## wpiea2023053-print-pdf ## Source details **Canonical URL:** [wpiea2023053-print-pdf](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023053-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023053-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023053-print-pdf.pdf.json) --- ### Sector-specific impacts of weather shocks - Agriculture is the most harmed sector by various weather shocks. - Abnormal temperature realizations (country-specific daily distribution): - An additional day above the 95th percentile of the daily temperature distribution reduces agricultural growth rate by 0.03 percentage points (11% of its sample mean). - Dryness (SPEI): - Droughts and dryness conditions have a negative and substantial effect on agricultural production. - Differential sectoral responses: - Drier conditions marginally benefit production in construction and in transport, storage and communication. - Tropical cyclones (Kunze (2021) wind-speed metric) harm agriculture and the “other activities” sector (government and financial sector). ### Channels of impact and network propagation - Direct channels of sectoral output losses: - Direct input effects (weather as input in crop production). - Reductions in labor supply and productivity. - Changes in total factor productivity. - Damages to assets and infrastructure. - Indirect channels via production networks: - Domestic shocks (within-country, weighted by intersectoral linkages) have strong negative effects on several sectors, notably construction; transport, storage and communication; wholesale, retail trade, restaurants and hotels. - Magnitude of domestic-shock effects is comparable to the direct effect on agricultural production. - Persistence: domestic shocks dampen sectoral growth up to ten periods after the shock. - Directional propagation by supply-chain position: - Agriculture (early-stage) is harmed by local shocks and by shocks in customer sectors that propagate upstream (demand shocks). - Later-stage sectors (wholesale, retail trade, restaurants and hotels; other activities) are negatively affected by shocks hitting supplier sectors and propagating downstream (supply-side productivity shocks). ### Quantified economic cost and importance of network channel - Recent warming (2000 onwards) accounting for trade interlinkages: - Heat shocks are responsible for an average annual output loss of 0.3%. - Sector-aggregation results (2001–2020 vs 1970–2000 counterfactual): - Considering only sector-specific local shocks: -0.02% average annual relative GVA per capita loss ( -0.04% median, IQR [-0.09,0.02]). - Accounting for indirect heat shocks (propagation through partners): 0.28% average annual relative GVA per capita loss (0.24% median, IQR [0.04,0.44]). - National aggregation (using each country’s average sectoral breakdown 2001–2020): - Accounting for indirect heat shocks: 0.3% mean, 0.26% median, IQR [0.16,0.37]. - Omitting heat shock propagation: -0.02% mean, -0.01% median, IQR [-0.04,0.05]. ### Data sources, measures and coverage - Sectoral production: - Source: National Accounts Main Aggregates (UNSD, 2022). - Coverage: 1970 through 2020; 205 countries (GVA in constant 2015 USD). - Six sector groups (ISIC revision 3.1): A&B; C-E; F; G–H; I; J–P. - Weather data: - ERA-5 (ECMWF) 0.25°×0.25° grid, hourly → daily. - Temperature: “2m Temperature” converted to °C. Precipitation: “Total precipitation” in meters. - Aggregation: population weights (Landscan 2000) for non-agriculture; cropland fraction (Global Agricultural Lands) for agriculture. - Dryness/wetness: - SPEI-12 (12-month time scale), grid 0.5°×0.5°. - Two measures: weighted average of negative monthly SPEI values; share of grid-months with SPEI ≤ -2 / SPEI ≥ 2 (yearly maximum-share month). - Tropical cyclones: - Kunze (2021) wind-speed based metric (non-linear including cube of wind speed above 92 km/h threshold; accounts for maximum sustained wind speed, forward speed, distance from storm center, radius of maximum wind). - Sectoral interlinkages: - IO data: EORA26 (26 sectors, 189 countries, 1990–2021); retain 1990 IO matrix year and aggregate to six sectors. ### Construction of network shocks (definitions and weights) - Types by geographic origin and supply-chain direction: - Domestic vs foreign; downstream vs upstream. - Four network shock definitions for sector s in country c at time t: - ShockDnD_{s,c,t} = Σ_{j≠s} ω_{s,c,j,c} ShockOwn_{j,c,t} - ShockUpD_{s,c,t} = Σ_{j≠s} bω_{s,c,j,c} ShockOwn_{j,c,t} - ShockDnF_{s,c,t} = Σ_{j} Σ_{k≠c} ω_{s,c,j,k} ShockOwn_{j,k,t} - ShockUpF_{s,c,t} = Σ_{j} Σ_{k≠c} bω_{s,c,j,k} ShockOwn_{j,k,t} - ShockOwn_{j,k,t} is a weather shock hitting sector j in country k in year t. - Weight formulas: - Downstream weights: ω_{s,c,j,k} = input_{jk→sc} / Σ_{lf∈Θ_{sc}} input_{lf→sc} - Upstream weights: bω_{s,c,j,k} = input_{sc→jk} / Σ_{lf∈bΘ_{sc}} input_{sc→lf} - Practical notes: - All six aggregated sectors included in both upstream and downstream given aggregation. - First-step analysis uses an unweighted average of upstream and downstream weights to construct network shocks. - Exception: agricultural weather shocks weighted by agricultural land coverage; other sectors in same country share same weather shock. ### Empirical approach and identification - Two-step strategy: 1. Estimate sector-specific response in per capita GVA growth to weather shocks using pooled sample across 184 countries over 50 years. 2. Analyze how weather shocks hitting customer/supplier sectors domestically and abroad affect sectoral production. - Baseline (first-differenced) specification: - ∆ log(GVA)_{sct} = f_s(∆W_{ct}) + α_{cs} + λ_{st} + ε_{sct} - Controls: country-sector FE (α_{cs}), sector-year FE (λ_{st}); standard errors clustered at country level. - Propagation specification: - ∆ log(GVA)_{sct} = γ_s Shock_Own_{sct} + Σ_J γ_J_s Shock_J_{sct} + α_{cs} + λ_{st} + η_{sct} - J disaggregated: {domestic downstream; domestic upstream; foreign downstream; foreign upstream}. - Identification: plausibly exogenous within-country year-to-year weather variation; first-differenced weather variables to avoid trends. ### Key empirical findings (selected, exact values preserved) - Sample and summary statistics: - Number of countries: 184 - Number of sectors: 6 - Number of years per country-sector: 47.926.271250 - Log GVA per capita: N 50,223; mean 6.147; SD 1.796; min -2.880; max 11.534 - GVA per capita growth rate: N 50,223; mean 0.016; SD 0.120; min -3.299; max 2.572 - Temperature above 95th percentile (days/year): N 8,572; mean 18.986; SD 16.501; min 0; max 152 - Temperature binary indicator results: - Agriculture: if daily average temperature is larger than in the previous year, agricultural GVA growth rate decreases by 0.7 percentage points (translates into a 284% decrease with respect to the sample average 0.002). - Agriculture benefits from more precipitation. - Transport, storage and communication marginally benefits from hotter and drier conditions (15% increase of sample mean for hotter; 17% increase of sample mean for drier). - Continuous-variable results: - A 0.01 ◦C daily increase with respect to the previous year is associated with a decrease in agricultural per capita growth rate by 3% of the sample mean. - Other sectors generally respond positively to increases in average daily temperatures (other activities; transport, storage and communication; wholesale, retail trade, restaurants and hotel), though some estimates are imprecise. - Abnormal temperature realizations (percentile-based): - Agriculture: an additional day above the 95th percentile reduces agricultural growth rate by 0.03 percentage points (11% of sample mean); an additional day below the 5th percentile reduces agricultural growth rate by 10% of sample mean. - Cold and hot shocks have similar magnitudes for agriculture. - Droughts (SPEI): - Agriculture is directly harmed by drought shocks: a 1 SD increase in dryness conditions associates with a 0.09 percentage point reduction in agricultural growth rate (sample mean is 0.002). - Own drought shocks strongly benefit production in construction; mining, manufacturing and utilities; transport, storage and communication. - Domestic shocks strongly harm mining, manufacturing and utilities. - Foreign shocks have sizable negative effects on other activities and wholesale, retail trade, restaurants and hotels. - Tropical cyclones: - Cyclone intensity damages agriculture and other activities directly. - Domestic cyclone shocks strongly negatively affect transport, storage and communication growth rate with magnitude comparable or larger than direct effect on other activities. - Foreign cyclone shocks null for most sectors; positive for wholesale, retail trade, restaurants and hotel. ### Network vs direct effects for abnormally hot shocks - Standardized-coefficient findings: - Only agriculture is directly harmed by abnormally hot temperatures. - Sectors are negatively harmed through domestic network shocks; domestic network effects on other sectors are comparable in magnitude to the direct damage on agriculture. - Foreign shocks have negative coefficients across most sectors but are never statistically significant. - Upstream/downstream decomposition: - Agriculture: suffers mostly from upstream shocks (domestic and foreign); benefits from downstream shocks (interpretable as demand-side propagation). - Later-stage sectors (other activities; wholesale, retail trade, restaurants and hotel): growth harmed by abnormally hot shocks in supplier sectors (downstream propagation); foreign upstream shocks can have positive effects (competition/price effects). - Persistence (local projections, horizon up to h = 10): - Domestic heat shocks have persistent effects on total value-added levels; coefficients small and noisy. - Several sectors do not recover from domestic partner abnormally hot shocks with strong negative growth effects in construction; mining, manufacturing and utilities; wholesale, retail trade, restaurants and hotels lasting multiple years. - Agriculture: direct shocks harm persistently but effect lasts only one year and dissipates thereafter. - Foreign shocks: small, positive, not distinguishable from zero at most lags; positive significant effect starting five years after shock at country level suggests competition/price effects. ### Robustness checks and spatial concerns - Spatial correlation: - Adding subregion-by-year and continent-by-year fixed effects: network effects