## Online Annex 5.1. Data Sources, Climate Science Overview, Descriptive Statistics ## Source details **Canonical URL:** [Online Annex 5.1. Data Sources, Climate Science Overview, Descriptive Statistics](https://www.imf.org/-/media/files/publications/gfsr/2020/april/english/onlineannex51.pdf) ## Other formats - [Markdown version](/-/media/files/publications/gfsr/2020/april/english/onlineannex51.pdf.md) - [Structured JSON version](/-/media/files/publications/gfsr/2020/april/english/onlineannex51.pdf.json) --- ### Data sources and key variables - Macroeconomic variables (IMF, World Economic Outlook unless noted): - Real Gross Domestic Product: Gross domestic product, constant prices in national currency - Nominal Gross Domestic Product: Gross domestic product, current prices in national currency - Nominal Gross Domestic Product, World: Gross domestic product, world, current prices in US dollars - Gross Domestic Product Deflator: Gross domestic product, current prices in national currency - Consumer Price Index: Consumer price index, percent - Nominal Private Consumption Expenditure: Private consumption expenditure, current prices - Real Private Consumption Expenditure: Private consumption expenditure, constant prices - Short-Term Nominal Interest Rate: Three-month treasury bill or interbank rate — Refinitiv Datastream; and Haver Analytics - Long-Term Government Bond Yield: Ten-year government bond yield — Refinitiv Datastream; and Haver Analytics - Government Debt: Public debt, percent of GDP - Demographic and gridded population: - Population: Total population in millions — IMF, World Economic Outlook - Population, gridded: Grid-level population estimates for the year 2000 — NASA SEDAC - Aggregate and firm-level financial variables (Refinitiv Datastream unless noted): - Market Capitalization, Stock Market Price Index, Price-to-Earnings Ratio, Total Return Index, Dividend Yield - Firm-level: Book Value per Share, Market Capitalization (US dollars), Stock Price (most recent closing price within the last sixty days, local currency), Total Return Index, Price-to-Book Ratio - Climate variables and indices: - Observed temperature and precipitation, gridded: monthly time-series 1900 to 2017 — University of Delaware, Department of Geography - Projected temperature and precipitation: RCP model projections by economy — KNMI, Climate Change Atlas - Δ Extreme heat: Number of Days with Dangerous Heat (Heat Index > 35°C). Anomaly to baseline period 1986-2005 — World Bank Group, Climate Change Knowledge Portal - Δ Extreme precipitation: Average count of days per month or year with at least 50mm of daily rainfall. Anomaly to baseline period 1986-2005 — World Bank Group, Climate Change Knowledge Portal - Δ Heatwave probability: Daily probability of a 3+ day sequence above long-term 95th percentile. Anomaly to baseline period 1986-2005 — World Bank Group, Climate Change Knowledge Portal - Δ Drought likelihood: Annual probability of at least severe drought (SPEI <-2). Anomaly to baseline period 1986-2005 — World Bank Group, Climate Change Knowledge Portal - Climate Change Hazard Index, Sea level rise index, Climate Change Adaptive Capacity Index, Climate Change Sensitivity Index, Climate Change Physical risk index — Verisk Maplecroft (definitions and ranges preserved exactly as in source) - Disaster variables (EM-DAT): - Disaster damages: Amount of damage to property, crops, and livestock in US$ (‘000) - Large disaster definition: rate of affected population > 0.5 percent, or damages > 0.05 percent of GDP - EM-DAT coverage notes: recorded disasters conform to at least one criterion: 10 or more deaths; 100 or more people affected; declaration of a state of emergency and/or call for international assistance. Non-climatic disasters excluded. Reported damages generally cover direct costs only; indirect costs and vulnerability-reduction investments are not considered. - Bond market and CAT bond data: - Spread to benchmark, Maturity, Size (Deal value in USD) — Dealogic - VIX: CBOE Volatility Index — Refinitiv Datastream - CAT Bond Issuance Volume: Face value in USD for bonds issued between 1997 and 2019 — Dealogic, Bloomberg and Artemis - CAT Bond Original Maturity, CAT Bond Spreads, CAT Bond Expected Loss, CAT Bond Hurricane Exposure — Dealogic; Bloomberg; Artemis; Lane Financial - Other indicators: - Worldwide Governance Indicators — World Bank Group - Insurance penetration: reported insurance premium volume as percent GDP — World Bank Group, Global Financial Development Database - Sovereign Rating: annual average of foreign currency long-term sovereign debt ratings by Moody’s, Standard & Poor’s, and Fitch Ratings — Kose and others 2017 (World Bank Group, fall 2019 version) - Assets Held by Sustainable Equity Funds: Fund size in USD of equity funds labeled “Sustainable Investment – Environmental” — Morningstar ### Sample economies and sample periods (exact listings) - Advanced Economies (each with exact period): Australia 1973–2019; Austria 1973–2019; Belgium 1973–2019; Canada 1973–2019; Cyprus 1992–2019; Czech Republic 1995–2019; Denmark 1973–2019; Estonia 1997–2019; Finland 1988–2019; France 1973–2019; Germany 1973–2019; Greece 1988–2019; Hong Kong SAR 1973–2019; Ireland 1973–2019; Israel 1993–2019; Italy 1973–2019; Japan 1973–2019; Korea 1987–2019; Lithuania 1998–2019; Luxembourg 1992–2019; Malta 2000–2019; Netherlands 1980–2019; New Zealand 1988–2019; Norway 1980–2019; Portugal 1990–2019; Singapore 1973–2019; Slovak Republic 2006–2019; Slovenia 1998–2019; Spain 1987–2019; Sweden 1982–2019; Switzerland 1973–2019; Taiwan Province of China 1987–2019; United Kingdom 1970–2019; United States 1973–2019. - Emerging Market and Developing Economies (each with exact period): Argentina 1988–2019; Bahrain 2003–2019; Brazil 1994–2019; Bulgaria 2000–2019; Chile 1989–2019; China 1993–2019; Colombia 1992–2019; Croatia 2005–2019; Egypt 1996–2019; Hungary 1991–2019; India 1990–2019; Indonesia 1990–2019; Jordan 2006–2019; Kuwait 2003–2019; Malaysia 1986–2019; Mexico 1988–2019; Morocco 1994–2019; Nigeria 2009–2019; Oman 2005–2019; Pakistan 1992–2019; Peru 1994–2019; Philippines 1987–2019; Poland 1994–2019; Qatar 2003–2019; Romania 1996–2019; Russia 1998–2019; Saudi Arabia 2005–2019; South Africa 1973–2019; Sri Lanka 1987–2019; Thailand 1987–2019; Turkey 1988–2019; United Arab Emirates 2003–2019; Venezuela 2012–2019; Vietnam 2007–2019. ### Climate change scenarios and observed past changes (IPCC 2014 — exact figures) - Representative Concentration Pathways (RCPs) — Likely Increase of Global Mean Surface Temperature