## High-Frequency Impact of Temperature on Economic Activity

## Source details

**Canonical URL:** [High-Frequency Impact of Temperature on Economic Activity](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024178-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024178-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024178-print-pdf.pdf.json)

---

### Introduction and motivation
- Climate change manifests through rising temperatures and more frequent droughts, floods, wildfires, and massive storms; understanding temperature’s impact on economic activity is fundamental to adaptation and mitigation.
- Prior literature:
  - Sachs and Warner, 1997; Gallup, Sachs, and Mellinger, 1999 — hotter countries tend to be poorer.
  - Dell et al., 2012; Acevedo et al., 2020; Colacito et al., 2019 — higher temperature reduces economic growth in poor countries and the United States.
  - Mechanisms: reduced total factor productivity growth (Letta and Tol, 2019), reduced investment and labor productivity (Acevedo et al., 2020; Kalkuhl and Wenz, 2020), reduced sectoral productivity (Lepore and Fernando, 2023).
  - Non-linear effects: Burke et al. (2015) — economic growth rises with average annual temperature until approximately 13 °C, then reverses.
- Limitations of average and aggregated measures:
  - Annual averages can mask seasonal variation.
  - Large-area averages (country or province) can conceal spatial heterogeneity in temperature, fluctuations, and economic sensitivity.

### Granular approach: temporal and spatial resolution
- Analysis conducted at fine disaggregation:
  - Temporal: monthly temperature effects.
  - Spatial: 30-kilometer by 30-kilometer grid cells.
- Rationale: grid-and-month analysis allows more precise estimation across different climate zones and captures localized heterogeneity.

### Proxy for local economic activity and data considerations
- Local economic activity proxied by nighttime lights intensity (VIIRS DNB monthly average radiance).
- Data sources and coverage:
  - ERA5 Monthly Aggregates: 2m air temperature in Celsius, monthly total precipitation; global coverage from January 1979 to June 2020; grid cell resolution about 30 kilometers.
  - VIIRS Day/Night Band (DNB) monthly average radiance composite images: available January 2014 to June 2022; grid cell resolution about 500 meters. Observations with snow cover > 50 percent in a month are excluded.
  - Analysis period for merged datasets: January 2014 to June 2020 (78 months) on 30-kilometer by 30-kilometer grid cells.
  - Population data at grid-cell level from CIESIN and SEDAC, 2018; all data collected via Google Earth Engine.

### Main empirical findings (contemporaneous effects)
- Preferred specification result:
  - A 1 °C increase in monthly-average temperature reduces the contemporaneous year-over-year growth of monthly nighttime lights by 0.77 percent.
- GDP-growth implication:
  - Common estimates of the elasticity of nighttime lights with respect to GDP imply this would be associated with a reduction in GDP growth of between 0.50 and 0.77 percentage points.
- Precipitation:
  - Higher-than-normal precipitation (excessive rainfall) is associated with a reduction in growth in nighttime lights; main temperature estimates are not sensitive to controlling for precipitation.
- Selected coefficient estimates (preferred specification, standard errors clustered at province level):
  - Temperature (t): -0.0077 (0.0022) [***: p < 0.01]
  - Temperature (t-12): 0.0026 (0.0026)
  - High precipitation: -0.0202 (0.0069) [***: p < 0.01]
  - Number of observations: 8,438,699; Number of cells: 197,166

### Heterogeneity by climate zone and income
- Contemporaneous effect by baseline temperature bin (Appendix Table 3, temp(month, cell) - lag0):
  - [0.0, 10.0): 0.0062
  - [10.0, 20.0): -0.0104***
  - [20.0, 30.0): -0.0104***
  - [30.0, 35.0): -0.0044
  - >35.0: -0.0032
- Lagged (t-12) effects by bin (Appendix Table 3):
  - [0.0, 10.0): 0.0101
  - [10.0, 20.0): 0.0065**
  - [20.0, 30.0): 0.0023
  - [30.0, 35.0): 0.0036
  - >35.0: -0.0065
- Contemporaneous effects by temperature and income group (Appendix Table 4; estimates shown as percentages):
  - [0 °C, 10 °C)
    - High income: 1.98%
    - Low & Middle income: 0.11%
  - [10 °C, 20 °C)
    - High income: -0.13%
    - Low & Middle income: -1.81%***
  - [20 °C, 30 °C)
    - High income: -0.85%*
    - Low & Middle income: -1.16%**
  - [30 °C, 35 °C)
    - High income: 1.06%*
    - Low & Middle income: -1.00%***
  - >35 °C
    - High income: 0.28%
    - Low & Middle income: -0.47%
- Interaction evidence:
  - Interaction terms show statistically significant larger negative effects in Low & Middle income cells for several bins (e.g., -2.07%*** in [30 °C, 35 °C) interaction).