persist, indicating trade transmission rather than spatially correlated weather shocks. - Additional robustness: - Accounting for incomplete exposure shares (interacting sector-year FE with sum of exposure shares) — effects robust. - Balanced panel estimation — results robust. - Excluding large countries (Brazil, China, India, Russia, US) — results robust. - Starting analysis from 1990 with predetermined IO — results robust. ### Simulation methodology for economic cost of recent warming (formulas preserved) - Country trend estimation (1970–2000): - T_ct = α_c + β_c t + ε_ct - Use estimated ˆβ_c to construct counterfactual ˜T95_sct. - Annual direct and spillover growth effects: - g_direct_sct = b γ_s (T95_sct − ˜T95_sct) (Equation (10)) - g_spillover_sct = (b γ_s T95_sct + c γ_D_s T95,D_sct + c γ_F_s T95,F_sct) − (b γ_s ˜T95_sct + c γ_D_s ˜T95,D_sct + c γ_F_s ˜T95,F_sct) (Equation (11)) - Counterfactual sectoral log GVA per capita evolution: - ˆY_p_sct = Y_sct−1 + y_sct + g_p_sct (Equation (12)) - Average relative loss in GVA (2001–2020): - %LOSS_p_sc = (1/T) Σ_{t=2001}^{2020} ( e^{ˆY_p_sct} − e^{Y_sct} ) / e^{Y_sct} (Equation (13)) - Country aggregation: - %LOSS_p_c = Σ_s λ_sc %LOSS_p_sc (Equation (14)) - Estimation details: - Sector-specific semi-elasticities estimated with 1000 bootstrap replications with replacement. ### Representative summary statistics and regression excerpts (selected exact entries) - Panel coverage: total 51,129 year-sectors in final sample (Table A1). - Weather shock summary (Table A4 highlights): - Temperature above 90th percentile (days/year): N 8,548; mean 37.487; sd 23.6100; min 0; max 222 - Temperature above 99th percentile (days/year): N 8,548; mean 3.851; sd 6.1450; min 0; max 94 - Negative temperature anomaly (◦C): N 7,817; mean 3.274; sd 3.353; min 0.00005; max 31.413 - Cubic wind speed (∆km3/h3) (tropical cyclones): N 7,371; mean 659.399; SD 672,647.087; min -17,838,080; max 17,856,152 - Unit-root tests (Table A5): - GVA growth rate: Statistic -6.072; p-value 0.000 - Abnormally hot temperature shock (p95): Statistic -6.258; p-value 0.000 - Selected coefficients (Table A6 / A7): - Temperature effect on Agriculture (binary/annual changes): -0.00676 ∗∗ (0.00297) and alternative -0.0351 ∗∗ (0.0144) - Precipitation effect on Agriculture: 0.0117 ∗∗∗ (0.00291) and alternative 0.0405 ∗∗∗ (0.0114) - Drought (SPEI) on Agriculture (standardized, Table A10): -0.119 ∗∗∗ (0.0197) - Tropical cyclones on Agriculture (Table A11): -0.0288 ∗∗ (0.0125) ### Limitations and avenues for future research - Adaptation: - Analysis provides modest evidence on adaptation but does not explicitly model adaptive investments, technological change, or fiscal policy responses. - Beliefs and expectations: - Study is silent on decision-makers’ climate beliefs and expectation formation; heterogeneous perceptions could affect adaptation and damages. - Spatial resolution: - Country-level aggregation may mask local variation; higher spatial resolution could change precipitation and local-impact estimates. - Input specificity and firm-level channels: - Current propagation via IO averages omits input specificity, search frictions, supplier concentration, and elasticity of substitution; firm-level integration is an open avenue. - Time-varying networks and reallocation: - Analysis uses exogenous time-invariant trade network; reallocation and adjustments in trade patterns as adaptation remain to be studied. *Source: wpiea2023053-print-pdf (excerpt from Working Paper No. WP/2023/053).* ### 2020. To identify sector-specific weather shocks, I exploit plausibly exogenous variations ### wpiea2023053-print-pdf - 2020. To identify sector-specific weather shocks, I exploit plausibly exogenous variations ### Sector-specific impacts of weather shocks - Agriculture is the most harmed sector by various weather shocks (in line with Acevedo et al., 2020; Dell et al., 2012). - Using a measure of abnormal weather realizations with respect to the country-specific daily distribution: - An additional day above the 95th percentile of the daily temperature distribution reduces agricultural growth rate by 11% of its sample mean. - Dryness measured by the Standardized Precipitation Evapotranspiration Index (SPEI) shows a negative and substantial effect of droughts and dryness conditions on agricultural production. - Drier conditions marginally benefit production in: - Construction. - Transport, storage and communication (tasks in outdoor “interface” areas are sensitive to abundance in precipitation). - Tropical cyclones (measured via wind speed per Kunze (2021)) harm: - Agriculture. - The “other activities” sector, comprising government and financial sector, reflecting destructive power beyond agriculture. ### Channels of impact across sectors - Sectoral output losses from climate change can occur via: - Direct input effects (weather as an input in crop production). - Reductions in labor supply and productivity. - Changes in total factor productivity. - Damages to assets and infrastructure. - Indirect effects through trade partners and production networks: - Domestic shocks (weather shocks within the same country weighted by intersectoral linkages) have strong negative effects on several sectors, notably construction; transport, storage and communication; wholesale, retail trade, restaurants and hotels. - The magnitude of domestic-shock effects is comparable to the direct effect on agricultural production. - Local projections show persistence: domestic shocks dampen sectoral growth up to ten periods after the shock. - Directional propagation depending on supply-chain position: - Agriculture (early-stage sector) is harmed by local shocks and also negatively affected by shocks in customer sectors that propagate upstream (interpretable as demand shocks). - Later-stage sectors (wholesale, retail trade, restaurants and hotels, and other activities) are negatively affected by shocks that hit supplier sectors and propagate downstream (supply-side productivity shocks), creating powerful downstream propagation. ### Quantified economic cost and importance of network channel - When computing the economic cost of recent warming from 2000 onwards and accounting for trade interlinkages: - Heat shocks are responsible for an average annual output loss of 0.3%. - The inclusion of sectoral linkages reveals a substantial underestimation of damages in estimates that omit these linkages, underlining the importance of propagation through production networks as a component of the total economic impact of climate change. ### Data sources and scope - Sectoral production data: - Source: National Accounts Main Aggregates (UNSD, 2022). - Coverage: 1970 through 2020 and for 205 countries (GVA in constant 2015 USD). - Six broad sector groups (ISIC revision number 3.1): agriculture, hunting, forestry, and fishing (A&B); mining, manufacturing and utilities (C-E); construction (F); wholesale, retail trade, restaurants, and hotels (G–H); transport, storage, and communication (I); other activities (J–P). - Weather data: - Temperature and precipitation from ERA-5 (ECMWF) on a 0.25°×0.25° resolution grid (≈28km at the Equator), original temporal frequency hourly aggregated to daily. - Temperature used: “2m Temperature” converted from Kelvin to °C. Precipitation: “Total precipitation” measured in meters. - Aggregation of grid-cell weather to country level uses time-invariant population weights from the 2000 Landscan dataset for non-agricultural sectors and cropland fraction (Global Agricultural Lands dataset) for the agricultural sector. - Dryness/wetness: - Measure: Standardized Precipitation Evapotranspiration Index (SPEI). - Version used: SPEI-12 (12-month time scale). - Grid resolution: 0.5°×0.5° (around 56 km at the Equator). - Two constructed measures: - Weighted average of negative monthly SPEI values to obtain average annual dryness relative to historical conditions. - Share of total grid-months subject to extreme droughts (SPEI ≤ -2) and extreme wetness (SPEI ≥ 2), using the share in the month where the share is at its maximum for each year. - Tropical cyclones: - Measure: Kunze (2021) wind-speed based metric. - Annual measure is a non-linear function including the cube of wind speed when wind speed is above a 92 km/h threshold; wind speed accounts for maximum sustained wind speed, forward speed, distance from storm center, and radius of maximum wind. - Sectoral interlinkages: - IO data: EORA26 (Kanemoto et al., 2011; Lenzen et al., 2012). - Coverage: 26 sectors for 189 countries from 1990 to 2021. - Use: retain first available IO matrix year (1990) and aggregate 26 sectors to the six-sector production classification to match sectoral GVA data. ### Construction of network shocks (definitions and weights) - Network shocks distinguished by geographic origin and supply-chain direction: - Geographic: domestic (same country) vs foreign (other countries). - Supply-chain: downstream (shocks originating in supplier sectors and propagating downstream) vs upstream (shocks originating in customer sectors and propagating upstream). - Four types of network shocks for sector s in country c at time t: - ShockDnD_{s,c,t} = Σ_{j≠s} ω_{s,c,j,c} ShockOwn_{j,c,t} - ShockUpD_{s,c,t} = Σ_{j≠s} bω_{s,c,j,c} ShockOwn_{j,c,t} - ShockDnF_{s,c,t} = Σ_{j} Σ_{k≠c} ω_{s,c,j,k} ShockOwn_{j,k,t} - ShockUpF_{s,c,t} = Σ_{j} Σ_{k≠c} bω_{s,c,j,k} ShockOwn_{j,k,t} - Where ShockOwn_{j,k,t} is a weather shock hitting sector j in country k in year t. - Weighting schemes: - Downstream weights (ω_{s,c,j,k}) = input_{jk→sc} / Σ_{lf∈Θ_{sc}} input_{lf→sc} - Ratio of inputs supplied by sector j (in country k) to sector s (in country c) over total inputs that sector s sources from its set of supplier sector-countries Θ_{sc} — captures relative importance of each supplier. - Upstream weights (bω_{s,c,j,k}) = input_{sc→jk} / Σ_{lf∈bΘ_{sc}} input_{sc→lf} - Ratio of inputs of sector s to each sector j over total inputs supplied by sector s to all sector-countries l in its set of customers bΘ_{sc} — captures importance of each customer. - Practical considerations: - All six aggregated sectors are included in both upstream and downstream weights given the aggregation level. - As a first step, the analysis considers network shocks based only on geographic location by taking an unweighted average of upstream and downstream weights. - Exception: weather shocks in the agricultural sector are obtained by weighting weather variables by agricultural land coverage; all other sectors in the same country share the same weather shock. ### Contributions and implications - Empirical contributions: - Provides jointly estimated sector-specific response functions to temperature and precipitation anomalies, dryness conditions, and cyclone intensity with global coverage of a six-sector economic production from 1970 through 2020. - Introduces and quantifies propagation of weather shocks through production networks across countries, distinguishing upstream vs downstream and domestic vs foreign channels. - Broader implications: - Demonstrates that accounting for sectoral interlinkages materially increases estimated economic damages from heat shocks (average annual output loss of 0.3% from 2000 onwards), indicating that estimates omitting linkages can substantially underestimate climate impacts. - Suggests an important channel for amplifying climate damages relevant for correct quantification of economic damages and computation of the social cost of carbon. *Source: wpiea2023053-print-pdf (excerpt).