by the End of the 21st Century (2081–2100) relative to 1986–2005: - RCP 2.6: 0.3°C to 1.7°C. - RCP 4.5: 1.1°C to 2.6°C. - RCP 6.0: 1.4°C to 3.1°C. - RCP 8.5: 2.6°C to 4.8°C. - Observed past changes and human contribution (IPCC 2014, Bouwer 2019) — exact phrasing preserved: - Warmer (and/or fewer cold) days and nights - Observed past changes: Very likely increase (decrease) in frequency over most land areas - Human contribution: Very likely - Heat Waves - Observed past changes: Medium confidence in increase on global scale; Likely increase in large parts of Europe, Asia, and Australia - Human contribution: Likely - Heavy Precipitation - Observed past changes: Likely increases over more land areas than decreases - Human contribution: Medium confidence - River Floods - Observed past changes: Limited to medium evidence for changes in frequency of river floods at the regional level; Low confidence for sign of change of river floods at the global level - Human contribution: Low confidence - Drought - Observed past changes: Low confidence in change on a global level - Human contribution: Low confidence - Tropical Cyclones - Observed past changes: Low confidence in increase in activity (intensity and frequency) on timescales of 100 years; Virtually certain increase in activity in North Atlantic since 1970 - Human contribution: Low confidence - Sea-levels - Observed past changes: Likely increase since 1970 - Human contribution: Likely ### Large climatic disasters — sample highlights (exact measures as presented) - Definition and data notes: - Disaster damages measured in EM-DAT as US$ (‘000) total over full duration. - Large disaster if affected population rate > 0.5 percent, or damages > 0.05 percent of GDP. - Selected sample entries (Country, Year, Disaster Type, Damage-to-GDP Ratio (percent), Damage insured (percent), Duration, Number of Deaths, Number of Affected, Market Return (T-5 to T+20; percent)) — entries preserved as in source (examples): - Australia 1981 — Drought — 3.2 — 2 years — 80,000 — -6.9 - Czech Republic 1997 — Flood 1997 Central European flood — 3.0 — 17~ 1 month — 291 — 02,107 — 5.0 - Czech Republic 2002 — Flood 2002 European floods — 2.9 — 50~ 1 month — 182 — 00,000 — -1.5 - Spain 1983 — Flood — 2.3 — < 1 month — 455 — 06,000 - Denmark 1999 — Storm Anatol — 1.58 — 1< 1 month — 73.6 - Austria 2002 — Flood 2002 European floods — 1.1 — 17< 1 month — 960 — 000 — 0.4 - Canada 1976 — Drought — 1.1 — 4 years — 00.03 - Portugal 2003 — Wildfire — 1.0 — 2 months — 141 — 50,000 — 2.5 - Greece 1990 — Drought — 1.0 — ~ 1 month — 6.7 - United States 2005 — Storm Hurricane Katrina — 1.04 — 8~ 1 month — 1833 — 500,000 — -0.4 - Thailand 2011 — Flood 2011 Thailand floods — 10.1 — 255 months — 81 — 39,500,000 — -6.5 - Oman 2007 — Storm Cyclone Gonu — 9.3 — 16.7 — < 1 month — 76 — 20,000 — 4.4 - Jordan 1992 — Cold wave — 7.4 — < 1 month — 15 - Pakistan 2010 — Flood 2010 Pakistan floods — 5.4 — 1.1 — < 1 month — 1985 — 20,359,496 — -6.5 - Peru 1983 — Landslide — 5.2 — ~ 1 month — 364 — 700,000 - Philippines 2013 — Storm Typhoon Haiyan (Yolanda) — 3.7 — 7< 1 month — 7354 — 16,106,870 — -8.1 - Venezuela 1999 — Flood Vargas tragedy — 3.2 — 12.7 — ~ 1 month — 30000 — 483,635 — 6.1 - Indonesia 1997 — Wildfire — 3.1 — ~ 1 month — 2403 — 2,070 — 7.4 - Vietnam 2015 — Drought — 3.1 — 1 year — 1,750,000 — -3.1 - China 1998 — Flood Yangtze river flood — 2.9 — 1.0 — 2 months — 3656 — 239,000,000 — -32.9 ### Event-study and regression methodologies for equity returns - Event study setup: - Pre-event estimation window: starts 12 months before a disaster and ends one month before the disaster. - Excess stock return modeled via global factor: - r_{i,c,t}^e = α_i + β_{i,c}^{global} ⋅ r_{global,t}^e + ε_{i,c,t} - Expected return: E[ r_{i,c,τ}^e ] = β̂_{i,c}^{global} ⋅ r_{global,τ}^e - Abnormal return: AR̂(r_{i,c,τ}^e) = r_{i,c,τ}^e − E[ r_{i,c,τ}^e ] - CAR: sum of abnormal returns from 21 trading days before the start date up to 60 trading days after the disaster start date. - CAAR: average of CARs across all disasters in the sample. - Regression analysis of equity market reaction: - Model: - CCAAR40_{i,c,d} = μ_i + γ_{1,i} Charr_{c,d} + γ_{2,i} X_d + ε_{i,c,d} - CCAAR40_{i,c,d} is 40-trading-day cumulative abnormal returns for Index i ∈ {Market, Banks, Non-life Insurance, Industrial} - Charr_{c,d} includes insurance penetration (ratio of total non-life insurance premiums to GDP, one year lagged) and sovereign bond rating (scale 1 to 21, 21 highest, one year lagged) - X_d controls for economic damage-to-GDP ratio for disaster d - Quantile regression for downside risk: - CCAAR40_{i,c,d}^q = μ_{i}^q + γ_{1,i}^q Charr_{c,d} + γ_{2,i}^q X_d + ε_{i,c,d}^q - q is the 10th percentile, capturing the left tail of the distribution - Notes on sample inclusion: - Only disasters with precise start dates are included. ### Additional controls, reinsurance CAARs, and cross-economy pricing approach - Additional business cycle controls: - Change in the short-term interest rate (3-month government bond yield) relative to the end of the previous year - Change in the long-term interest rate (10-year government bond yield) relative to the end of the previous year - Reinsurance sector CAAR analysis: - Sample restricted to storms and floods; included all globally listed re-insurance companies and all disasters in the sample - Result: Cumulative average abnormal returns (CAARs) are negative and significant (see Online Annex Figure 5.2.1) - Abnormal returns computed using single global factor model with estimation using historical data one year before each disaster; X-axis trading days with time 0 as disaster start; dashed lines are 90th percent confidence intervals - Cross-economy regression approach (summary): - Dependent variables: log(P/EPS), log(P/BPS), log(P/Div) - Key regressor: predicted change in climatic hazard occurrence (Haarh measure) - Controls: proxies for expected future earnings growth, equity risk premium, risk-free interest rate (as in IMF (2019)) - Valuation controls include mean annual growth of earnings per share (past 5 years), standard deviation of annual growth of EPS (past 5 years), 3-month govt. bond yield (Average of monthly observations in 2019.) - Climatic hazard indicators: ΔExtreme heat, ΔExtreme precipitation, ΔHeatwave probability, ΔDrought likelihood, Sea level rise index, Climate Change Hazard Index - Estimations repeated for RCP 2.6, RCP 4.5, RCP 6.0, RCP 8.5 - Sample includes 50 economies in total - Augmented specification: - Adds indicator 1{High} equal to one when Climate Change