### Dynamic (persistence) results
- Local linear projections (Jordà, 2005) findings for a 1 °C shock:
  - Nighttime lights decrease by on average 0.5 percent during the first quarter after the shock.
  - Statistically significant reduction persists for three quarters.
  - Effects become statistically insignificant in the fourth quarter after the shock.
- Temperature persistence:
  - Temperatures remain elevated for 3 months after a monthly shock.
- Interpretation:
  - Persistence in nighttime-lights response lasts longer than persistence in temperature itself, suggesting channels beyond temperature persistence drive economic-activity persistence.

### Robustness checks and alternate specifications
- Contemporaneous temperature coefficients across robustness models (Table 3, Models 1.1–1.5):
  - Model 1.1 (omit 2017): -0.0073**  (0.0029)
  - Model 1.2 (omit 1% lowest and 1% highest growth in nighttime lights): -0.0042**  (0.0019)
  - Model 1.3 (omit both 1% lowest/highest growth and 1% lowest/highest levels): -0.0043**  (0.0019)
  - Model 1.4 (omit 30-km grid cells with population < 100): -0.0066***  (0.0022)
  - Model 1.5 (no precipitation control): -0.0071***  (0.0022)
- Selected precipitation-related coefficients (Table 3):
  - High precipitation:
    - Model 1.1: -0.0281***  (0.0072)
    - Model 1.2: -0.0129**  (0.0057)
    - Model 1.3: -0.0131**  (0.0057)
    - Model 1.4: -0.0282***  (0.0062)
    - Model 1.5: N/A
  - Low precipitation:
    - Model 1.1: 0.0196**  (0.0098)
    - Model 1.2: 0.0031  (0.007)
    - Model 1.3: 0.0031  (0.007)
    - Model 1.4: 0.0141*  (0.0074)
    - Model 1.5: N/A
- Robustness summary:
  - Contemporaneous and dynamic relationships between nighttime lights and temperature are robust across the five alternative specifications.
  - Additional results confirm robustness to using monthly average rather than monthly maximum temperatures (available upon request in the source).

### Data and sample statistics (selected)
- Analysis period: January 2014 to June 2020 (78 months).
- Overall sample (selected statistics):
  - Observations (cell × month): YoY Growth in Nighttime Lights 8,438,699
  - Number of cells: 197,166 (YoY Growth in Nighttime Lights), 200,648 (other variables)
  - Mean YoY Growth in Nighttime Lights: 0.1167
  - Median YoY Growth in Nighttime Lights: 0.0405
  - Mean Temperature: 28.5310 °C
  - Median Temperature: 30.3510 °C
  - Temperature has increased by 0.10°C (or 0.46 percent) annually for the typical (median) cell.
  - Median YoY growth rate of nighttime light is approximately 4 percent, with significant variation (94 percentage points).
- Appendix Table 1 — observations, cells, countries by temperature group:
  - [0 °C, 10 °C): Number of observations 322,271; Number of cells 64,396; Number of countries 72
  - [10 °C, 20 °C): Number of observations 1,337,642; Number of cells 110,371; Number of countries 95
  - [20 °C, 30 °C): Number of observations 2,327,248; Number of cells 143,240; Number of countries 180
  - [30 °C, 35 °C): Number of observations 2,332,987; Number of cells 125,112; Number of countries 166
  - >35 °C: Number of observations 2,118,551; Number of cells 85,935; Number of countries 136
- Sample sizes used in temperature-group contemporaneous estimates (Appendix Table 3 repeated):
  - Number of observations by bin: 322,271; 1,337,642; 2,327,248; 2,332,987; 2,118,551
  - Number of cells by bin: 64,396; 110,371; 143,240; 125,112; 85,935
- Table 3 sample sizes (Models 1.1–1.5):
  - Number of observations: 7,039,914; 8,269,925; 8,157,719; 7,127,418; 8,438,699
  - Number of cells: 197,103; 197,157; 196,502; 145,627; 197,166