* ### 2.4 Summary Statistics ### 2.4 Summary Statistics ### Dataset coverage and sectoral production summary (Table 1) - Panel covers all countries in the world for most of the fifty available years; on average, information for each sector-country tuple is reported for 48 years. - Number of countries: 184 - Number of sectors: 6 - Number of years per country-sector: 47.926.271250 - Summary statistics (N mean SD min max): - Log GVA per capita: 50,223 6.147 1.796 -2.880 11.534 - GVA per capita growth rate: 50,223 0.016 0.120 -3.299 2.572 - io Downstream (ω): 1,284,822 0.00083 0.01429 2.68e-11 0.99976 - io Upstream (bω): 1,284,822 0.00084 0.01432 8.50e-11 0.9944 - io Total (ω): 1,284,822 0.00083 0.01132 1.65e-10 0.73665 - Sectoral GVA per capita (N mean SD min max): - Agriculture, hunting, forestry, fishing (ISIC A-B): 7,472 0.003 0.104 -1.691 0.745 - Mining, Manufacturing, Utilities (ISIC C-E): 8,572 0.012 0.125 -3.169 2.430 - Construction (ISIC F): 8,566 0.015 0.167 -3.299 2.572 - Wholesale, retail trade, restaurants and hotels (ISIC G-H): 8,522 0.016 0.108 -1.639 1.502 - Transport, storage and communication (ISIC I): 8,519 0.027 0.109 -2.514 2.030 - Other Activities (ISIC J-P): 8,572 0.019 0.085 -1.513 1.261 - Note: Most sectors are covered for the whole time period except for recent geopolitical changes. ### Weather shock measures and summary (Table 2) - Summary statistics computed using country-year observations using only population-weighted weather shocks. - Temperature and precipitation (N mean SD min max): - Positive difference in daily temperature sum {0;1}: 8,572 0.524 0.49901 - Positive difference in daily precipitation sum {0;1}: 8,572 0.497 0.50001 - Changes in daily temperature sum (∆ ◦C): 8,572 9.556197.755 -1594.597 1704.612 - Changes in daily precipitation sum (∆m): 8,572 0.0008 0.010 -0.092 0.095 - Temperature above 95th percentile (days/year): 8,572 18.986 16.501 0 152 - Temperature below 5th percentile (days/year): 8,572 17.870 14.185 0 156 - Precipitation above 95th percentile (days/year): 8,572 18.244 6.613 1 78 - Precipitation below 5th percentile (days/year): 8,572 15.633 10.182 0 86 - Dryness and wetness: - Average dryness (∆ SPEI <0): 6,204 -0.009 0.496 -2.181 1.924 - Extreme drought prevalence (∆share grid-months SPEI<−2): 8,448 0.002 0.252 -11 - Extreme wetness prevalence (∆share grid-months SPEI>2): 8,448 -0.001 0.195 -11 - Tropical cyclones: - Cubic wind speed (∆km3/h3): 7,371 659.399 672,647.087 -17,838,080 17,856,152 - Notes: Where ∆ is indicated, variables are in first-difference, measuring changes in weather conditions from the previous year. ### Empirical approach (Sections 3, 3.1, 3.2) - Two-step empirical strategy: 1. Estimate sector-specific response in per capita GVA growth rate to weather shocks using a pooled sample across 184 countries over 50 years. 2. Analyze how weather shocks hitting customer/supplier sectors domestically and abroad affect sectoral economic production. - Baseline specification (first-differenced growth regression, Equation (7)): - ∆ log(GVA)_{sct} = f_s(∆W_{ct}) + α_{cs} + λ_{st} + ε_{sct} - Controls: country-sector fixed effects (α_{cs}) and sector-year fixed effects (λ_{st}); standard errors clustered at the country level. - First-differenced, stationary weather variables are used to avoid reintroducing trends. - Country-specific linear (and quadratic) time trends and lagged dependent variables are tested in robustness checks. - Propagation specification (Equation (8)): - ∆ log(GVA)_{sct} = γ_s Shock_Own_{sct} + Σ_J γ_J_s Shock_J_{sct} + α_{cs} + λ_{st} + η_{sct} - Shocks J disaggregated by geographic location and supply chain position: J ∈ {domestic; foreign} and further J ∈ {domestic downstream; domestic upstream; foreign downstream; foreign upstream}. - Purpose: quantify trade-induced exposure to weather shocks and the role of network propagation channels. - Identification relies on plausibly exogenous within-country variation in year-to-year weather changes; omission of spatial weather considerations could bias estimates due to violations of SUTVA. ### Key empirical findings (Section 4 and 4.1) - Method: pooled, multi-country, sector-specific response functions estimated for 6 sectors to changes in annual temperature and precipitation distribution. - Binary indicator results (positive vs negative annual changes): - Agriculture: if daily average temperature is larger than in the previous year, agricultural GVA growth rate decreases by 0.7 percentage points (point estimates reported in Table A6). This translates into a 284% decrease with respect to the sample average (0.002). - Agriculture benefits from more precipitation. - Transport, storage and communication marginally benefits from hotter and drier conditions (15% increase of sample mean for hotter; 17% increase of sample mean for drier). - Standardized continuous-variable results: - A 0.01 ◦C daily increase with respect to the previous year’s temperature (around 30% of the sample mean) is associated with a decrease in agricultural per capita growth rate by 3% of the sample mean. - Other sectors generally respond positively to increases in average daily temperatures (other activities; transport, storage and communication; wholesale, retail trade, restaurants and hotel), though some estimates are imprecise. - Production in other sectors generally does not respond to changes in precipitations, except transportation benefits from drier conditions. - Heterogeneity by income and climate (adaptation potential): - Income groups: interaction with World Economic Outlook income groups (advanced economies; emerging market economies; low-income developing countries). - Advanced economies are not harmed by increases in temperature distribution; agricultural production is sheltered in advanced economies (coefficient positive and statistically significant). - Construction; mining, manufacturing, utilities; transport, storage and communication; wholesale, retail trade, restaurants and hotels also benefit from temperature increases among the richest countries. - The coefficient of temperature increases on agriculture remains strongly negative for emerging market economies and low-income developing countries. - Low-income developing countries’ estimates are always smaller in magnitude than for emerging market economies. - Climate terciles (cold, temperate, hot average long-run temperature): - Agriculture experiences persistent negative effects of temperature increases across climates (smaller in absolute magnitude for cold countries; imprecise for hot countries). - Increases in temperature harm other sectors in hot climate countries (construction; other activities; wholesale, retail trade, restaurants and hotels). - In cold climate countries, temperature increases benefit production in industrial and services sectors (construction; mining, manufacturing, utilities; other activities; wholesale, retail trade, restaurants, hotel). *Source: wpiea2023053-print-pdf - 2.4 Summary Statistics* ### 4.2 Abnormal weather realizations ### 4.2 Abnormal weather realizations ### Methodology and definitions - Abnormal weather realizations are defined as the number of days in a year that belong to the pth-percentile of the country-specific temperature and precipitation daily distribution over the fifty-year period, where p ∈ {1; 5; 10; 90; 95; 99}. - This location-specific, time-invariant threshold approach: - Compares any abnormal realization to the country-specific normal distribution. - Accounts for long-run adaptation by using a fifty-year climate norm as baseline. - Interprets bottom percentiles as abnormally cold or dry and top percentiles as abnormally hot or wet. - Identification: estimate the impact of increases in the number of abnormally cold and hot, dry and wet days in a pooled multi-sector stacked regression, using days in the rest of the distribution as the baseline category. - Robustness checks: - Definitions of “abnormal” varied using the 1st/99th, 5th/95th, and 10th/90th percentiles produce similar results. - Results are robust to estimating the baseline equation in a balanced panel and to excluding large countries (Brazil, China, India, Russia, US). ### Key empirical findings — temperature shocks - Agriculture: - An additional day above the 95th percentile of the daily temperature distribution reduces the agricultural growth rate by 0.03 percentage points (11% of its sample mean). - An additional day below the 5th percentile reduces the agricultural growth rate by 10% of its sample mean. - Hot and cold shocks have very similar magnitudes for agriculture, providing little evidence of asymmetry. - Other sectors: - Most sectors show no statistically significant response to hot or cold temperature shocks. - Construction: - Growth rate decreases by 1.2% of sample mean for increases in temperature realizations above the 95th percentile. - Growth rate increases by 1.8% of sample mean from cold temperature shocks. - Interpretation: results confirm prior literature that agriculture is most harmed by heat shocks; cold shocks also harm crops that cannot grow below certain temperatures. ### Key empirical findings — precipitation shocks - Wet precipitation shocks: - Generally do not affect sectoral production. - Exception: an additional day of precipitation above the 95th percentile has a positive effect on agricultural production. - Potential