Sensitivity Index or Climate Change Adaptive Capacity Index is above median - Interaction term: climatic hazard × 1{High} ### Robustness, sectoral tests, and discounting hypothesis - Robustness checks performed: - Alternative valuation measures: price-to-book, price-to-dividend - Yearly averages and longer averages (e.g. 5-10 years) - Augmentation with credit-to-GDP gap and Output gap - Interaction of climatic indicators with sovereign ratings - Addition of transition risk proxies - Different projection horizons of climatic hazard indicators - Robustness results: - Results are robust to these tests - No change in the pricing of climate change physical risk over time - No difference in pricing in economies with higher attention to climate change (measured by Google search index for “climate change” or share of sustainable fund investors relative to total market capitalization) - Sectoral tests and discounting hypothesis: - Agents with lower discount rates hypothesized to be more sensitive to long-horizon developments - Cross-sectional regressions by sector: insurance and industrials show some pricing of Climatic Hazard Risk and future Heat Wave likelihoods, but these effects do not hold once financial cycle controls are included - Specification with adaptive capacity and sensitivity indicators unchanged ### Temperature sensitivity and predictable equity returns (two-stage approach) - Sample and period: - 27 economies selected (criterion: at least 50 listed firms in the first month of 1998) - Regressions use 1993-2017 for all economies - Same mispricing pattern present when allowing earlier starting months for economies with data for at least 50 listed firms before 1998m1 - Stage 1 (rolling window 5 years / 60 months): - Regress firm excess returns on excess market return and temperature anomaly - Temperature anomaly = current temperature − average temperature over past 30 years in same month - Yields firm temperature sensitivity parameter θ_{i,c}; key variable |θ_{i,c}^T| - Stage 2 (Fama-Macbeth style, monthly): - Firms ranked by θ_{i,c}^T and grouped into quintiles; top quintile = highest temperature sensitivity - Dummy HighTempSens_{i,c,t} = 1 if firm in top quintile in month t - Next-month excess return regressed on HighTempSens dummy and controls X (Fama-French factors: excess market returns, SMB, HML) - Aggregate measure λ̂_c = (∑_{t=1}^{240} λ_{c,t})/240 over 1998m1 to 2017m12; significantly negative λ̂_c implies mispricing (high sensitivity stocks earn abnormal lower returns) - Robustness checks: - Add RMW, CMA, MOM; include firm-level controls (lagged returns, market cap, book-to-market) - Pricing anomaly remains - t-statistics computed using Newey-West adjusted standard errors with 3-month lags ### CAT bond market pricing of hurricane/storm risk - Data and sample: - 778 CAT bonds issued between 1997 and 2018 from Tomunen (2019); final sample after exclusions comprises 656 observations - Bonds classified by exposure to hurricane and storm risk (including European Windstorms and Pacific Typhoons) or not - Regression (OLS): - Dependent variable: spread over floating benchmark quoted of bond b on day t, measured in basis points - Regressors include: - annualized expected loss (in percentage points) from underlying actuarial risk model - dummy: bonds exposed to hurricane and storm risk - dummy: risk in the U.S. - maturity at issuance in years - log(total face value in U.S. dollars) - trigger type dummies: Parametric, Industry Loss Index, Modelled Loss, Indemnity, Multiple Triggers - year fixed effects - Coefficient of interest: effect of hurricane/storm exposure on bond spread ### Long-Run Risk (LRR) model with climatic disasters - Framework: - Builds on Bansal and Yaron (2004) with Epstein and Zin (1989) recursive preferences and early resolution of uncertainty - Introduces temperature-induced very large climatic disasters affecting consumption growth (Barro (2009) style) - Temperature informs probability of future very large climatic disasters - Key model elements: - Preferences include parameters δ (time preference), γ (risk aversion), ψ (IES) with θ = (1−γ)/(1−1/ψ) - Wealth evolution: W_{t+1} = (W_t − C_t) A_{C,t+1} - Consumption and dividends dynamics include μ_c drift, persistent E_t, common volatility σ_t, and drop D_{t+1} from temperature-induced disasters - Shocks η, T, w, V assumed i.i.d. normal and orthogonal - Log IMRS, linear approximations yield log returns for aggregate and market portfolios; risk premium determined by covariation with pricing kernel - Unknowns solved numerically via fixed-point problem - Estimation and calibration: - Parameter vector Θ includes {γ, ψ, δ, μ_c, ρ, φ_e, σ^2, v, σ_w, μ_d, φ_d, φ_d, h} - Macro parameters estimated using auxiliary model with generalized method of moments-type criterion via indirect inference; LRR-implied data simulated - Climatic scenarios considered: RCP 2.6, RCP 6.0, RCP 8.5 - Risk premia reported using average, 5th percentile and 95th percentile of temperature projections from CMIP5 ensemble for each scenario and compared to counterfactual with no future climate risk and to market-implied risk premia from a standard discounted cash-flow valuation model - Bootstrap simulations used to address small sample bias ### Calibration of very large climatic disasters and disaster probability - Very large disaster definition: - Disaster is very large if ratio of affected population to total population, ratio of deaths to total population, or damage-to-GDP ratio are above the 95th percentile of historical distribution - Impact on consumption: - Estimated via local projections (Jorda, 2005) - Indicator = 1 if very large disaster occurred at time t - Impact reported as deviation from a no-disaster scenario (percent) - Blue dots in figure indicate coefficients statistically significant at the 10 percent level or lower - Calibration of likelihood of climatic disasters: - Panel logit specification: - Pr(DisastersVT_{c,t} = 1) = Φ(f(T_{c,t}) + f(P_{c,t}) + θ_c + ε_{c,t}) - f(T_{c,t}) quadratic in temperature; f(P_{c,t}) quadratic in precipitation; θ_c economy fixed effects - Disaster probability calibration uses only wildfires, droughts and heat waves to define “T-Induced” - Logit-implied probabilities scaled by number of very large T-induced disasters over total observed T-induced disasters in EM-DAT to match prior disaster-size calibration - Model-implied quarterly probability figure notes: - Reports marginal quarterly probability of all T-Induced disasters for different temperature levels - T-induced disasters = Heat