### Theoretical and empirical framework (brief)
- Theoretical motivation follows Dell et al. (2012): economic activity Yt = e^{βTt} A_t^{α}; temperature has a direct level effect β and can affect productivity growth via λ and σ. After log differencing, short-run growth effect is (β + αλ) and long-run effect is α(λ + σ).
- Empirical contemporaneous specification (Eq. 4):
  - Dependent variable: log(L_{c,t}) − log(L_{c,t−12}) where L is nighttime lights.
  - Main regressors: T_{c,t} and T_{c,t−12}; controls include precipitation anomaly indicators, month-specific fixed effects (F E_t), and cell × calendar-month fixed effects (F E_{c,m}).
  - Standard errors clustered at the province (first-tier administrative) level.
- Nonlinear/asymmetric specifications and dynamic local linear projections (Eq. 5–7) are used to estimate positive/negative shock asymmetries and impulse responses up to 24 months.

### Key implications and context
- A negative and highly statistically significant effect of temperature on economic activity (proxied by nighttime lights) is found, larger than often reported in the literature.
- Negative and statistically significant effects prevail except in colder areas; warmer areas exhibit more negative impacts.
- Within temperature bins, poorer areas (low- and middle-income countries) suffer relatively more severe effects.
- Historical short-run responses may differ from future long-run climate-change impacts due to magnitude, non-linearities, tipping points, and adaptation limits.
- Stronger adaptation (e.g., drought-resistant seeds, diversification into resilient sectors) may soften impacts; limits to adaptation and large-scale behavioral changes (e.g., migration) could amplify future damages.

*Source: IMF Working Paper “High-Frequency Impact of Temperature on Economic Activity” (selected content).*

### References .............................................................................................................

### High-Frequency Impact of Temperature on Economic Activity

### Introduction and motivation
- Climate change manifests through rising temperatures and more frequent droughts, floods, wildfires, and massive storms; understanding temperature’s impact on economic activity is fundamental to adaptation and mitigation.
- Prior literature:
  - Early cross-country work: Sachs and Warner, 1997; Gallup, Sachs, and Mellinger, 1999 — hotter countries tend to be poorer (potential omitted-variable concerns).
  - Within-country fluctuation studies: Dell et al., 2012; Acevedo et al., 2020; Colacito et al., 2019 — higher temperature reduces economic growth in poor countries and the United States.
  - Mechanisms identified: reduced total factor productivity growth (Letta and Tol, 2019), reduced investment and labor productivity (Acevedo et al., 2020; Kalkuhl and Wenz, 2020), reduced sectoral productivity (Lepore and Fernando, 2023).
  - Non-linear effects: Burke et al. (2015) — economic growth rises with average annual temperature until approximately 13 °C, then reverses.
- Limitations of average and aggregated measures:
  - Annual averages can mask seasonal variation (example: Washington, D.C. with annual average close to 21 °C but cold winter and hot summer).
  - Large-area averages (country or province) can conceal spatial heterogeneity in temperature, fluctuations, and economic sensitivity.

### Granular approach: temporal and spatial resolution
- Analysis conducted at fine disaggregation:
  - Temporal: monthly temperature effects (rather than annual averages).
  - Spatial: 30-kilometer by 30-kilometer grid cells.
- Rationale: grid-and-month analysis allows more precise estimation across different climate zones and captures localized heterogeneity in both temperature and economic responses.

### Proxy for local economic activity and data considerations
- Local economic activity proxied by nighttime lights intensity.
  - Motivating literature: Chen and Nordhaus, 2011; Hu and Yao, 2022; Martinez, 2022; Asher et al., 2021.
  - Related work using nighttime lights: Felbermayr et al. (2022) — impacts of storms, excessive precipitation, droughts, and cold spells on economic activity proxied by nighttime lights.
- Acknowledged challenge: scarcity of economic data at local levels; nighttime lights used to overcome this limitation.