explanations for null/noisy precipitation results: - Excessive and insufficient precipitation may not be adequate indicators of water availability. - High spatial variation of precipitation and country-level aggregation may mask meaningful local variation. - Data limitations: - Lack of globally comparable, fine-grained sectoral production data prevents overcoming aggregation concerns. ### Time-varying climate norms and adaptation - Construct time-varying country-specific distributions over the preceding m years for each t, where m ∈ {20; 30; 40}. - Weather data exploited from 1950; official World Meteorological Organization norm corresponds to thirty years. - To compare different time-varying norms, sectoral production data from 1990 onward are used. - Findings: - Results using time-varying historical norms are very similar to baseline estimates: agriculture is negatively affected by hot temperature shocks. - Assuming different speeds of change for the historical climate distribution (20-, 30- or 40-year) does not significantly alter point estimates. - The negative effect of heat shocks on agricultural production is persistent, suggesting adaptation has not entirely offset climate damages. - Some suggestive, small degree of adaptation in other sectors (transport, storage and communication; mining, manufacturing and utilities; wholesale, retail trade, restaurants and hotel) where output losses are mitigated or converted into gains for faster time-varying climate norms. ### Extensions to physical extreme-event measures (preview of Section 4.3) - The section motivates moving from percentile-based abnormal days to physical measures of extremes (droughts, cyclones) to better capture water availability and event intensity; results summarized further in Section 4.3. *Italic: Source — wpiea2023053-print-pdf, Section 4.2 Abnormal weather realizations* ### 5.1 Abnormally hot temperature shocks ### 5.1 Abnormally hot temperature shocks ### Measurement and empirical approach - Heat shocks measured as the number of days above the 95th percentile of the country-specific temperature daily distribution. - Main estimation extends Equation (8) to include an average of heat shocks in domestic trade partners weighted by input-output interlinkages (unweighted average of upstream and downstream interlinkages). - Rationale: direct sectoral impacts omit amplification and transmission through intersectoral reliance; network effects may turn negligible direct effects into important indirect impacts. ### Domestic shocks: direct vs. network effects - All coefficients are standardized. - Key findings: - Sectors are negatively harmed by abnormally hot temperatures through network shocks. - The only sector directly harmed by weather shocks is agriculture. - Comparing direct shocks estimates with estimates without network shocks: agriculture direct effect is negative and slightly smaller in magnitude in absolute value (-0.036 instead of -0.039). - Coefficients on other sectors (direct shocks) become positive but not statistically different from zero when accounting for network shocks. - Implications: - Sector-specific estimates that account only for direct impacts may be downward biased because they omit propagated shocks. - Accounting for input-output interlinkages shows weather shocks are amplified across the economy and indirectly affect other sectors. - The effect of domestic network shocks on other sectors is comparable in magnitude with the direct damage estimated on agriculture, implying prior economy-wide estimates have been largely underestimated. ### Foreign shocks and decomposition into domestic vs. foreign - Foreign shocks constructed as the average of weather shocks in foreign sector-countries weighted by input-output interlinkages. - Findings (Figure 4 referenced): - Domestic shocks have a negative and sizable effect on economic production across all sectors. - Domestic coefficients are larger in absolute value (though imprecisely estimated for agriculture and mining, manufacturing and utilities). - Foreign shocks have a negative effect across all sectors except construction, but never statistically significant (large imprecision). - Including foreign shocks corrects for upward bias in previous estimations and contradicts the hypothesis that geographically distant weather shocks are idiosyncratic and orthogonal to local ones. - Endogenous selection of trade partners by firms based on location/exposure would bias results against finding any effect, so it is not a threat to identification. ### Robustness to spatial correlation and sample composition - Spatial correlation concern addressed by adding subregion-by-year and continent-by-year fixed effects (subregions defined into 17 zones). - This isolates weather variation local to the unit and uncorrelated within subregion/continent; network effects persist, indicating transmission via trade interlinkages rather than spatially correlated shocks. - Strong negative domestic network effects robust in construction; other activities; transport, storage and communication; wholesale, retail trade, restaurants and hotel. - Additional robustness checks: - Interact period fixed effects with the sum of exposure shares to account for incomplete trade exposure shares (Borusyak et al., 2022) — effects robust (Figure A14a). - Balanced panel estimation — results robust (Figure A14b). - Excluding large countries (Brazil, China, India, Russia, US) — results robust (Figure A14c). - Starting analysis from 1990 to allow strongly exogenous pre-determined sectoral interlinkages — results robust (Figure A14d). ### Upstream and downstream propagation - Domestic and foreign shocks decomposed into upstream and downstream using relative inputs sourced from or supplied by a sector (Section 2.3.1). - Sector examples (Figure 5 referenced): - Agriculture (early stage of supply chain): - Suffers mostly from upstream weather shocks (negative for domestic and foreign). - Benefits from downstream weather shocks (positive for domestic and foreign, though foreign estimates imprecise). - Interpretation: weather shocks affect early-stage sectors as demand-side shocks that propagate upstream; agricultural damages exceed crop productivity impacts due to downstream input demand changes. - Sectors in later stages (other activities; wholesale, retail trade, restaurants and hotel): - Growth harmed by abnormally hot shocks in supplier sectors both domestically and abroad (downstream propagation). - Foreign and domestic upstream shocks have positive effects, with shocks in other countries larger in magnitude and statistically significant, indicating potential mitigation via competition and price effects. - Downstream propagation confirms firm-level evidence (Barrot & Sauvagnat, 2016); upstream propagation from customer sectors reveals a demand-side transmission channel. ### Persistence of network shocks over time (local projections) - Local projections estimated (Jordà, 2005) to obtain impulse response functions up to horizon h = 10 years; when h = 0 equation reduces to Equation (8). - Country-level results (Figure A16 referenced): - Domestic heat shocks (average number of days above the 95th percentile weighted by within-country trade interlinkages) have a persistent effect on total value-added levels; coefficient does not increase in absolute value over time. - Estimates are small, noisy, and not distinguishable from zero. - Foreign heat shocks have a positive significant effect starting from five years after the shock, suggesting competition and price effects. - Sector-specific impulse responses to a standardized domestic heat shock (Figure 6 referenced): - Aggregation conceals heterogeneous effects across sectors. - Several sectors do not recover from domestic partner abnormally hot shocks, with strong negative growth effects in: - construction; - mining, manufacturing and utilities; - wholesale, retail trade, restaurants and hotels. - Other activities: negative growth effect lasts five years then becomes statistically indistinguishable from zero. - Own and foreign shocks (Figures A17 and A18 referenced): - Direct shocks generally do not have persistent growth effects on sectoral production. - Agriculture is the only sector harmed persistently by direct shocks, but the negative effect lasts only one year and dissipates thereafter (no visible long-run growth effects). - Foreign shocks estimates are very small, positive but not distinguishable from zero at any lag — consistent with competition and price effects possibly benefiting sectoral production. - Suggestion: a more spatially disaggregated definition of foreign shocks is a promising avenue for future research. - Overall conclusion on persistence: - Trade-induced propagation of weather shocks importantly affects local economic outcomes and the persistence of impacts varies markedly across sectors; some downstream-intensive sectors exhibit multi-year negative growth responses to domestic network heat shocks. *Source: wpiea2023053-print-pdf - 5.1 Abnormally hot temperature shocks* ### 5.2 Other extreme weather events ### 5.2 Other extreme weather events ### Droughts - Dryness conditions directly harm agriculture and marginally benefit sectors that would be less productive under wetter conditions than the historical average, such as transportation and construction. - Agriculture is the only sector directly harmed by drought shocks, with a sizeable negative effect of 0.09 p.p. (sample mean is 0.002) associated with a 1 SD increase in the dryness conditions in the country. - Own drought shocks strongly benefit economic production in other sectors (construction; mining, manufacturing and utilities; and transport, storage and communication sectors), improving the precision of the positive estimates obtained when omitting network shocks. - Industries in the tertiary sector at later stages of the value chain (wholesale, retail trade, restaurants and hotel, and other activities) are virtually not impacted at all by their own drought shocks, with a coefficient very close to zero. - Network (propagation) effects: - Domestic shocks have a strong negative effect only on economic production in mining, manufacturing and utilities. - Domestic shocks have imprecisely estimated negative effects on construction; transport, storage and communication; and wholesale, retail trade, restaurants and hotel. - Foreign shocks have a sizable negative effect on other activities and wholesale, retail