Waves, Wildfires and Droughts - Vertical lines show 95 percent confidence bounds - Temperature weighted by population in 2000 - Figure axis scale includes probabilities ranging from 0 to 0.7 and temperature values shown across a range including 0, 4, 8, 12, 16, 20, 24, 28 ### Pricing of physical risk in sovereign bonds — baseline specification and sample - Baseline regression (notation preserved from source): - Spread_{g,c,t} = α_{c,t} + β1 CViTaVT_c + β2 LongTerm_g,c,t + β3 CViTaVT_c × LongTerm_g,c,t + β4 Controls_g,c,t + ε_{g,c,t} - Dependent variable: spread between coupon rate and benchmark (basis points) - Long-term dummy: bonds with maturities after 2040 - Controls: log(issue size), log(maturity), VIX, sovereign credit rating - α_{c,t} are economy-year fixed effects - Seven climate measures used (one per regression): 1. projected change in number of extreme heat days 2. projected change in number of extreme precipitation days 3. projected change in heatwave likelihood 4. projected change in drought likelihood 5. Sea Level Rise Index 6. Climate Change Hazard Index 7. Climate Change Physical Risk Index - Coefficient of interest: β3 (interaction) - Projection horizons, scenario choice, and sample: - Projected changes available for four projection horizons up to year 2100 and four emission scenarios - Analysis concentrates on horizon 2020–39 and emission scenario RCP 8.5; results robust to other horizons/scenarios - Sample includes 41,211 bonds issued in 121 economies - 94 percent of bonds are issued by advanced economies - 54 percent of bonds are issued by the United States alone - Results robust to excluding the United States and to excluding economies not in main-text analysis ### Robustness and ancillary notes - Results remain broadly unchanged for robustness checks referenced - Disaster data and reported damages measured imperfectly; primarily direct costs covered; indirect costs and vulnerability-reduction investments generally excluded *International Monetary Fund | Online Annex 5.1 — GLOBAL FINANCIAL STABILITY REPORT: Climate Change: Physical Risk and Equity Prices (April 2020) — onlineannex51.pdf* ### References ____________________________________________________________ 23 ### Online Annex 5.1. Data Sources, Climate Science Overview, Descriptive Statistics ### Data sources (selected variables and exact sources) - Macroeconomic Variables - Real Gross Domestic Product: Gross domestic product, constant prices in national currency — IMF, World Economic Outlook - Nominal Gross Domestic Product: Gross domestic product, current prices in national currency — IMF, World Economic Outlook - Nominal Gross Domestic Product, World: Gross domestic product, world, current prices in US dollars — IMF, World Economic Outlook - Gross Domestic Product Deflator: Gross domestic product, current prices in national currency — IMF, World Economic Outlook - Consumer Price Index: Consumer price index, percent — IMF, World Economic Outlook - Nominal Private Consumption Expenditure: Private consumption expenditure, current prices — IMF, World Economic Outlook - Real Private Consumption Expenditure: Private consumption expenditure, constant prices — IMF, World Economic Outlook - Short-Term Nominal Interest Rate: Three-month treasury bill or interbank rate — Refinitiv Datastream; and Haver Analytics - Long-Term Government Bond Yield: Ten-year government bond yield — Refinitiv Datastream; and Haver Analytics - Government Debt: Public debt, percent of GDP — IMF, World Economic Outlook - Demographic Variables - Population: Total population in millions — IMF, World Economic Outlook - Population, gridded: Grid-level population estimates for the year 2000 — NASA Socioeconomic Data and Applications Center (SEDAC) - Aggregate Financial Variables (Refinitiv Datastream unless noted) - Market Capitalization: Market values calculated from constituents (US dollars) - Stock Market Price Index: Price as percentage of base date, adjusted for capital changes - Price-to-Earnings Ratio: Total market value divided by total earnings - Total Return Index: Return index assuming dividends re-invested - Dividend Yield: Total dividend amount as percent of total market value - Firm-level Variables (Refinitiv Datastream) - Book Value per Share: Proportioned common equity divided by outstanding shares - Market Capitalization: Current total market value (US dollars) - Stock Price: Most recent closing price within the last sixty days (local currency) - Total Return Index: Return index for a stock (dividends re-invested) - Price-to-Book Ratio: Ratio of market price to book value per share - Climate Variables and Indices - Observed temperature and precipitation, gridded: Grid-level monthly time-series 1900 to 2017 — University of Delaware, Department of Geography - Projected temperature and precipitation: Model projections under different Representative Concentration Pathways (RCPs) by economy — Koninklijk Nederlands Meteorologisch Instituut (KNMI), Climate Change Atlas - Δ Extreme heat: Number of Days with Dangerous Heat (Heat Index > 35°C). Anomaly to the baseline period of 1986-2005 — World Bank Group, Climate Change Knowledge Portal - Δ Extreme precipitation: Average count of days per month or year with at least 50mm of daily rainfall. Anomaly to the baseline period of 1986-2005 — World Bank Group, Climate Change Knowledge Portal - Δ Heatwave probability: Daily probability of a 3+ day sequence above long-term 95th percentile. Anomaly to baseline period 1986-2005 — World Bank Group, Climate Change Knowledge Portal - Δ Drought likelihood: Annual probability of at least severe drought (SPEI <-2). Anomaly to baseline period 1986-2005 — World Bank Group, Climate Change Knowledge Portal - Climate Change Hazard Index, Sea level rise index, Climate Change Adaptive Capacity Index, Climate Change Sensitivity Index, Climate Change Physical risk index — Verisk Maplecroft (definitions and ranges preserved exactly as in source) - Disaster Variables (EM-DAT) - Disaster damages: Amount of damage to property, crops, and livestock in US$ (‘000) — EM-DAT: The Emergency Events Database - Large disaster definition: A disaster is “large” if the rate of affected population > 0.5 percent, or damages > 0.05 percent of GDP — EM-DAT; IMF staff calculations - Notes on EM-DAT coverage: Recorded disasters conform to at least one criterion: 10 or more deaths; 100 or more people affected; declaration of a state of emergency and/or call for international assistance. Non-climatic disasters excluded. Reported damages generally cover direct costs only; indirect