## The Economic Costs of Temperature Uncertainty (wpiea2025026-print-pdf)

## Source details

**Canonical URL:** [The Economic Costs of Temperature Uncertainty (wpiea2025026-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025026-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025026-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025026-print-pdf.pdf.json)

---

### Major findings and quantitative outcomes
- Climate-related world-wide economic costs may arise to over 125 trillion dollars by 2050 if climate policy actions are not intensified.
- Temperature volatility is on the rise globally and in the US has increased even more than the average temperature in the past 30 years.
- A 1-degree Celsius increase in temperature volatility causes:
  - a reduction in firms’ investment of 0.4 percent after four quarters.
  - a reduction in firms’ investment of about 1.4 percent after 20 quarters.
- No statistically significant effects on firms’ investment were found for:
  - increases in average temperature levels,
  - the occurrence of heat or cold waves,
  - initial levels of temperature volatility.
- Broader negative effects of temperature volatility on firms’ outcomes:
  - sales, labor productivity, and employment decline.
  - Using Metropolitan Area employment and unemployment data produces effects more than double those obtained with publicly listed firms data.
- Cross-study benchmark:
  - Kotz et al. (2021) find that an extra degree of intra-annual temperature variability reduces average regional growth by five percentage points (across 1,537 larger subnational regions in 77 countries over 40 years).

### Data sources, scope, and key sample sizes
- Temperature data:
  - EU Copernicus satellite and ground-based measurements at ~27.64km x 27.64km at the equator.
  - US focus for period 1961-2018.
  - Aggregated to US zip code level and collapsed to quarterly/monthly/yearly measures as needed (within-quarter standard deviation of daily temperatures is baseline measure).
- Firm-level data:
  - Compustat quarterly data for approximately 33 thousand firms, period 1961Q1-2018Q4 (baseline capital expenditure analysis).
  - Final quarterly dataset: approximately 1.3 million observations at the firm/quarter level, 1961Q1-2018Q4.
  - Annual-frequency firm employment and labor productivity are used (employment data available yearly).
- Metropolitan Statistical Area (MSA) data:
  - Bureau of Labor Statistics monthly data from 1/1990 to 1/2022 for 373 metropolitan areas.
  - Merged dataset: 143,605 observations (monthly MSA statistics with monthly standard deviation and average daily temperatures).
- Power outage data:
  - EAGLE-I Power Outage Data (Brelsford et al. 2024) covering 2014–2022, 3,243 counties.
  - Merged and collapsed to monthly county/month panel: approximately 350 thousand observations.
- Teachers’ sickness days (labor productivity proxy):
  - Illinois State Board of Education data on sickness days for public-school teachers, 2013–2020.
  - Final school-district × year dataset: approximately 6.8 thousand observations (849 school districts × 8 years).
- Firm uncertainty and risk exposure measures:
  - Firm realized volatility from annualized 12-month standard deviation of daily CRSP returns (1992-2019) for ca. 10 thousand US listed firms (Alfaro et al., 2021).
  - Textual analysis measures from earnings conference-call transcripts for 7,357 US listed firms (2002-2016) and a climate-exposure measure from Sautner et al. (2023) at quarterly frequency (2002-2020).

### Empirical approach and identification
- Baseline empirical framework:
  - Local projection framework (Jorda 2005) to estimate impulse-response functions of firms’ outcomes to temperature volatility shocks.
  - Baseline dependent variable: cumulative percent variation of (log) firms’ capital expenditure between t+k and t-1, with k = 0,...,20 (quarters).
  - Baseline measure of temperature uncertainty: standard deviation of daily temperatures within quarter at the firm-location/quarter level.
  - Controls include four lags of the dependent variable, temperature volatility, and average temperatures at firm/quarter level (l = 1,...,4). Robustness controls: min and max temperatures, number of heat (≥ 35-degree Celsius) and cold (≤ 0-degree Celsius) days.
  - Fixed effects: firm fixed effects (α_f^k) and quarter fixed effects (γ_t^k).
  - Estimated coefficients β_k indicate percent variation in capital expenditure (t-1 to t+k) in response to a 1-degree Celsius increase in temperature volatility at time t-1.
- Identification strengths:
  - High geographical granularity (zip-code level) with firm fixed effects reduces reverse causality concerns.
  - Long time span (1961–2018) and quarterly frequency mitigate selection and contemporaneous endogeneity issues.
  - Use of unexpected temperature residuals (from regressions of daily temperatures on date fixed effects) for robustness.