trade, restaurants and hotel, suggesting strong propagation of droughts through the economy and across countries in later stages of the supply chain. - Mining, manufacturing and utilities sector is strongly harmed by domestic droughts, with a magnitude comparable to the coefficient associated with own shocks, implying the net effect of droughts in a country on this sector is not as positive as own shocks alone indicated. - Accounting for both domestic and foreign network shocks is necessary to reveal the overall sectoral damage due to droughts hitting partner sectors. ### Tropical cyclones - Tropical cyclone intensity (wind speed) has the widest impact across sectors, damaging agriculture and other activities. - Agriculture and other activities remain the only two sectors directly harmed by tropical cyclones. - Domestic cyclone shocks: - Do not appear to have strong effects at earlier stages of the value chain. - Have a strong negative effect on transport, storage and communication growth rate with an effect of magnitude comparable and larger than the direct effect on other activities. - Foreign network shocks: - Have a null effect for most sectors. - Are positive for wholesale, retail trade, restaurants and hotel, indicating that damages in other countries due to cyclone shocks may improve sectoral economic production in that sector, which may benefit from international competition. - Trade linkages may either amplify or mitigate aggregate damage from cyclones, given potential direct impacts on capital stock. ### Discussion and conclusion — propagation through production networks - The analysis identifies a new component of climate damages arising from propagation of weather shocks through production networks across sectors and countries, and over time. - Key findings: - Sectors not negatively affected by local weather shocks can suffer losses due to interdependence with other sectors hit by weather shocks domestically or abroad. - Sectors at later stages of the supply chain (transport, storage and communication; wholesale, retail trade, restaurants and hotels; and other activities) are negatively impacted by heat shocks in other sectors, with losses comparable in magnitude to the direct impact on agriculture. - There is a strong negative persistent growth effect of domestic shocks in certain sectors’ output (construction; mining, manufacturing and utilities; wholesale, retail trade, restaurants and hotels) up to ten years since the shock occurred. - Climate damages may be larger than indicated by standard empirical approaches and integrated assessment models when network propagation is considered. ### Quantification of recent warming impacts (heat shocks, 2001-2020 vs 1970-2000 counterfactual) - Simulation setup: assess how much slower or faster each sector would have grown over the 2001-2020 period compared to a counterfactual in which the number of abnormally hot days in each country evolves linearly from its 1970-2000 long-run average; comparisons omit and account for propagation of shocks (see Figure A21; Appendix Section A.3). - Aggregate sectoral results (average annual relative GVA per capita loss across sectors due to recent warming): - Considering only sector-specific local shocks: -0.02% ( -0.04% median, IQR [-0.09,0.02]). - Accounting for indirect heat shocks (propagation through partners): 0.28% (0.24% median, IQR [0.04,0.44]). - Sectoral detail: - Only agriculture and mining, manufacturing and utilities are damaged by local heat shocks compared to the counterfactual; other services sectors appear sheltered by local shocks. - Accounting for heat shocks in trade partners, damages increase particularly in sectors sheltered from local shocks (mining, manufacturing, utilities; other activities; transport, storage and communications). - Construction and wholesale, retail, hotel and restaurants show larger heterogeneity: larger damages in Sub-Saharan Africa and South-East Asia are offset by modest benefits in Latin America and Europe. - National aggregation (using each country’s average sectoral breakdown of total GVA between 2001 and 2020): - Accounting for indirect heat shocks, national damages are substantial: 0.3% mean, 0.26% median, IQR [0.16,0.37]. - Omitting heat shocks propagation substantially underestimates damages: -0.02% mean, -0.01% median, IQR [-0.04,0.05]. - Estimation details: sector-specific semi-elasticities used were estimated with 1000 bootstrap replications with replacement. ### Limitations and avenues for further research - Adaptation: - The analysis provides modest but suggestive evidence of adaptation’s role but does not explicitly model adaptive investments, technological change, or other sector-specific adaptive responses (e.g., irrigation, sea walls...) that may heterogeneously affect response functions and mitigate climate damage. - Fiscal policy responses to climate-induced economic losses are not accounted for. - The study examines transmission of weather shocks, not transmission of shock-induced damages in other sectors and countries; estimates should be interpreted as the net effect. - Beliefs and expectations: - The analysis is mostly silent about decision-makers’ climate beliefs and expectation formation processes, which can affect adaptation and explain substantial economic losses due to inadequate adaptation. - Future work should allow for heterogeneous perceptions and expectations of climatic conditions in production networks and supply-chain relationships. - Spatial resolution: - The spatial level of the analysis may mask substantial variation in economic responses and local weather fluctuations. - High spatial resolution particularly matters for estimating precipitation effects; replicating the analysis on disaggregated sector-level sub-national data could yield new estimates and insight into regional propagation. - Input specificity and firm-level channels: - Transmission is studied through relative importance of trade partners in input-output interlinkages, but input specificity and elasticity of substitution (search frictions, relation-specific investments, supplier concentration and diversification) are key drivers documented at the firm level and merit integration into sector-level analyses. - Time-varying networks and reallocation: - The analysis focuses on propagation through an exogenous time-invariant trade network. - Sectoral reallocation and adjustments in trade patterns as adaptive margins to climate change remain an open avenue for research. *Source: wpiea2023053-print-pdf - 5.2 Other extreme weather events* ### References ### References ### Key themes in the cited literature - Climate impacts on economic activity and growth: - "Quantifying economic damages from climate change" (Auffhammer, 2018). - "Temperature shocks and economic growth" and related studies (Dell, Jones, & Olken, 2012; Dell, Jones, & Olken, 2014; Burke, Hsiang, & Miguel, 2015; Burke & Tanutama, 2019). - Studies on temperature–GDP relationships and climate damages (Newell, Prest, & Sexton, 2021; Kahn et al., 2021; Tol, 2019; Tol, 2022). - Sectoral and spatial heterogeneity in climate impacts (Desmet & Rossi-Hansberg, 2015; Henseler & Schumacher, 2019; Holtermann, 2020; Proctor et al., 2022). - Weather shocks, extreme events, and natural disasters: - Cyclones, droughts, floods, and heat waves and their economic consequences (Hsiang, 2010; Hsiang & Jina, 2014; Kunze, 2021; Loayza et al., 2012; Miller et al., 2021; Kousky, 2014). - Estimation of meteorological and drought indices (Palmer, 1965; Vicente-Serrano et al., 2010; Vicente-Serrano et al., 2012; McKee et al., 1993). - Tropical cyclone intensity measures and damages (Kunze, 2021). - Supply chains, firm networks, and propagation of shocks: - Input linkages, production networks, and shock propagation (Barrot & Sauvagnat, 2016; Boehm, Flaaen, & Pandalai-Nayar, 2019; Carvalho et al., 2021; Giroud & Mueller, 2019; Eaton, Kortum, & Kramarz, 2022). - Cross-border spillovers through trade and supply chains (Feng & Li, 2021; Kashiwagi, Todo, & Matous, 2021; Shughrue, Werner, & Seto, 2020; Osberghaus & Schenker, 2022). - Global multi-region input–output modeling and mapping world economic structure (Kanemoto et al., 2011; Lenzen et al., 2012). - Methods and empirical approaches: - Use of weather data, climate model output, and climate econometrics (Auffhammer et al., 2013; Hsiang, 2016; Kolstad & Moore, 2020; Lemoine, 2021). - Quasi-experimental and panel methods: shift-share designs, local projections, unit-root tests, and panel inference (Borusyak, Hull, & Jaravel, 2022; Jordà, 2005; Im, Pesaran, & Shin, 2003; Deschenes & Meng, 2018). - Adaptation, valuation of mortality consequences, and accounting for adaptation costs and benefits (Carleton et al., 2022; Shrader, 2021; Zappalà, 2022a; Zappalà, 2022b). - Data sources and climate normals: - ERA5 hourly data (Muñoz Sabater, 2019). - NOAA’s 1981–2010 US climate normals (Arguez et al., 2012). - Landscan global 2000 [data set] (Bright & Coleman, 2001). - Global Agricultural Lands: Croplands, 2000 (Ramankutty et al., 2010). - UNSD Gross Value Added by Kind of Economic Activity at constant (2015) prices - US dollars. ISIC Revision 3.1 (UNSD, 2022). - Fiscal and macro-fiscal responses to disasters: - Public spending and revenues after natural disasters; fiscal rules for small states prone to disasters (Noy & Nualsri, 2011; Nakatani, 2021; Felbermayr & Gröschl, 2014; Cunado & Ferreira, 2014). ### Appendix A.1 — Figures (figure descriptions and notes) - Figure A1: Average weights across countries - (a) Upstream - (b) Downstream - Notes: "The figure shows the average weights across countries by sector for upstream and downstream weights. Both upstream and downstream weights are constructed from the perspective of Source sectors in the x-axis." - Figure A2: Countries in the sample by climatic zone - Notes: "The map represents the countries in the sample divided by climatic zones, defined as terciles of the average annual temperature from 1970 through 2020. The classification is implemented in order to compute heterogeneous treatment effects as reported in Figure A3." - Figure A3: Heterogeneity in the GVA response to changes in temperature distribution - (a) Income groups - (b) Climate terciles - Notes: "The figure shows the (standardized) coefficients associated with the response of sectoral GVA per capita growth rate to an increase in the sum of average daily temperature in