### Main empirical findings
- Effect on nighttime lights growth:
  - Under the fully specified model, a 1 °C increase in monthly-average temperature reduces the contemporaneous year-over-year growth of monthly nighttime lights by 0.77 percent.
- Implications for GDP growth:
  - Common estimates of the elasticity of nighttime lights with respect to GDP suggest the above effect would be associated with a reduction in GDP growth, relative to baseline, of between 0.5 and 0.77 percentage points.
- Comparison with prior findings:
  - The estimated magnitude is larger than many studies using aggregated data.
  - Consistent with Acevedo et al. (2020), which report that a 1 °C increase in annual temperature reduces output growth by 0.6– [text truncated in source].

### Heterogeneity of effects
- Significant negative effects of temperature on nighttime-lights growth are observed even in locations with relatively lower average temperatures.
- Statistically larger effects found in poorer areas, though significant effects occur across countries throughout the income distribution.

*International Monetary Fund — IMF WORKING PAPERS: High-Frequency Impact of Temperature on Economic Activity*

### 1.2 percentage points for low- and middle-income countries, although the effect is largely not statistically

### High-Frequency Impact of Temperature on Economic Activity

### Major findings: contemporaneous effects of temperature
- A 1 °C increase in temperature causes a contemporaneous 0.77 percent contraction in nighttime light intensity (estimate from fully specified model controlling for month fixed effects, cell × calendar-month fixed effects, and precipitation anomalies).
- Common estimates of the elasticity of nighttime lights with respect to GDP imply this would be associated with a reduction in GDP growth of between 0.50 and 0.77 percentage points.
- Higher-than-normal precipitation (excessive rainfall) is associated with a reduction in growth in nighttime lights; main temperature estimates are not sensitive to controlling for precipitation.
- Table 2 estimates (selected coefficients, standard errors clustered at province level):
  - Temperature (t): -0.0077 (0.0022) [***: p < 0.01] in the preferred specification.
  - Temperature (t-12): 0.0026 (0.0026) in the preferred specification.
  - High precipitation: -0.0202 (0.0069) [***: p < 0.01] in the preferred specification.
  - Number of observations: 8,438,699; Number of cells: 197,166.

### Heterogeneity by climate zone and income
- Using six temperature bins (less than 0 °C, 0 °C–10 °C, 10 °C–20 °C, 20 °C–30 °C, 30 °C–35 °C, and greater than 35 °C), the contemporaneous effect of a 1 °C increase varies by baseline temperature:
  - Approximately a 1 percent reduction in growth of nighttime lights for the temperature groups [10 °C, 20 °C) and [20 °C, 30 °C).
  - About a 0.3–0.45 percent reduction for the temperature groups [30 °C, 35 °C) and [>=35 °C).
- Controlling for temperature, income levels also matter: within each temperature bin, temperature has more negative effects in grid cells that are part of a low- or middle-income country.
- The literature cited finds mixed results on uneven temperature effects by baseline climate: Burke et al. (2015) reported positive impacts below 13 °C and negative above 13 °C, but subsequent work (Nath et al., 2023) shows those uneven effects disappear when controlling for lagged temperature. The paper emphasizes the need for granular analyses to disentangle hot-climate effects from low economic development effects.

### Data sources and key descriptive statistics
- Temperature and precipitation: ERA5 Monthly Aggregates dataset (ERA5 Monthly provides 2m air temperature converted to Celsius, monthly total precipitation; global coverage from January 1979 to June 2020; grid cell resolution about 30 kilometers).
- Nighttime lights: VIIRS Day/Night Band (DNB) monthly average radiance composite images (available January 2014 to June 2022; grid cell resolution about 500 meters). Observations with snow cover > 50 percent in a month are excluded.
- Analysis period for merged datasets: January 2014 to June 2020 (78 months) on 30-kilometer by 30-kilometer grid cells.
- Population data at grid-cell level from CIESIN and SEDAC, 2018; all data collected via Google Earth Engine.
- Selected summary statistics (Table 1; units percentage points except observations and number of cells):
  - Observations (cell × month): YoY Growth in Nighttime Lights 8,438,699; Temperature (°C) 12,517,650; YoY Change in Temperature (°C) 10,585,842; Total monthly precipitation (meter) 12,517,650.
  - Number of cells: 197,166 (YoY Growth in Nighttime Lights), 200,648 (other variables).
  - Mean YoY Growth in Nighttime Lights: 0.1167.
  - Median YoY Growth in Nighttime Lights: 0.0405.
  - Mean Temperature: 28.5310 °C.
  - Median Temperature: 30.3510 °C.
  - Temperature has increased by 0.10°C (or 0.46 percent) annually for the typical (median) cell.
  - Median year-on-year (YoY) growth rate of nighttime light is approximately 4 percent, with significant variation (94 percentage points).