### Transmission channels tested and empirical evidence
- Energy disruptions and higher energy costs:
  - Temperature volatility increases frequency of power outages (EAGLE-I data), disrupting production and supply chains.
  - County-level monthly analysis shows a 1-degree Celsius rise in temperature volatility increases the number of power outages by 9 percent per month.
- Labor productivity reductions:
  - Increased sickness and absenteeism (teachers’ sickness-days data) reduce labor productivity even for occupations primarily indoors.
  - A 1-degree Celsius increase in temperature volatility is associated with an increase in the number of sick days per contracted days of about 8 percentage points (school-district/year analysis).
- Financial constraints and uncertainty:
  - Temperature volatility raises firms’ realized stock market volatility and exposure to environment-related and non-political risks (textual analysis), which attenuates investment—effects larger for financially constrained firms.
  - Using annual firm realized volatility measures, a 1-degree Celsius rise in temperature volatility increases firms’ uncertainty by approximately 1.5 percent.
  - Quarterly textual measures: temperature volatility increases firm-level exposure to environmental risks by +1.3 percent and to non-political risks by +1.2 percent.

### Heterogeneity across sectors and firm characteristics
- Sectoral heterogeneity:
  - Effect on investment is larger for firms operating in manufacturing than in services.
  - Larger negative effects in heat-sensitive industries (NAICS: 311, 3122, 322, 3361, 2122, 48, 23, 221).
  - Temperature volatility has statistically significant effects on investment in both heat-sensitive and non-sensitive sectors, larger and more precisely estimated for heat-sensitive sectors.
- Firm-level heterogeneity (financial constraints amplify effects):
  - Five firm characteristics considered: age, size, share of liabilities in short-term maturities, share of liquid assets, and Tobin’s Q.
  - Smooth transition modeling (γ=1.5, F(z)=0.5 cutoff) interpretation:
    - β_L = low-regime (young, small, low liquidity, low maturity, low Tobin’s Q).
    - β_H = high-regime (old, large, high liquidity, high maturity, high Tobin’s Q).
  - Quantitative effects:
    - A 1-degree Celsius increase in temperature volatility leads to medium-term investment losses of about 4 percent for firms with a low Tobin’s Q.
    - The same shock leads to medium-term investment losses of about 2½ for small and young firms, and for firms with lower liquidity and higher share of short-term debt maturity.
  - Tobin’s Q and size are the most statistically significant firm characteristics shaping investment losses due to temperature volatility.

### Robustness and complementary analyses
- Alternative measures of temperature volatility all confirm the main result:
  - (i) log of the standard deviation of temperature;
  - (ii) difference between max and min temperatures;
  - (iii) residuals from regressing temperatures on date fixed effects (“unexpected” temperature volatility).
- Lag parametrization and clustering:
  - Impulse response functions robust to alternative lags (1/2; 1/3; and 0/4) and to standard errors clustered at zip code level.
  - Excluding lags of the dependent variable confirms baseline results.
- Distributional and sample robustness:
  - Winsorizing the dependent variable to range between the 1st and the 99th percentiles does not affect results.
  - Sub-sample starting in 2000 yields identical results.
- Additional controls:
  - Including time-varying firm characteristics (age, size, liquidity, debt maturity and Tobin’s Q), firm-specific time trends, firm-quarter dummies, firm-region dummies, and region-specific time trends confirm main findings.
- Quarterly impulse responses computed for horizons k = 0,...,20 with robust standard errors clustered at the firm/location level.
- Temperature volatility persistence: correlation with one-quarter lag ≈ 0.31 and with four-quarter lag ≈ 0.83.

### Alternative economic outcomes and magnitude comparisons
- Investment ratio: a 1-degree Celsius increase in temperature volatility causes a reduction in the investment ratio of about 1 percentage point after 20 quarters.
- Firms’ sales: a 1-degree Celsius increase in temperature volatility reduces sales by about 0.1 percent in the first year and about 0.5 percent 20 quarters after.
- Employment and labor productivity (annual Compustat-based measures):
  - Employment: a 1-degree Celsius increase in temperature volatility causes a medium-term reduction in employment of about 0.7 percent after 5 years.
  - Productivity: a 1-degree Celsius increase in temperature volatility causes a medium-term reduction in productivity of about 1.2 percent after 5 years.
- Annual-frequency replication: magnitudes are larger—about 5 percent for investment and 2 percent for sales after 5 years.
- MSA analysis:
  - Monthly MSA data: a 1-degree Celsius increase in temperature volatility causes a medium-term reduction in monthly employment of about 0.06 percent after 60 months.
  - The same shock causes an increase in the unemployment rate of about 0.02 percentage point after 60 months.
  - Annual MSA data: a 1-degree Celsius increase reduces employment by about 1.7 percent and increases the unemployment rate by about 0.6 percentage point after 5 years.
  - Employment effect from MSAs is more than double that from Compustat, indicating Compustat-based estimates likely under-estimate true negative effects.