different sub-samples split by income groups according to the World Economic Outlook (IMF, 2022) and by climate split into terciles using the long-run average temperature. All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A4: Abnormal weather realizations using 1st and 99th percentiles - (a) Hot and cold temperature shocks - (b) Wet and dry precipitation shocks - Notes: "The figure shows the (standardized) regression estimates for the country-average number of days above the 99th and below the 1st percentile of the daily distribution in temperature (Panel (a)) and in precipitation (Panel (b)). All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A5: Abnormal weather realizations using 10th and 90th percentiles - (a) Hot and cold temperature shocks - (b) Wet and dry precipitation shocks - Notes: "The figure shows the (standardized) regression estimates for the country-average number of days above the 90th and below the 10th percentile of the daily distribution in temperature (Panel (a)) and in precipitation (Panel (b)). All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A6: Robustness: Abnormal temperature realizations - (a) Balanced panel - (b) Excluding “large” countries - Notes: "The figure shows the (standardized) regression estimates for the country-average number of days above the 95th and below the 5th percentile of the daily distribution in temperature using a sector-country balanced panel (Panel (a)) and excluding large countries (Brazil, China, India, Russia, US) (Panel (b)). All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. Bins represent the 90% confidence intervals around point estimates." - Figure A7: Abnormal weather realizations from time-varying climate norms using 5th and 95th percentiles - (a) Temperature - 20-year - (b) Precipitation - 20-year - (c) Temperature - 30-year - (d) Precipitation - 30-year - (e) Temperature - 40-year - (f) Precipitation - 40-year - Notes: "The figure shows the (standardized) regression coefficients on the number of days above the 90th and below the 10th percentile of the daily distribution in temperature (Panels (a-c-e)) and in precipitation (Panels (b-d-f)) using time-varying distributions (respectively, 20-year, 30-year and 40-year). The estimation sample starts from 1990. All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A8: Abnormal weather realizations from time-varying climate norms using 10th and 90th percentiles - (a) Temperature - 20-year - (b) Precipitation - 20-year - (c) Temperature - 30-year - (d) Precipitation - 30-year - (e) Temperature - 40-year - (f) Precipitation - 40-year - Notes: "The figure shows the (standardized) regression coefficients on the number of days above the 90th and below the 10th percentile of the daily distribution in temperature (Panels (a-c-e)) and in precipitation (Panels (b-d-f)) using time-varying distributions (respectively, 20-year, 30-year and 40-year). The estimation sample starts from 1990. All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A9: Abnormal weather realizations from time-varying climate norms using 1st and 99th percentiles - (a) Temperature - 20-year - (b) Precipitation - 20-year - (c) Temperature - 30-year - (d) Precipitation - 30-year - (e) Temperature - 40-year - (f) Precipitation - 40-year - Notes: "The figure shows the (standardized) regression coefficients on the number of days above the 99th and below the 1st percentile of the daily distribution in temperature (Panels (a-c-e)) and in precipitation (Panels (b-d-f)) using time-varying distributions (respectively, 20-year, 30-year and 40-year). The estimation sample starts from 1990. All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A10: Extreme drought and wetness prevalence and sectoral production - Notes: "The figure shows the (standardized) coefficients from a stacked multi-sector regression model where changes in dryness and wetness variables are sector-specific. All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A11: Tropical cyclone intensity and sectoral production - Notes: "The figure shows the (standardized) sector-specific coefficients from a stacked multi-sector regression model where the main regressor is changes in damage intensity measure of tropical cyclones constructed from wind speed from Kunze (2021). All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects and controlling for country-specific linear time trends. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A12: Domestic abnormally hot temperature shocks and sectoral production - Notes: "Bars represent the (standardized) sector-specific coefficients associated with direct shocks and domestic shocks, using the average number of days above the 95th percentile of the daily temperature distribution. Domestic shocks are constructed as the average weather shock in the other sectors in the same country as the sector of interest weighted by the average of upstream and downstream interdependence with each sector. All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects and accounting for sector-specific responses to temperature realizations below the 5th percentile and sector-specific responses to precipitation realizations below the 5th and above the 95th percentile. Bins represent the 90% confidence intervals with standard errors clustered at the country-level." - Figure A13: Robustness: Spatial correlation - (a) Subregion-by-year FE - (b) Continent-by-year FE - Notes: "The figure shows the (standardized) sector-specific coefficients associated with direct shocks and domestic and foreign shocks, using the average number of days above the 95th percentile of the daily temperature distribution. Domestic shocks are constructed as the average weather shock in the other sectors in the same country as the sector of interest weighted by the average of upstream and downstream interdependence with each sector. Symmetrically, foreign shocks are constructed as the average weather shock in the other sectors in all the other countries weighted by the average of upstream and downstream interdependence with each sector. All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects and accounting for sector-specific responses to temperature realizations below the 5th percentile and sector-specific responses to precipitation realizations below the 5th and above the 95th percentile. Panel (a) shows the estimates in a regression that additionally accounts for subregion-by-year fixed effects, Panel (b) shows the estimates in a regression that additionally accounts for continent-by-year fixed effects. Bins represent the 90% confidence intervals around point estimates." - Figure A14: Robustness: Domestic and foreign shocks - (a) Accounting for incomplete shares - (b) Balanced panel - (c) Excluding “large” countries - (d) After 1990 sample - Notes: "The figure shows the (standardized) sector-specific coefficients associated with direct shocks and domestic and foreign shocks, using the average number of days above the 95th percentile of the daily temperature distribution. Domestic shocks are constructed as the average weather shock in the other sectors in the same country as the sector of interest weighted by the average of upstream and downstream interdependence with each sector. Symmetrically, foreign shocks are constructed as the average weather shock in the other sectors in all the other countries weighted by the average of upstream and downstream interdependence with each sector. All sector-specific coefficients are estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects and accounting for sector-specific responses to temperature realizations below the 5th percentile and sector-specific responses to precipitation realizations below the 5th and above the 95th percentile. Panel (a) shows the estimates controlling for sector-year FE interacted with the sum of exposure shares. Panel (b) uses sector-country balanced panel, Panel (c) excludes large countries (Brazil, China, India, Russia, US), Panel (d) uses only the sample after" *Content derived from "References" and "Appendix A.1 Figures" in the source PDF.* ### 1990. Bins represent the 90% confidence intervals around point estimates. ### wpiea2023053-print-pdf - 1990. Bins represent the 90% confidence intervals around point estimates. ### Network and direct weather shock effects on sectoral production - Figure A15: Network abnormally hot temperature shocks and sectoral production - Sector-specific coefficients represent standardized effects of direct shocks and domestic and foreign shocks distinguished between upstream and downstream, using the average number of days above the 95th percentile of the daily temperature distribution. - Domestic upstream (resp. downstream) shocks: average weather shock in the other sectors in the same country weighted by upstream (resp. downstream) interdependence. - Foreign upstream (resp. downstream) shocks: average weather shock in other sectors abroad weighted by upstream (resp. downstream) interdependence. - Coefficients estimated in a stacked regression model fully saturated with country-sector and sector-year fixed effects, accounting for sector-specific responses to temperature realizations below the 5th percentile and sector-specific responses to precipitation realizations below the 5th and above the 95th percentile. - Bins represent the 90% confidence intervals with standard errors clustered at the country-level. - Figure A19: Direct and network drought shocks and sectoral production - Bars: standardized sector-specific coefficients for direct shocks and domestic and foreign shocks using changes in extreme drought prevalence. - Domestic shocks: average shock in other sectors in the same country weighted by the average of upstream and downstream interdependence. - Foreign shocks: average shock in other sectors in all other countries weighted by the average of upstream and downstream interdependence. - All sector-specific coefficients estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. - Bins represent the 90% confidence intervals with standard errors clustered at the country-level. - Figure A20: Direct and network cyclones shocks and sectoral production - Bars: standardized sector-specific coefficients associated with direct shocks and domestic and foreign shocks, using the changes in the wind speed measure by Kunze (2021). - Domestic shocks: average shock in other sectors in the same country weighted by the average of upstream and downstream