### Theoretical and empirical framework
- Theoretical motivation builds on Dell et al. (2012): economic activity Yt = e^{βTt} A_t^{α}; temperature has a direct level effect β and can affect productivity growth via λ and σ. After log differencing, short-run growth effect is (β + αλ) and long-run effect is α(λ + σ).
- Main empirical contemporaneous specification (Eq. 4):
  - Dependent variable: log(L_{c,t}) − log(L_{c,t−12}) where L is nighttime lights.
  - Main regressors: T_{c,t} and T_{c,t−12} (monthly temperature and same month previous year); controls include precipitation anomaly indicators, month-specific fixed effects (F E_t), and cell × calendar-month fixed effects (F E_{c,m}).
  - Standard errors clustered at the province (first-tier administrative) level.
- Nonlinear and asymmetric specifications:
  - Separate impacts for positive vs. negative temperature shocks (Eq. 5) using Positive_{c,t} indicator for nonnegative temperature changes.
  - Separate impacts for large positive and large negative changes defined as deviations more than one standard deviation from the cell-specific average (Eq. 6).
- Dynamic effects estimated with local linear projections (Jordà, 2005) (Eq. 7) to obtain impulse responses of nighttime lights to a temperature increase over horizons up to 24 months; baseline controls include temperature and nighttime lights lags up to eleven months and forward values of temperature.

### Dynamic (persistence) results
- Local projection analysis shows that even a temporary shock to monthly temperatures has a relatively persistent effect on economic activity (proxied by nighttime lights), which remains statistically significant for almost a year, although the effect wanes as the time horizon expands.

*Source: IMF Working Paper — High-Frequency Impact of Temperature on Economic Activity (selected content).*

### Appendix Table 3 for the corresponding detailed estimation results).

### Appendix Table 3 for the corresponding detailed estimation results

### Main findings: contemporaneous effect of temperature on economic activity (nighttime lights)
- Preferred specification finding (reported in Conclusions): a 1 °C increase in temperature reduces the contemporaneous growth of nighttime lights by 0.77 percent.
- Appendix Table 3 (estimates corresponding to Figure 2) — temp(month, cell) - lag0 by temperature group:
  - [0.0, 10.0): 0.0062
  - [10.0, 20.0): -0.0104***
  - [20.0, 30.0): -0.0104***
  - [30.0, 35.0): -0.0044
  - >35.0: -0.0032
- Appendix Table 3 — temp(month, cell) - lag12 by temperature group:
  - [0.0, 10.0): 0.0101
  - [10.0, 20.0): 0.0065**
  - [20.0, 30.0): 0.0023
  - [30.0, 35.0): 0.0036
  - >35.0: -0.0065
- Note on significance notation: ***: p < 0.01, **: p < 0.05, *: p < 0.1.

### Heterogeneity by temperature bin and income level
- Across temperature bins, impacts are more negative in low- and middle-income countries than in high-income countries (Figure 4; Appendix Table 4).
- Appendix Table 4 — contemporaneous effect (temp(month, cell) - lag0) by temperature and income group (estimates shown as percentages):
  - [0 °C, 10 °C)
    - High income: 1.98%
    - Low & Middle income: 0.11%
  - [10 °C, 20 °C)
    - High income: -0.13%
    - Low & Middle income: -1.81%***
  - [20 °C, 30 °C)
    - High income: -0.85%*
    - Low & Middle income: -1.16%**
  - [30 °C, 35 °C)
    - High income: 1.06%*
    - Low & Middle income: -1.00%***
  - >35 °C
    - High income: 0.28%
    - Low & Middle income: -0.47%
- Interaction terms (temp(month, cell) - lag0 × I(Low & Middle Income) and × I(High Income)) reported in Appendix Table 4 indicate statistically significant differences favoring larger negative effects in Low & Middle income cells for several bins (e.g., -2.07%*** in [30 °C, 35 °C) interaction).