### Potential non-linearities and spatial heterogeneity
- Interaction with volatility quartiles:
  - Regression with interactions across quartiles (<=25th; 25th–50th; 50th–75th; >75th) shows point estimates typically higher when volatility is initially high (4th quartile), but effects are not statistically different across quartiles.
- Heterogeneity by subnational increases in volatility:
  - Dummies based on evolution of temperature volatility between 1990 and 2018 (quartiles of ∆TeV) find no statistically significant differences across coefficients.

### Mechanisms summary and policy-relevant conclusions
- Identified channels through which temperature volatility reduces economic activity:
  - Increasing power/energy disruptions.
  - Reducing labor productivity (sickness and absenteeism).
  - Increasing firms’ riskiness and economic uncertainty, leading to reduced investment—effects amplified for financially constrained firms.
- Policy-relevant implications drawn in the analysis:
  - Existing literature focusing on average temperature changes or extreme events may underestimate true economic costs because climate variability (temperature volatility) is an important channel.
  - The negative impact of climate change on firms’ economic activity in the US is driven by temperature fluctuations rather than average temperature changes (though average temperature effects may be significant in developing economies).
  - Financially constrained firms suffer disproportionately, suggesting targeted resilience and financial support policies could mitigate amplification effects.

*Italic source: wpiea2025026-print-pdf*

### REFERENCES .............................................................................................................

### REFERENCES

### Major findings and quantitative outcomes
- Climate-related world-wide economic costs may arise to over 125 trillion dollars by 2050 if climate policy actions are not intensified.
- Temperature volatility is on the rise globally and in the US has increased even more than the average temperature in the past 30 years.
- A 1-degree Celsius increase in temperature volatility causes:
  - a reduction in firms’ investment of 0.4 percent after four quarters.
  - a reduction in firms’ investment of about 1.4 percent after 20 quarters.
- No statistically significant effects on firms’ investment were found for:
  - increases in average temperature levels,
  - the occurrence of heat or cold waves,
  - initial levels of temperature volatility.
- Evidence of broader negative effects of temperature volatility on firms’ outcomes:
  - sales, labor productivity, and employment decline.
  - Using Metropolitan Area employment and unemployment data produces effects more than double those obtained with publicly listed firms data.
- Cross-study benchmark: Kotz et al. (2021) find that an extra degree of intra-annual temperature variability reduces average regional growth by five percentage points (across 1,537 larger subnational regions in 77 countries over 40 years).

### Data sources, scope, and key sample sizes (preserve original temporal and spatial precision)
- Temperature data:
  - EU Copernicus satellite and ground-based measurements at ~27.64km x 27.64km at the equator.
  - US focus for period 1961-2018.
  - Aggregated to US zip code level and collapsed to quarterly/monthly/yearly measures as needed (within-quarter standard deviation of daily temperatures is baseline measure).
- Firm-level data:
  - Compustat quarterly data for approximately 33 thousand firms, period 1961Q1-2018Q4 (baseline capital expenditure analysis).
  - Final quarterly dataset: approximately 1.3 million observations at the firm/quarter level, 1961Q1-2018Q4.
  - Annual-frequency firm employment and labor productivity are used (employment data available yearly).
- Metropolitan Statistical Area (MSA) data:
  - Bureau of Labor Statistics monthly data from 1/1990 to 1/2022 for 373 metropolitan areas.
  - Merged dataset: 143,605 observations (monthly MSA statistics with monthly standard deviation and average daily temperatures).
- Power outage data:
  - EAGLE-I Power Outage Data (Brelsford et al. 2024) covering 2014–2022, 3,243 counties.
  - Merged and collapsed to monthly county/month panel: approximately 350 thousand observations.
- Teachers’ sickness days (labor productivity proxy):
  - Illinois State Board of Education data on sickness days for public-school teachers, 2013–2020.
  - Final school-district × year dataset: approximately 6.8 thousand observations (849 school districts × 8 years).
- Firm uncertainty and risk exposure measures:
  - Firm realized volatility from annualized 12-month standard deviation of daily CRSP returns (1992-2019) for ca. 10 thousand US listed firms (Alfaro et al., 2021).
  - Textual analysis measures of environment-related and non-political risks from earnings conference-call transcripts for 7,357 US listed firms (2002-2016) and a climate-exposure measure from Sautner et al. (2023) at quarterly frequency (2002-2020).