costs and vulnerability-reduction investments are not considered. - Bond Market and CAT Bond Data - Spread to benchmark, Maturity, Size (Deal value in USD) — Dealogic - VIX: CBOE Volatility Index — Refinitiv Datastream - CAT Bond Issuance Volume: Face value in USD for bonds issued between 1997 and 2019 — Dealogic, Bloomberg and Artemis - CAT Bond Original Maturity, CAT Bond Spreads, CAT Bond Expected Loss, CAT Bond Hurricane Exposure — Dealogic; Bloomberg; Artemis; Lane Financial (as specified) - Other indicators - Worldwide Governance Indicators — World Bank Group - Insurance penetration: Reported insurance premium volume as percent GDP — World Bank Group, Global Financial Development Database - Sovereign Rating: Annual average of foreign currency long-term sovereign debt ratings by Moody’s, Standard & Poor’s, and Fitch Ratings — Kose and others 2017 (World Bank Group, fall 2019 version) - Assets Held by Sustainable Equity Funds: Fund size in USD of equity funds labeled “Sustainable Investment – Environmental” — Morningstar ### Sample economies and sample periods (exact listings) - Advanced Economies (each with exact period) - Australia 1973–2019 - Austria 1973–2019 - Belgium 1973–2019 - Canada 1973–2019 - Cyprus 1992–2019 - Czech Republic 1995–2019 - Denmark 1973–2019 - Estonia 1997–2019 - Finland 1988–2019 - France 1973–2019 - Germany 1973–2019 - Greece 1988–2019 - Hong Kong SAR 1973–2019 - Ireland 1973–2019 - Israel 1993–2019 - Italy 1973–2019 - Japan 1973–2019 - Korea 1987–2019 - Lithuania 1998–2019 - Luxembourg 1992–2019 - Malta 2000–2019 - Netherlands 1980–2019 - New Zealand 1988–2019 - Norway 1980–2019 - Portugal 1990–2019 - Singapore 1973–2019 - Slovak Republic 2006–2019 - Slovenia 1998–2019 - Spain 1987–2019 - Sweden 1982–2019 - Switzerland 1973–2019 - Taiwan Province of China 1987–2019 - United Kingdom 1970–2019 - United States 1973–2019 - Emerging Market and Developing Economies (each with exact period) - Argentina 1988–2019 - Bahrain 2003–2019 - Brazil 1994–2019 - Bulgaria 2000–2019 - Chile 1989–2019 - China 1993–2019 - Colombia 1992–2019 - Croatia 2005–2019 - Egypt 1996–2019 - Hungary 1991–2019 - India 1990–2019 - Indonesia 1990–2019 - Jordan 2006–2019 - Kuwait 2003–2019 - Malaysia 1986–2019 - Mexico 1988–2019 - Morocco 1994–2019 - Nigeria 2009–2019 - Oman 2005–2019 - Pakistan 1992–2019 - Peru 1994–2019 - Philippines 1987–2019 - Poland 1994–2019 - Qatar 2003–2019 - Romania 1996–2019 - Russia 1998–2019 - Saudi Arabia 2005–2019 - South Africa 1973–2019 - Sri Lanka 1987–2019 - Thailand 1987–2019 - Turkey 1988–2019 - United Arab Emirates 2003–2019 - Venezuela 2012–2019 - Vietnam 2007–2019 ### Climate change scenarios and past changes in extreme weather events (IPCC 2014 — exact figures) - Representative Concentration Pathways (RCPs) — Likely Increase of Global Mean Surface Temperature by the End of the 21st Century (2081–2100) relative to 1986–2005 - RCP 2.6: 0.3°C to 1.7°C. - RCP 4.5: 1.1°C to 2.6°C. - RCP 6.0: 1.4°C to 3.1°C. - RCP 8.5: 2.6°C to 4.8°C. - Observed past changes and human contribution (IPCC 2014, Bouwer 2019) — exact phrasing preserved: - Warmer (and/or fewer cold) days and nights - Observed past changes: Very likely increase (decrease) in frequency over most land areas - Human contribution: Very likely - Heat Waves - Observed past changes: Medium confidence in increase on global scale; Likely increase in large parts of Europe, Asia, and Australia - Human contribution: Likely - Heavy Precipitation - Observed past changes: Likely increases over more land areas than decreases - Human contribution: Medium confidence - River Floods - Observed past changes: Limited to medium evidence for changes in frequency of river floods at the regional level; Low confidence for sign of change of river floods at the global level - Human contribution: Low confidence - Drought - Observed past changes: Low confidence in change on a global level - Human contribution: Low confidence - Tropical Cyclones - Observed past changes: Low confidence in increase in activity (intensity and frequency) on timescales of 100 years; Virtually certain increase in activity in North Atlantic since 1970 - Human contribution: Low confidence - Sea-levels - Observed past changes: Likely increase since 1970 - Human contribution: Likely ### Large climatic disasters—sample highlights (exact measures as presented) - Definition and data notes: - Disaster damages measured in EM-DAT as US$ (‘000) total over full duration. - Large disaster if affected population rate > 0.5 percent, or damages > 0.05 percent of GDP. - Selected sample entries (Country, Year, Disaster Type, Damage-to-GDP Ratio (percent), Damage insured (percent), Duration, Number of Deaths, Number of Affected, Market Return (T-5 to T+20; percent)) — entries preserved as in source: - Australia 1981 — Drought — 3.2 — 2 years — 80,000 — -6.9 - Czech Republic 1997 — Flood 1997 Central European flood — 3.0 — 17~ 1 month — 291 — 02,107 — 5.0 - Czech Republic 2002 — Flood 2002 European floods — 2.9 — 50~ 1 month — 182 — 00,000 — -1.5 - Spain 1983 — Flood — 2.3 — < 1 month — 455 — 06,000 - Denmark 1999 — Storm Anatol — 1.58 — 1< 1 month — 73.6 - Austria 2002 — Flood 2002 European floods — 1.1 — 17< 1 month — 960 — 000 — 0.4 - Canada 1976 — Drought — 1.1 — 4 years — 00.03 - Portugal 2003 — Wildfire — 1.0 — 2 months — 141 — 50,000 — 2.5 - Greece 1990 — Drought — 1.0 — ~ 1 month — 6.7 - United States 2005 — Storm Hurricane Katrina — 1.04 — 8~ 1 month — 1833 — 500,000 — -0.4 - Thailand 2011 — Flood 2011 Thailand floods — 10.1 — 255 months — 81 — 39,500,000 — -6.5 - Oman 2007 — Storm Cyclone Gonu — 9.3 — 16.7 — < 1 month — 76 — 20,000 — 4.4 - Jordan 1992 — Cold wave — 7.4 — < 1 month — 15 - Pakistan 2010 — Flood 2010 Pakistan floods — 5.4 — 1.1 — < 1 month — 1985 — 20,359,496 — -6.5 - Peru 1983 — Landslide — 5.2 — ~ 1 month — 364 — 700,000 - Philippines 2013 — Storm Typhoon Haiyan (Yolanda) — 3.7 — 7< 1 month — 7354 — 16,106,870 — -8.1 - Venezuela 1999 — Flood Vargas tragedy — 3.2 — 12.7 — ~ 1 month — 30000 — 483,635 — 6.1 - Indonesia 1997 — Wildfire — 3.1 — ~ 1 month — 2403 — 2,070 — 7.4 - Vietnam 2015 — Drought — 3.1 — 1 year — 1,750,000 — -3.1 - China 1998 — Flood Yangtze river flood — 2.9 — 1.0 — 2 months — 3656 — 239,000,000 — -32.9 ### Large climatic disasters and equity returns — methodology (exact windows and model specifications) - Event study setup - Pre-event estimation window: starts 12 months before a disaster and ends one month before the disaster. - Excess stock return for index i in economy c on day t (r_{i,c,t}^e) is modeled