interdependence. - Foreign shocks: average weather shock in other sectors in all other countries weighted by the average of upstream and downstream interdependence. - All sector-specific coefficients estimated jointly in a stacked regression model fully saturated with country-sector and sector-year fixed effects. - Bins represent the 90% confidence intervals with standard errors clustered at the country-level. ### Impulse responses: domestic vs foreign abnormally hot temperature shocks - Figure A16: Local projections of domestic and foreign heat shocks on total value added - Impulse response functions of per capita total value added growth rate to a 1 SD increase in heat shocks. - Estimated in a stacked regression model with country and year fixed effects and accounting for abnormally cold temperature shocks (below the 5th percentile) and precipitation realizations below the 5th and above the 95th percentile. - Horizon 0 is the year of the shock. - Shaded areas represent the 90% confidence intervals with standard errors clustered at the country level. - Panel (a) shows estimates for domestic shocks; Panel (b) shows estimates for foreign shocks. - Figure A17: Local projections of own abnormally hot temperature shocks on sectoral production - Sector-specific impulse response of sectoral per capita GVA growth rate to a 1 SD increase in own abnormally hot temperature shocks. - Estimated in a stacked regression model fully saturated with country-sector and sector-year fixed effects and accounting for sector-specific responses to domestic and foreign abnormally hot temperature shocks, abnormally cold temperature shocks (below the 5th percentile), and precipitation realizations below the 5th and above the 95th percentile. - Horizon 0 is the year of the shock. - Shaded areas represent the 90% confidence intervals with standard errors clustered at the country-level. - Figure A18: Local projections of foreign abnormally hot temperature shocks on sectoral production - Sector-specific impulse response of sectoral per capita GVA growth rate to a 1 SD increase in foreign abnormally hot temperature shocks. - Estimated in a stacked regression model fully saturated with country-sector and sector-year fixed effects and accounting for sector-specific responses to direct and domestic abnormally hot temperature shocks, abnormally cold temperature shocks (below the 5th percentile), and precipitation realizations below the 5th and above the 95th percentile. - Horizon 0 is the year of the shock. - Shaded areas represent the 90% confidence intervals with standard errors clustered at the country-level. ### Estimated average annual sectoral GVA per capita losses from recent warming (2001-2020) - Figure A21: Average annual relative sectoral GVA pc losses (%) due to recent warming - Shows average annual losses (in red) and gains (in blue) in sectoral per capita GVA due to abnormally hot temperature shocks in the 2001-2020 period compared to a counterfactual where heat shocks evolved linearly from their 1970-2000 averages. - Panel (a): using sector-specific local heat shocks estimates. - Panel (b): accounting for shocks in other partner sectors estimated from Equation (8). - Averages obtained from 1000 bootstrap estimations. - In panel (a), only estimates in Agriculture are statistically significant at 95% level. - Table A12 (referenced) reports estimated average losses at the country-sector level significant at the 95% level. ### Data coverage and sample composition - Table A1: Countries and year-sectors in final sample - Total 51,129 year-sectors. - Example entries (Country : Number of years-sectors): Afghanistan 300; Albania 300; Algeria 300; ...; Zimbabwe 300. (Full table lists many countries with counts such as North Korea 200, North Macedonia 180, South Sudan 72, Sudan 72, Tajikistan 178, Turkmenistan 180, Ukraine 180, United States 300, China 300, India 300, etc.) ### Sector mapping and classifications - Table A2: Mapping between EORA26 sectors and UNSD industries - Agriculture -> Agriculture, hunting, forestry, fishing (ISIC A-B) - Fishing -> Agriculture, hunting, forestry, fishing (ISIC A-B) - Mining and Quarrying -> Mining, Manufacturing, Utilities (ISIC C-E) - Electricity, Gas and Water -> Mining, Manufacturing, Utilities (ISIC C-E) - Food & Beverages -> Mining, Manufacturing, Utilities (ISIC C-E) - Textiles and Wearing Apparel -> Mining, Manufacturing, Utilities (ISIC C-E) - Wood and Paper -> Mining, Manufacturing, Utilities (ISIC C-E) - Petroleum, Chemical and Non-Metallic Mineral Products -> Mining, Manufacturing, Utilities (ISIC C-E) - Metal Products -> Mining, Manufacturing, Utilities (ISIC C-E) - Electrical and Machinery -> Mining, Manufacturing, Utilities (ISIC C-E) - Transport Equipment -> Mining, Manufacturing, Utilities (ISIC C-E) - Other Manufacturing -> Mining, Manufacturing, Utilities (ISIC C-E) - Recycling -> Mining, Manufacturing, Utilities (ISIC C-E) - Construction -> Construction (ISIC F) - Maintenance and Repair -> Construction (ISIC F) - Wholesale Trade -> Wholesale, retail trade, restaurants and hotels (ISIC G-H) - Retail Trade -> Wholesale, retail trade, restaurants and hotels (ISIC G-H) - Hotels and Restaurants -> Wholesale, retail trade, restaurants and hotels (ISIC G-H) - Transport -> Transport, storage and communication (ISIC I) - Post and Telecommunications -> Transport, storage and communication (ISIC I) - Financial Intermediation and Business Activities -> Other Activities (ISIC J-P) - Public Administration -> Other Activities (ISIC J-P) - Education, Health and Other Services -> Other Activities (ISIC J-P) - Private Households -> Other Activities (ISIC J-P) - Others -> Other Activities (ISIC J-P) - Re-export & Re-import -> Other Activities (ISIC J-P) - Notes: Author’s classification based on Kunze (2021) and adapted to six UNSD sectors. - Table A3: Classification of countries by income group - Advanced Economies: e.g., Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark, ..., United States. - Emerging Market Economies: e.g., Albania, Algeria, Angola, Antigua and Barbuda, Argentina, Armenia, ..., Uruguay, Vanuatu, Venezuela. - Low-Income Developing Countries: e.g., Afghanistan, Bangladesh, Benin, Bhutan, Bolivia, Burkina Faso, ..., Zambia, Zimbabwe. - Notes: Author’s classification based on IMF World Economic Outlook (IMF, 2022). An anomaly is defined as the sum of the monthly deviations from the monthly 25-year moving average, distinguishing between positive and negative. ### Weather shock summary statistics and stationarity - Table A4: Summary statistics on weather shocks (N, mean, sd, min, max) - Temperature above 90th percentile (days/year): N 8,548; mean 37.487; sd 23.6100; min 0; max 222 - Temperature below 10th percentile (days/year): N 8,548; mean 35.907; sd 21.0230; min 0; max 210 - Precipitation above 90th percentile (days/year): N 8,548; mean 36.458; sd 9.9077; min 0; max 111 - Precipitation below 10th percentile (days/year): N 8,548; mean 32.390; sd 16.3670; min 0; max 114 - Temperature above 99th percentile (days/year): N 8,548; mean 3.851; sd 6.1450; min 0; max 94 - Temperature below 1st percentile (days/year): N 8,548; mean 3.563; sd 4.8920; min 0; max 54 - Precipitation above 99th percentile (days/year): N 8,548; mean 3.659; sd 2.5390; min 0; max 29 - Precipitation below 1st percentile (days/year): N 8,548; mean 2.474; sd 3.1870; min 0; max 32 - Negative temperature anomaly (◦C): N 7,817; mean 3.274; sd 3.353; min 0.00005; max 31.413 - Positive temperature anomaly (◦C): N 7,817; mean 6.33; sd 4.728; min 0.0007; max 32.898 - Negative precipitation anomaly (mm): N 7,817; mean 6.980; sd 5.308; min 0.009; max 48.944 - Positive precipitation anomaly (mm): N 7,817; mean 6.836; sd 6.322; min 0.00008; max 89.33 - Notes: Abnormal weather realizations and anomalies are measured in levels. - Table A5: Im-Pesaran-Shin unit-root test for main variables (Statistic, p-value) - GVA growth rate: -6.072 0.000 - Abnormally dry precipitation shock (p1): -6.782 0.000 - Abnormally dry precipitation shock (p5): -6.464 0.000 - Abnormally dry precipitation shock (p10): -6.456 0.000 - Abnormally wet precipitation shock (p90): -6.571 0.000 - Abnormally wet precipitation shock (p95): -6.600 0.000 - Abnormally wet precipitation shock (p99): -6.832 0.000 - Abnormally cold temperature shock (p1): -6.541 0.000 - Abnormally cold temperature shock (p5): -6.134 0.000 - Abnormally cold temperature shock (p10): -6.128 0.000 - Abnormally hot temperature shock (p90): -6.156 0.000 - Abnormally hot temperature shock (p95): -6.258 0.000 - Abnormally hot temperature shock (p99): -6.575 0.000 - Notes: Null hypothesis is that all panels contain unit roots against alternative that some panels are stationary. Tests remove cross-sectional means and include a time trend; no lags included. Growth rate test on balanced sector-country-year panel; weather variables on balanced country-year panel using population-weighted weather variables. ### Sector-specific regression results: temperature and precipitation impacts - Table A6: Sector-specific impact of positive annual temperature and precipitation changes on GVA per capita growth rate (columns (1)-(3)) - Temperature coefficients (selected, standard errors in parentheses): - Agriculture, hunting, forestry, fishing: -0.00676 ∗∗ (0.00297); -0.00726 ∗∗ (0.00305); -0.00773 ∗∗ (0.00300) - Construction: 0.000787 (0.00401); 0.000861 (0.00403); 0.000352 (0.00403) - Mining, Manufacturing, Utilities: 0.00229 (0.00251); 0.00205 (0.00253); 0.00162 (0.00256) - Transport, storage and communication: 0.00410 (0.00266); 0.00423 (0.00271); 0.00370 (0.00272) - Wholesale, retail trade, restaurants and hotels: 0.00284 (0.00260); 0.00266 (0.00264); 0.00220 (0.00266) - Precipitation coefficients (selected): - Agriculture, hunting, forestry, fishing: 0.0117 ∗∗∗ (0.00291); 0.0122 ∗∗∗ (0.00299); 0.0117 ∗∗∗ (0.00293) - Transport, storage and communication: -0.00514 ∗∗ (0.00233); -0.00460 ∗ (0.00240); -0.00505 ∗∗ (0.00238) - GVA growth rate t−1: 0.0618 ∗∗ (0.0264) in column (1); 0.0399 (0.0257) in column (3). - Fixed effects: Country-Sector FE ✓, Sector-Year FE ✓; columns include country linear and quadratic time trends in some specifications. - N: 5127350; 16250; 162 (presented values reflect table formatting). - adj.R2: 0.043; 0.046; 0.060 - Notes: Standard errors clustered at country-level. Significance levels: ∗ p <0.1, ∗∗ p <0.05, ∗∗∗ p <0.01. - Table A7: Sector-specific impact of annual temperature and precipitation changes (alternative specification, columns (1)-(3)) - Temperature Changes (selected): - Agriculture, hunting, forestry, fishing: -0.0351 ∗∗ (0.0144); -0.0383 ∗∗ (0.0149); -0.0379 ∗∗ (0.0149) - Construction: 0.0402 ∗∗∗ (0.0153); 0.0360 ∗∗ (0.0157); 0.0362 ∗∗ (0.0155) - Mining, Manufacturing, Utilities: 0.0220 ∗ (0.0112); 0.0189 (0.0119); 0.0193 (0.0118) - Transport, storage and communication: 0.0230 ∗ (0.0124); 0.0200 (0.0127); 0.0205 (0.0126) - Precipitation Changes (selected): - Agriculture, hunting, forestry, fishing: 0.0405 ∗∗∗ (0.0114); 0.0417 ∗∗∗ (0.0119); 0.0409 ∗∗∗ (0.0117) - Transport, storage and communication: -0.00857 (0.00821); -0.00713 (0.00867); -0.00744 (0.00851) - GVA growth rate t−1: 0.0616 ∗∗ (0.0264) in column (1); 0.0400 (0.0257) in column (3). - Fixed effects: Country-Sector FE ✓, Sector-Year FE ✓; country linear and quadratic time trends included in some specifications. - N: 5022349; 13349; 133 (presented values reflect table formatting). - adj.R2: 0.044; 0.047; 0.060 - Notes: Standard errors clustered at country-level. A graphical representation of column (2) coefficients is reported in Figure 1. Significance levels: ∗ p <0.1, ∗∗ p <0.05, ∗∗∗ p <0.01. *Italic: Content derived from wpiea2023053-print-pdf - 1990. Bins represent the 90% confidence intervals around point estimates.