### Dynamic effects (local projections, 1 °C shock)
- After a 1 °C temperature shock, nighttime lights:
  - Decrease by on average 0.5 percent during the first quarter after the shock.
  - Exhibit a statistically significant reduction that persists for three quarters.
  - Become statistically insignificant in the fourth quarter after the shock.
- Temperature persistence after a monthly shock:
  - Temperatures remain elevated for 3 months (Figure 6).
- Interpretation: persistence in nighttime-lights response lasts longer than persistence in temperature itself, suggesting other channels (not temperature persistence alone) drive the persistence in economic activity response.

### Robustness checks and alternate specifications
- Table 3 (robustness models 1.1–1.5) — contemporaneous temperature coefficients (Temperature - lag0):
  - Model 1.1 (omit 2017): -0.0073**  (standard error (0.0029))
  - Model 1.2 (omit 1% lowest and 1% highest growth in nighttime lights): -0.0042**  (0.0019)
  - Model 1.3 (omit both 1% lowest/highest growth and 1% lowest/highest levels): -0.0043**  (0.0019)
  - Model 1.4 (omit 30-km grid cells with population < 100): -0.0066***  (0.0022)
  - Model 1.5 (no precipitation control): -0.0071***  (0.0022)
- Selected precipitation-related coefficients in Table 3:
  - High precipitation:
    - Model 1.1: -0.0281***  (0.0072)
    - Model 1.2: -0.0129**  (0.0057)
    - Model 1.3: -0.0131**  (0.0057)
    - Model 1.4: -0.0282***  (0.0062)
    - Model 1.5: N/A
  - Low precipitation:
    - Model 1.1: 0.0196**  (0.0098)
    - Model 1.2: 0.0031  (0.007)
    - Model 1.3: 0.0031  (0.007)
    - Model 1.4: 0.0141*  (0.0074)
    - Model 1.5: N/A
- Robustness summary:
  - The contemporaneous and dynamic relationships between nighttime lights and temperature are robust across the five alternative specifications (omitting 2017, trimming extremes, removing sparsely populated cells, and excluding precipitation control).
  - Additional results (available upon request per the source) confirm robustness to using monthly average rather than monthly maximum temperatures.

### Data and sample statistics (Appendix A)
- Appendix Table 1 — Number of observations, number of cells, number of countries by temperature group:
  - [0 °C, 10 °C): Number of observations 322,271; Number of cells 64,396; Number of countries 72
  - [10 °C, 20 °C): Number of observations 1,337,642; Number of cells 110,371; Number of countries 95
  - [20 °C, 30 °C): Number of observations 2,327,248; Number of cells 143,240; Number of countries 180
  - [30 °C, 35 °C): Number of observations 2,332,987; Number of cells 125,112; Number of countries 166
  - >35 °C: Number of observations 2,118,551; Number of cells 85,935; Number of countries 136
- Appendix Table 3 — sample sizes used in the temperature-group contemporaneous estimates (repeated):
  - Number of observations by bin: 322,271; 1,337,642; 2,327,248; 2,332,987; 2,118,551
  - Number of cells by bin: 64,396; 110,371; 143,240; 125,112; 85,935
- Table 3 (robustness models) — sample sizes and cells:
  - Number of observations across Models 1.1–1.5: 7,039,914; 8,269,925; 8,157,719; 7,127,418; 8,438,699
  - Number of cells across Models 1.1–1.5: 197,103; 197,157; 196,502; 145,627; 197,166

### Key implications and context (from Conclusions)
- The study uses a new dataset and higher geographic/temporal resolution to estimate temperature impacts on economic activity proxied by nighttime lights.
- Main implications:
  - A negative and highly statistically significant effect of temperature on economic activity is found, larger than often reported in the literature.
  - Negative and statistically significant effects prevail except in colder areas; warmer areas exhibit more negative impacts.
  - Within temperature bins, poorer areas (low- and middle-income countries) suffer relatively more severe effects.
  - Historical short-run responses may differ from future long-run climate-change impacts due to magnitude, non-linearities, tipping points, and adaptation limits.
  - Stronger adaptation (e.g., drought-resistant seeds, diversification into resilient sectors) may soften impacts; limits to adaptation and large-scale behavioral changes (e.g., migration) could amplify future damages.

*Source: IMF Working Paper “High-Frequency Impact of Temperature on Economic Activity” — Appendix tables and figures as provided in the supplied content.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024178-print-pdf.pdf_