via a global factor model: - r_{i,c,t}^e = α_i + β_{i,c}^{global} ⋅ r_{global,t}^e + ε_{i,c,t} - where r_{global,t}^e is the excess return for the global stock index (excess of the risk-free rate of the U.S.), and β_{i,c}^{global} is the loading on the global factor. - Estimation is conducted at the daily frequency. Returns use the Stock Market Price Index. - Expected return at time τ after the disaster: E[ r_{i,c,τ}^e ] = β̂_{i,c}^{global} ⋅ r_{global,τ}^e - Abnormal return at time τ: AR̂(r_{i,c,τ}^e) = r_{i,c,τ}^e − E[ r_{i,c,τ}^e ] - Cumulative Abnormal Return (CAR) computed by summing abnormal returns from 21 trading days before the start date of the disaster up to 60 trading days after the disaster start date. - Cumulative Average Abnormal Return (CAAR) is the average of the CARs across all disasters in the sample. - Regression analysis of equity market reaction to large disasters - Model estimated: - CCAAR40_{i,c,d} = μ_i + γ_{1,i} Charr_{c,d} + γ_{2,i} X_d + ε_{i,c,d} - where CCAAR40_{i,c,d} is the 40-trading-day cumulative abnormal returns for Index i ∈ {Market, Banks, Non-life Insurance, Industrial} in economy c after disaster d. - Charr_{c,d} includes economy characteristics at the time of disaster d: insurance penetration (ratio of total non-life insurance premiums to GDP, one year lagged), and sovereign financial strength (sovereign bond rating on a scale from 1 to 21, 21 being the highest, one year lagged). - X_d controls for the economic damage-to-GDP ratio for disaster d. - Quantile regression to examine downside risk: - CCAAR40_{i,c,d}^q = μ_{i}^q + γ_{1,i}^q Charr_{c,d} + γ_{2,i}^q X_d + ε_{i,c,d}^q - where q is the 10th percentile, capturing the left tail of the distribution of cumulative abnormal returns. - Notes on sample inclusion - Only disasters for which precise start dates are available are included in the analysis. ### Robustness and ancillary notes - Results remain broadly unchanged for the robustness checks listed (robustness checks referenced; details in the source). - Disaster data and damage measures: reported damages measured imperfectly; primarily direct costs covered; indirect costs and investment for vulnerability reduction generally excluded. *International Monetary Fund | Online Annex 5.1 (References) — GLOBAL FINANCIAL STABILITY REPORT—Climate Change: Physical Risk and Equity Prices* ### 1. To account for business cycle effects and government policy responses, the change in the short-term ### onlineannex51 - 1. To account for business cycle effects and government policy responses, the change in the short-term ### Additional control variables for business cycle and policy effects - Added controls: - Change in the short-term interest rate (3-month government bond yield) relative to the end of the previous year. - Change in the long-term interest rate (10-year government bond yield) relative to the end of the previous year. ### CAAR analysis for disasters and reinsurance sector - Sample restriction to storms and floods to capture more acute disaster effects. - Repeated CAAR analysis including all globally listed re-insurance companies and all disasters in the sample. - Result: Cumulative average abnormal returns (CAARs) are negative and significant (see Online Annex Figure 5.2.1). - Figure specifics: - Plots CAARs for the global reinsurance sector during trading days around large climatic disasters. - Global reinsurance sector return is simple average of daily returns of all publicly-traded reinsurance companies globally. - Abnormal returns computed using a single global factor model with estimation using historical data one year before each disaster. - X-axis: trading days; time 0 is the start day of disasters. - Dashed lines are the 90th percent confidence intervals. - Data sources: Refinitiv Datastream and IMF staff estimates. ### Cross-economy evidence: empirical approach (Online Annex 5.3) - Estimated cross-sectional regression (Equation (1)): - Dependent variable: 푉푉푎푎푉푉푉푉 푎푎푉푉 푖푖푉푉푛푛 푐푐 — either log(P/EPS), log(P/BPS), or log(P/Div). - Key regressor: 퐻퐻푎푎퐻퐻 푎푎푟푟퐻퐻 푐푐 — measure of predicted change in climatic hazard occurrence. - Controls (푉푉푎푎푉푉푉푉 푎푎푉푉 푖푖푉푉푛푛 퐶퐶푉푉푛푛푉푉푟푟푉푉푉푉 푠푠 푐푐): proxies for expected future earnings growth, the equity risk premium, and the risk-free interest rate (as in IMF (2019)). - Valuation metrics and climatic hazard indicators: - Valuation metrics: - Log of Price-to-Earnings per Share Ratio (P/EPS) - Log of Price-to-Book per Share Ratio (P/BPS) - Log of Price-to-Dividend per Share Ratio (P/Div) - Climatic hazard indicators: - Measures of future changes in hazard occurrence (Maplecroft indicators, World Bank indicators) - Valuation controls: - mean annual growth of earning per share (past 5 years) - standard deviation of annual growth of earnings per share (past 5 years) - 3-month govt. bond yield (Average of monthly observations in 2019.) - Scenarios and sample: - Estimations repeated for RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5. - Sample of economies with available data includes 50 economies in total. - Augmented specification (Equation (2)) to assess climate change sensitivity and adaptive capacity: - Includes indicator ퟙퟙ{퐻퐻푖푖 퐻퐻ℎ}푐푐 equal to one when the Climate Change Sensitivity Index or the Climate Change Adaptive Capacity Index is above the median of the cross-economy distribution. - Interaction term: 퐻퐻푎푎퐻퐻푎푎푟푟퐻퐻 푐푐 X ퟙퟙ{퐻퐻푖푖 퐻퐻ℎ}푐푐. - Note on climatic hazard risks considered: - ΔExtreme heat, ΔExtreme precipitation, ΔHeatwave probability, ΔDrought likelihood, Sea level rise index, Climate Change Hazard Index (See Online Annex Table 5.1.1 for data sources and definitions). ### Robustness analysis (Cross-economy pricing) - Robustness checks performed: - Use of price-to-book and price-to-dividend ratios as alternative market valuation measures. - Regressions using yearly averages of market valuation metrics and averages over longer periods (e.g. 5-10 years) to control for aggregate state of the economy. - Augmentation with credit-to-GDP gap and Output gap measures to examine confounding effects from the economic and financial cycle. - Inclusion of interaction terms of climatic hazard indicators with sovereign ratings to test confounding due to political risk and fiscal capacity. - Addition of transition risk proxies as control variables. - Use of different projection horizons of the climatic hazard indicators. - Robustness results: - Results are robust to these additional tests. - No change in the pricing of climate change physical risk over time. - No difference in pricing in economies