* ### 2. Significance levels: ### 2. Significance levels: ### Significance thresholds - ∗ p <0.1 - ∗∗ p <0.05 - ∗∗∗ p <0.01 ### Table A8 — Heterogeneous effect of annual changes in temperature and precipitation by income groups (sectoral GVA per capita growth rate) - Sample size: N46243, N45235, N? (table columns report N46243 45235 ? — preserve printed sequence: N462434523545235) - adj. R2 reported: 0.0470, 0.0500, 0.064 (for columns (1)(2)(3) respectively) - Covariates and fixed effects: Country-Sector FE ✓✓✓; Sector-Year FE ✓✓✓; Country linear time trends ✓; Country quadratic time trends ✓ - Key temperature coefficients (selected, preserving exact printed values and significance): - Advanced Economies: - Agriculture: 0.0272 0.0287 ∗ 0.0284 ∗ (standard errors: (0.0176) (0.0153) (0.0151)) - Construction: 0.0668 ∗∗∗ 0.0653 ∗∗∗ 0.0651 ∗∗∗ ( (0.0198) (0.0170) (0.0167) ) - Mining, Manufacturing, Utilities: 0.0153 ∗∗ 0.0176 ∗∗ 0.0180 ∗∗ ( (0.00689) (0.00886) (0.00856) ) - Wholesale, retail trade, restaurants and hotels: 0.0223 ∗∗∗ 0.0176 ∗∗ 0.0177 ∗∗ ( (0.00844) (0.00720) (0.00716) ) - Emerging Economies: - Agriculture: -0.0804 ∗∗∗ -0.0844 ∗∗∗ -0.0845 ∗∗∗ ( (0.0191) (0.0202) (0.0202) ) - Transport, storage and communication: 0.0440 ∗ 0.0395 0.0389 ( (0.0254) (0.0261) (0.0260) ) - Low-Income Developing Countries: - Agriculture: -0.0762 ∗∗ -0.0888 ∗∗ -0.0852 ∗∗ ( (0.0354) (0.0380) (0.0384) ) - Wholesale, retail trade, restaurants and hotels: -0.0119 -0.0172 -0.0128 ( (0.0331) (0.0350) (0.0347) ) - Key precipitation coefficients (selected): - Advanced Economies: - Other Activities: -0.0076 -0.0148 ∗ -0.0143 ∗ ( (0.00644) (0.00756) (0.00739) ) - Emerging Economies: - Agriculture: 0.0225 ∗ 0.0222 ∗ 0.0217 ( (0.0132) (0.0133) (0.0132) ) - Low-Income Developing Countries: - Agriculture: 0.0466 ∗∗ 0.0488 ∗∗ 0.0477 ∗∗ ( (0.0195) (0.0204) (0.0203) ) - Additional reported regressor: GVA growth rate t−1 = 0.0566 ∗∗ 0.0344 ( (0.0280) (0.0272) ) ### Table A9 — Heterogeneous effect by climate terciles (sectoral GVA per capita growth rate) - Sample size: N50223 49133 49133? (printed sequence: N502234913349133) - adj. R2 reported: 0.0440, 0.0470, 0.060 (columns (1)(2)(3)) - Fixed effects: Sector FE ✓✓✓; Country FE ✓✓✓; Year FE ✓✓✓; Country linear time trends ✓; Country quadratic time trends ✓ - Key temperature coefficients (selected): - Cold Climate: - Construction: 0.0769 ∗∗∗ 0.0680 ∗∗∗ 0.0675 ∗∗∗ ( (0.0196) (0.0194) (0.0193) ) - Wholesale, retail trade, restaurants and hotels: 0.0392 ∗∗ 0.0353 ∗∗ 0.0351 ∗∗ ( (0.0172) (0.0174) (0.0174) ) - Temperate Climate: - Agriculture: -0.101 ∗∗∗ -0.103 ∗∗∗ -0.0998 ∗∗∗ ( (0.0312) (0.0319) (0.0321) ) - Transport, storage and communication: 0.0488 ∗ 0.0509 ∗ 0.0520 ∗ ( (0.0283) (0.0291) (0.0291) ) - Hot Climate: - Wholesale, retail trade, restaurants and hotels: -0.0555 ∗∗ -0.0585 ∗∗ -0.0552 ∗∗ ( (0.0235) (0.0234) (0.0234) ) - Key precipitation coefficients (selected): - Cold Climate: - Agriculture: 0.0389 ∗ 0.0395 ∗ 0.0405 ∗∗ ( (0.0205) (0.0207) (0.0205) ) - Mining, Manufacturing, Utilities: 0.0179 0.0213 0.0234 ∗ ( (0.0137) (0.0131) (0.0131) ) - Temperate Climate: - Agriculture: 0.0417 ∗ 0.0428 ∗ 0.0411 ∗ ( (0.0216) (0.0224) (0.0222) ) - Wholesale, retail trade, restaurants and hotels: 0.0208 ∗ 0.0218 ∗ 0.0211 ∗ ( (0.0122) (0.0127) (0.0125) ) - Hot Climate: - Agriculture: 0.0271 ∗ 0.0279 ∗ 0.0277 ∗ ( (0.0156) (0.0164) (0.0162) ) - Transport, storage and communication: -0.0319 ∗∗ -0.0314 ∗ -0.0313 ∗ ( (0.0153) (0.0163) (0.0159) ) - Additional reported regressor: GVA growth rate t−1 = 0.0620 ∗∗ 0.0404 ( (0.0264) (0.0258) ) ### Table A10 — Dryness and wetness shocks and sectoral GVA (standardized sector-specific coefficients) - Sample size: N35911 49578 49578? (printed sequence: N359114957849578) - adj. R2 reported: 0.0490, 0.0470, 0.046 (columns (1)(2)(3)) - Fixed effects: Country-Sector FE ✓✓✓; Sector-Year FE ✓✓✓ - Measures (columns): - (1) Average dryness (average of monthly negative realizations of SPEI) - (2) Extreme drought prevalence (maximum share of grid-months with SPEI<-2) - (3) Extreme wetness prevalence (maximum share of grid-months with SPEI>2) - Key coefficients (selected): - Agriculture, hunting, forestry, fishing: - (1): -0.119 ∗∗∗ ( (0.0197) ) - (2): -0.0733 ∗∗∗ ( (0.0126) ) - (3): -0.00346 ( (0.0116) ) - Construction: - (1): 0.0184 ( (0.0156) ) - (2): 0.0281 ∗∗ ( (0.0135) ) - (3): -0.00293 ( (0.0133) ) - GVA growth rate t−1: 0.0687 ∗∗ 0.0605 ∗∗ 0.0605 ∗∗ ( (0.0282) (0.0263) (0.0263) ) ### Table A11 — Tropical cyclones and sectoral GVA (sector-specific standardized coefficients) - Note: Tropical cyclones data are available until 2015. - Sample size printed: N44167 44167 44167 (sequence: N441674416744167) - adj. R2 reported: 0.0530, 0.0530, 0.053 (columns (1)(2)(3)) - Fixed effects: Country-Sector FE ✓✓✓; Sector-Year FE ✓✓✓; Country linear time trends ✓✓; Country quadratic time trends ✓✓ - Key coefficients (selected): - Agriculture, hunting, forestry, fishing: - (1): -0.0288 ∗∗ ( (0.0125) ) - (2): -0.0297 ∗∗ ( (0.0126) ) - (3): -0.0315 ∗∗∗ ( (0.0119) ) - Other Activities: - (1): -0.00500 ∗ ( (0.00278) ) - (3): -0.00603 ∗∗ ( (0.00289) ) ### Table A12 — Sector-country damages (% loss GVA per capita) significant at 95% level - Presentation: Average loss for each sector as a % loss in GVA per capita relative to observed production between 2001 and 2020; 95% confidence intervals from 1000 bootstrap estimates of Equation 8. - Representative extracted entries (preserving exact printed values and 95% CI brackets): - Afghanistan — Agriculture: 0.48 [ 0.14 0.81 ] - Angola — Agriculture: 0.91 [ 0.18 1.66 ] - Argentina — Agriculture: 0.44 [ 0.04 0.86 ] - Brazil — Agriculture: 0.63 [ 0.10 1.18 ] - Colombia — Agriculture: 1.26 [ 0.15 2.37 ] - Ecuador — Agriculture: 1.13 [ 0.37 1.85 ] - India — Agriculture: 0.37 [ 0.04 0.70 ] - Indonesia — Agriculture: 0.97 [ 0.45 1.52 ] - Mexico — Agriculture: 0.43 [ 0.08 0.78 ] - USA — Wholesale, retail, hotel, restaurant: 1.81 [ 0.07 3.51 ] - Netherlands — Construction: 1.21 [ 0.23 2.17 ] - Spain — Construction: 1.73 [ 0.34 3.11 ] - Selected negative significant entries (preserving signs and intervals): - Azerbaijan — Agriculture: -0.70 [ -1.35 -0.07 ] - Bosnia and Herzegovina — Agriculture: -0.11 [ -0.18 -0.03 ] - Turkmenistan — Construction: -1.21 [ -2.22 -0.20 ] - Czech Republic — Construction: -2.02 [ -3.87 -0.17 ] - Kazakhstan — Construction: -0.64 [ -1.28 -0.05 ] - Notes: Table reports average loss for each sector as % loss in GVA per capita over 2001–2020, accounting for own, domestic and foreign heat shocks. ### A.3 Computing the economic cost of the propagation of recent warming — methodology and formulas - Objective: Simulate how much slower or faster each sector would have grown annually over 2001-2020 compared to a counterfactual where number of hot days above the 95th percentile increases according to its historical trend of 1970-2000. - Country-specific trend estimation: - Estimate T_ct = α_c + β_c t + ε_ct in 1970-2000 sample, where T_ct is number of days above the 95th percentile of the country-specific daily average temperature distribution. - Use estimated ˆβ_c to construct counterfactual ˜T95_sct (assumes linear trend and no change in volatility). - Computation of annual direct and spillover growth effects (preserve formulas exactly): - g_direct_sct = b γ_s (T95_sct − ˜T95_sct) (Equation (10)) - g_spillover_sct = (b γ_s T95_sct + c γ_D_s T95,D_sct + c γ_F_s T95,F_sct) − (b γ_s ˜T95_sct + c γ_D_s ˜T95,D_sct + c γ_F_s ˜T95,F_sct) (Equation (11)) - T95_ct is observed number of days above 95th percentile in sector s in country c in year t. - ˜T95_sct is counterfactual predicted number had 1970-2000 average evolved linearly. - T95,J_sct is weighted average number of days above 95th percentile in trade partners J∈{Foreign, Domestic}. - γ_s’s are sector-specific coefficient estimates for own, domestic and foreign heat shocks from Equation (8). - Counterfactual sectoral log GVA per capita evolution (preserve formula exactly): - ˆY_p_sct = Y_sct−1 + y_sct + g_p_sct (Equation (12)) - where p ∈ {direct, spillover}; hat indicates counterfactual; Y is (log) GVA per capita; y is observed growth rate. - Average relative loss in GVA for sector s in country c over 2001-2020: - %LOSS_p_sc = (1/T) sum_{t=2001}^{2020} ( e^{ˆY_p_sct} − e^{Y_sct} ) / e^{Y_sct} (Equation (13)) - Aggregation to country level (preserve formula exactly): - %LOSS_p_c = Σ_s λ_sc %LOSS_p_sc (Equation (14)) - where λ_sc is the average share of total GVA of sector s in country c between 2001 and 2020. - Outputs: - Sector-level %LOSS_p_sc reported in Figure A21 (not reproduced here). - Country-level losses omitting and accounting for indirect heat shocks reported in Figure 7 (not reproduced here). - Caveat stated in text: Computation does not necessarily represent differential impact of recent anthropogenic warming accounting for network shocks and is agnostic to the cause of recent warming. *Source: Sectoral impact and propagation of weather shocks — Working Paper No. WP/2023/053* --- _Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023053-print-pdf.pdf_