed those of the Net Zero 2050 scenario. - Credit losses are largely driven by sectors where DTIs have their largest exposures: other business services, transport, and construction. - Looking at relative change in losses by sector highlights that oil & gas and electricity non-renewable sectors, the two heaviest emitters, are the ones most impacted by the transition in the banks’ NFC portfolio. Figure and data notes referenced in the text: - Figure 27: Firms’ Probability of Defaults Level Change (2030, 2035, 2040; with and without Oil & Gas). - Figure 28: Firms’ Loss Given Defaults Level Change (2030, 2035, 2040; with and without Oil & Gas). - Figure 29: Firms’ PD and LGD Percentage Change in 2040. - Figure 30: Deposit Taking Institutions’ Loan Losses (box plot of loan losses change for the seven institutions). - Figure 31: Deposit Taking Institutions’ Credit Losses by Sector (Net Zero 2050 and DT; relative change by sector). ### Wildfire risk analysis: First-best methodology (Annex I) - Physical risks are modeled as the interaction of three components: hazard, exposure, and vulnerability. Changes in the climate system and socioeconomic processes including adaptation and mitigation drive hazards, exposure, and vulnerability. - Exposure focus: residential properties in Canada serving as collateral for DTIs’ mortgages. Ideal data elements: - Address-level geocoded locations (latitude and longitude). - Structure characteristics (type of roof, exposed exterior vents). - Building code information. - Projections of how exposures change in the future (structure, location, value). - Hazard data needs: likelihood (probability for a location to experience a wildland fire) and intensity (amount of energy produced by a fire) under current and future climate for different climate scenarios at sufficiently granular geographic resolution across Canada. - Example model: probabilistic model Burn P-3 used in Erni et al. (2024). - Wang et al. (2016) used Burn P-3 to simulate future climate impacts for south-central British Columbia; results show future climate increases number of fires and fire-conducive weather, leading to widespread increases in burn probability in that region. - Wang et al. (2020) find that fire size and annual area burned across Canada may increase by 20–64 percent and 25–93 percent respectively under future climate. - Future fire behavior depends on weather changes, vegetation, and ignition; additional influencing factors include pest outbreaks, fire suppression, and other mitigation actions. There remains limited understanding of how these factors impact fire behavior. - Vulnerability: use of damage functions quantifying rate of damage of exposure for varying levels of hazard intensity. Ideally, damage functions would be calibrated for Canada’s residential buildings and account for building characteristics and surrounding conditions (vegetation, other structures). - Combining hazard, exposure, and vulnerability datasets would allow estimation of wildfire risk under current and future climate and quantify expected economic losses to residential buildings at each location in Canada. - Due to lack of these datasets, the FSAP team adopted a simplified methodology. ### Methodological and dataset references cited - Burn P-3 probabilistic modeling referenced in Erni et al. (2024) and Wang et al. (2016). - Quantitative projections: fire size and annual area burned increases of 20–64 percent and 25–93 percent respectively under future climate (Wang et al. (2020)). - Comparison alignment: Bank of Canada transition risk analysis study (Bruneau et al (2023)). *Source: IMF staff calculations and Annex I methodology text from the Canada FSAP chapter.* --- _Source: https://www.imf.org/-/media/files/publications/cr/2025/english/1canea2025006-source-pdf.pdf_