with higher levels of attention to climate change (measured using the Google search index for the topic “climate change” or with the share of sustainable fund investors relative to total market capitalization). - Sectoral tests and discounting hypothesis: - Hypothesis: Agents with lower discount rates are more sensitive to developments expected several years in the future. - Cross-sectional regressions for sectors (banking, insurance, industrials, real estate, utilities): - Insurances and industrials show some degree of pricing of the Climatic Hazard Risk and future Heat Wave likelihoods. - These results do not hold once controls for the financial cycle are included in the baseline specification. - Specification with adaptive climate capacity and climate change sensitivity remains unchanged. ### Temperature sensitivity and predictable equity returns (Online Annex 5.4) - Sample and period: - 27 economies selected (criterion: at least 50 listed firms in the first month of 1998). - Regressions use sample period 1993-2017 for all economies. - Note: same pattern of mispricing present after allowing earlier starting months for economies with data for at least 50 listed firms before 1998m1. - Two-stage regression approach (Kumar, Xin, and Zhang (2019); asreg in STATA): - Stage 1 (rolling window estimation, 5 years / 60 months): - For each month V, regress firm excess returns on excess market return and temperature anomaly. - Temperature anomaly = current temperature − average temperature over past 30 years in the same month. - Equation (1) yields firm temperature sensitivity parameter 휃휃 푖푖,푐푐; key variable is absolute value 휃휃 푖푖,푐푐 푇푇. - Variables: 푟푟 푖푖,푐푐,푡푡 firm monthly equity return in US dollars; 푟푟 푓푓,푡푡 3-month U.S. treasury yield; 푟푟 푚푚푚푚푡푡,푐푐,푡푡 monthly market returns in economy c in U.S. dollars. Returns calculated using the Total Return Index. - Stage 2 (cross-sectional Fama-Macbeth style regressions for each month t): - Firms in each economy ranked by temperature sensitivity 휃휃 푖푖,푐푐 푇푇 and grouped into quintiles; top quintile = highest temperature sensitivity. - Dummy 퐻퐻푖푖 퐻퐻ℎ푇푇푇푇 푖푖,푐푐,푡푡 = 1 if firm is in top quintile in month t. - Regression (Equation (2)) for next-month excess return on the high-sensitivity dummy and controls. - Controls X include exposures to three Fama-French factors (excess market returns, SMB, HML). - Aggregate measure: 휆휆 ̂ 푐푐 = (∑_{t=1}^{240} 휆휆 푐푐,푡푡)/240 — average of estimated parameter over period 1998m1 to 2017m12. - Interpretation: if 휆휆 ̂ 푐푐 is significantly negative, stocks highly sensitive to temperature anomalies earn abnormal lower returns (mispricing). - Robustness checks for temperature-sensitivity pricing: - Add three additional standard risk factors: RMW (robustness minus weak), CMA (conservative minus aggressive), and MOM (momentum). - Include firm-level characteristics: lagged firm returns, firm market cap, firm book-to-market. - With all variables included, the pricing anomaly remains. - Note: t-statistics calculated using Newey-West adjusted standard errors with 3-month lags. ### Pricing of hurricane and storm risk in the catastrophe bond market (Online Annex 5.5) - Data: - 778 CAT bonds issued between 1997 and 2018 from Tomunen (2019); after exclusions (mortgage risk) final sample comprises 656 observations. - Bonds classified by exposure to hurricane and storm risk (including European Windstorms and Pacific Typhoons) or not. - Regression model (OLS, Equation (1)): - Dependent variable: 푇푇푇푇푟푟푇푇푎푎 퐻퐻 푔푔,푡푡 — spread over floating benchmark quoted of bond b issued on day t, measured in basis points. - Regressors include: - 퐸퐸퐸퐸 푇푇푇푇 퐸퐸푉푉푇푇퐻퐻 퐸퐸푉푉푠푠푠푠 푔푔,푡푡 — annualized expected loss (in percentage points) from underlying actuarial risk model. - 퐻퐻푉푉푟푟푟푟푖푖퐸퐸푎푎푛푛푇푇 퐸퐸푉푉푠푠푉푉푟푟 푇푇 푔푔,푡푡 — dummy: bonds exposed to hurricane and storm risk. - 푈푈푇푇퐸퐸퐸퐸푇푇푉푉푠푠푉푉푟푟푇푇 푔푔,푡푡 — dummy: risk in the U.S. - 푀푀푎푎푉푉푉푉푟푟푖푖푉푉푇푇 푔푔,푡푡 — maturity at issuance of the bond in years. - log(퐴퐴푇푇푉푉푉푉푛푛푉푉 퐴퐴푠푠푠푠푉푉푇푇 퐻퐻 푔푔,푡푡) — total face value in U.S. dollars of the bond. - 푇푇푟푟푖푖퐻퐻퐻퐻 푇푇푟푟푇푇푇푇푇푇 푇푇 푔푔,푡푡 — set of dummy variables describing trigger type: Parametric, Industry Loss Index, Modelled Loss, Indemnity, Multiple Triggers. - Year fixed effects 휇휇 푡푡. - Coefficient of interest: 훽훽 2 (effect of hurricane/storm exposure on bond spread). ### Long-Run Risk (LRR) model with climatic disasters (Online Annex 5.6) - Framework: - Builds on Bansal and Yaron (2004) LRR framework with Epstein and Zin (1989) recursive preferences and preference for early resolution of uncertainty. - Introduces temperature-induced very large climatic disasters affecting consumption growth (similar to Barro (2009), Wachter and others (2015), Bansal and others (2019)). - Temperature provides information about probability of future very large climatic disasters. - Representative agent preferences: - Epstein and Zin (1989) recursive utility specified (Equation (1)); parameters include 훿훿 (time preference), γ (risk aversion), 휓휓 (IES) with 휃휃 = (1−γ)/(1−1/ψ). - Budget constraint: - Wealth evolution W_{t+1} = (W_t − C_t) A_{C,t+1} (Equation (2)), where A_{C,t+1} is return on all invested wealth. - Consumption and dividends dynamics (Equation (3)): - ΔE_{t+1} and ΔH_{t+1} denote growth rates of consumption and dividends. - Components: - μ_c positive drift; E_t persistent component capturing long-run risks. - σ_t common time-varying volatility component for consumption and dividends. - D_{t+1} drop in consumption growth due to temperature-induced very large disasters. - Shocks η, T, w, V assumed i.i.d. normal and orthogonal. - Model solutions: - Log IMRS defined (Equation (4)). - Linear approximations (Campbell and Shiller (1988)) yield log returns for aggregate wealth portfolio r_{c,t+1} and market portfolio r_{m,t+1} (Equations (5) and (6)), depending on log price-consumption ratio H_t and log price-dividend ratio H_{m,t}. - Risk premium determined by covariation of return innovation with pricing kernel innovation (Equation (7)). - Unknowns solved by numerically solving a fixed-point problem for z. - Estimation and calibration: - Parameter vector Θ includes: {γ, ψ, δ, μ_c, ρ, φ_e, σ^2, v, σ_w, μ_d, φ_d, φ_d, h}. - Macro parameters estimated using an auxiliary model with generalized method of moments-type criterion via indirect inference; simulations of LRR-implied data required. - Draws from standard normal used for shock innovations; probability of climatic disasters mapped to current temperature level in simulations. - Time series length for simulations determined by number of observations, sampling frequency, investor decision frequency, and projected climate scenario length (available until 2100). - Bootstrap simulations used to overcome small sample bias. - Climatic scenarios considered: RCP 2.6, RCP 6.0, RCP 8.5. - Risk premia reported using average, 5th percentile and 95th percentile of temperature projections from CMIP5 ensemble for each scenario; compared to a counterfactual with no future climate risk and to market-implied risk premia from a standard discounted cash-flow valuation model. - Calibration details for disaster frequency and size provided in the Annex (local projections method, Equation (9)). - Calibration of very large climatic disasters (local projections specification, Equation (9)): - Dependent variable: log difference in real consumption per capita between t-1 and t+h. - Indicator ퟙퟙ{DisasVTr} = 1 if economy c was hit by a very large disaster in quarter t. - Controls X_{c,t} include net official development assistance and official aid received (percent of GDP). - Fixed effects: economy fixed effects θ_c and quarter fixed effects θ_t included. - Sample: quarterly data with h between 0 and 16 quarters. *Italic source: Online Annex — IMF Global Financial Stability Report: Climate Change: Physical Risk and Equity Prices (April 2020) — onlineannex51.pdf* ### 1975. The impact of disasters is calibrated based on the estimates for very large disasters reported (훽훽 ### 1975. The impact of disasters is calibrated based on the estimates for very large disasters reported (훽훽 ### Impact of very large disasters on consumption — estimation and calibration - Estimated coefficients measure the impact of disasters on real per capita consumption using local projections (Jorda, 2005). - Indicator dummy equals one if a very large disaster occurred at time t. - A disaster is defined as very large if the ratio of affected population to total population, the ratio of deaths to total population, or the damage-to-GDP ratio, are above the 95th percentile of the historical distribution. - Blue dots in the figure indicate coefficients statistically significant at the 10 percent level or lower. - The impact is reported as deviation from a no-disaster scenario (percent). ### Market-implied equity risk premiums — method - Implied methods derived from the standard dividend discount model (DDM) by inverting discounted cash-flow valuation formulas. - Market implied equity risk premium computed as: - the difference between the 18 months ahead forecasts of earnings per share-to-price ratio and the 10-year government bond yield (inflation adjusted). - The measure is then averaged over the last 15 years. ### Calibration of the likelihood of climatic disasters — model specification - The relationship between temperature and occurrence of disasters is calibrated using a panel logit model: - Pr(DisastersVTc,t = 1) = Φ(f(Tc,t) + f(Pc,t) + θc + εc,t) (equation (10)) - where: - c is economy, t is a quarter. - f(Tc,t) is a quadratic function of temperature. - f(Pc,t) is a quadratic function of precipitations. - θc are economy fixed effects. - εc,t is the error term. - The model is also tested for each disaster type separately. - Disaster probability calibration uses only wildfires, droughts and heat waves events to define the category “T-Induced”. - Estimates are robust to different fixed effects structures and to inclusion of economy specific time trends. - Within model simulation, logit-implied probabilities are scaled by the number of very large T-induced disasters over the total number of T-induced disasters observed in the EM-DAT database to match prior disaster-size calibration. - The (marginal) implied quarterly probability of all T-Induced disasters at different temperature levels is reported in Online Annex Figure 5.6.2. ### Model-implied quarterly probability of T-Induced disasters — notes on the figure - The figure reports the marginal quarterly probability of all temperature induced (“T-induced”) disasters for different levels of temperature. - T-induced disasters refer to Heat Waves, Wildfires and Droughts. - Vertical lines delineate the 95 percent confidence bounds around the point estimates. - Temperature is weighted by population in 2000. - The figure axis scale includes probabilities ranging from 0 to 0.7 and temperature values shown across a range including 0, 4, 8, 12, 16, 20, 24, 28 (as indicated in the figure context). ### The pricing of physical risk into sovereign bonds — baseline specification - Baseline regression specification (equation (1)): - BVnH TTrTaH aV isssVaanE T g,c,t = αc,t + β1 CViTaVT c + β2 VVVnH VTrT g,c,t + β3 CViTaVT c X VVVnH VTrT g,c,t + β4 CVnVrVVs g,c,t + εg,c,t - where: - b is bond, c is economy, t is year. - BVnH TTrTaH aV isssVaanE T g,c,t is the spread between the coupon rate of the security and the US Treasury or another Benchmark bond, measured in basis points. - Long-term is a dummy for bonds with maturities after 2040. - Control variables: log of issue size, log of maturity, VIX, and sovereign credit rating. - αc,t are economy-year fixed effects. - There is no perfect single measure of climate change physical risk available; analysis uses seven measures (CViTaVT T c), one of which is included per regression: 1. projected change in the number of extreme heat days 2. projected change in the number of extreme precipitation days 3. projected change in heatwave likelihood 4. projected change in drought likelihood 5. Sea Level Rise Index 6. Climate Change Hazard Index 7. Climate Change Physical Risk Index - The coefficient of interest is β3, the interaction coefficient. ### Projection horizons, scenario choice, and sample - Projected changes in climate hazard variables (variables 1–4) are available for four different projection horizons up to the year 2100, and for four different emission scenarios. - Analysis concentrates on the horizon 2020–39 and emission scenario RCP 8.5; results are robust to other horizons/scenarios. - Sample includes 41,211 bonds, issued in 121 economies. - 94 percent of the bonds are issued by advanced economies. - 54 percent of the bonds are issued by the United States alone. - Results are robust to excluding the United States from the sample and to excluding economies not included in the analysis discussed in the main text of the chapter. *Source: Online Annex 5.6–5.7, GLOBAL FINANCIAL STABILITY REPORT—Climate Change: Physical Risk and Equity Prices, International Monetary Fund | April 2020* --- _Source: https://www.imf.org/-/media/files/publications/gfsr/2020/april/english/onlineannex51.pdf_