### Empirical approach and identification
- Baseline empirical framework:
  - Local projection framework (Jorda 2005) to estimate impulse-response functions of firms’ outcomes to temperature volatility shocks.
  - Baseline dependent variable: cumulative percent variation of (log) firms’ capital expenditure between t+k and t-1, with k = 0,...,20 (quarters).
  - Baseline measure of temperature uncertainty: standard deviation of daily temperatures within quarter at the firm-location/quarter level.
  - Controls include four lags of the dependent variable, temperature volatility, and average temperatures at firm/quarter level (l = 1,...,4). Additional robustness controls: min and max temperatures, number of heat (≥ 35-degree Celsius) and cold (≤ 0-degree Celsius) days.
  - Fixed effects: firm fixed effects (α_f^k) and quarter fixed effects (γ_t^k).
  - Estimated coefficients β_k indicate percent variation in capital expenditure (t-1 to t+k) in response to a 1-degree Celsius increase in temperature volatility at time t-1.
- Identification strengths emphasized:
  - High geographical granularity (zip-code level) with firm fixed effects to control for time-invariant heterogeneity and reduce reverse causality concerns at the zip-code-quarter scale.
  - Long time span (1961–2018) and quarterly frequency mitigate selection and contemporaneous endogeneity issues.
  - Use of unexpected temperature residuals (from regressions of daily temperatures on date fixed effects) for robustness.

### Transmission channels tested and empirical evidence
- Three primary channels examined with granular data:
  - Energy disruptions and higher energy costs:
    - Temperature volatility increases frequency of power outages (EAGLE-I data), disrupting production and supply chains.
  - Labor productivity reductions:
    - Increased sickness and absenteeism (teachers’ sickness-days data) lower labor productivity even for occupations primarily indoors.
  - Financial constraints and uncertainty:
    - Temperature volatility raises firms’ realized stock market volatility and exposure to environment-related and non-political risks (textual analysis of earnings calls), which attenuates investment—effects larger for financially constrained firms.
- Empirical findings highlight that all these channels play a key role in driving the effect of temperature uncertainty on economic outcomes.

### Heterogeneity of effects across sectors and firm characteristics
- Sectoral heterogeneity:
  - The effect of temperature volatility on investment is larger for firms operating in manufacturing than in services.
  - Larger negative effects in heat-sensitive industries (as identified by Graff-Zivin and Neidell 2014).
- Firm-level heterogeneity:
  - Effects are larger for firms that face higher financial constraints: smaller firms, younger firms, and firms with a higher share of short-term liabilities.
  - Publicly listed firms’ estimates likely understate the true economy-wide effect, given larger impacts found using MSA employment/unemployment data.

### Robustness and complementary analyses
- Alternative measures of temperature volatility used in robustness checks: difference between maximum and minimum temperatures, log of the standard deviation, and within-quarter statistics of temperature residuals (mean, min, max, standard deviation) to capture unexpected temperature changes.
- Results are robust to a battery of sensitivity checks and alternative specifications.
- Quarterly impulse responses are computed for horizons k = 0,...,20 with robust standard errors clustered at the firm/location level.

*Source: IMF Working Paper excerpt (REFERENCES, APPENDIX and selected sections from The Economic Costs of Temperature Uncertainty).*

### 0.4  percent  in  the  first  year  and  about  1.4  percent  20  quarters  after.   This  effect  is  not  only  highly

### The Economic Costs of Temperature Uncertainty

### A. Main empirical findings on firms’ investment
- A 1-degree Celsius increase in temperature volatility reduces firms’ investment by 0.4 percent in the first year and about 1.4 percent 20 quarters after.
- The increase in temperature volatility observed in the US between 2000 and 2018 (from a standard deviation of 4.39 to 5.55 Celsius) is estimated to have led to a drop in firms’ investment by 1.6 percent in the medium term—that is, about 20USD billion for the entire US economy.
- The framework controls for average temperature at the firm location and finds:
  - The effect of average temperature is not statistically different from zero for any considered time horizons.
  - Excluding average temperature as a control does not affect the economic impact of temperature volatility.
- Controlling for extreme temperatures:
  - When controlling for minimum temperatures, the effect of a 1-degree Celsius increase in temperature uncertainty is about -0.15 percent in the first year and about -0.7 percent 20 quarters after.
  - When controlling for maximum temperatures, the effect is about -0.2 percent in the first year and about -1.5 percent 20 quarters after.
  - A 1-degree Celsius decline in minimum temperatures is associated with a reduction in investment by 0.1 percent in the first year and about 0.35 percent 20 quarters after.

### B. Robustness checks and specification tests
- Temperature volatility persistence: correlation with one-quarter lag ≈ 0.31 and with four-quarter lag ≈ 0.83.
- Alternative measures of temperature volatility all confirm the main result:
  - (i) log of the standard deviation of temperature;
  - (ii) difference between max and min temperatures;
  - (iii) residuals from regressing temperatures on date fixed effects (“unexpected” temperature volatility).
- Lag parametrization and clustering:
  - Impulse response functions robust to alternative lags (1/2; 1/3; and 0/4) and to standard errors clustered at zip code level.
  - Excluding lags of the dependent variable confirms baseline results.
- Distributional and sample robustness:
  - Winsorizing the dependent variable to range between the 1st and the 99th percentiles does not affect results.
  - Sub-sample starting in 2000 yields identical results.
- Additional controls:
  - Including time-varying firm characteristics (age, size, liquidity, debt maturity and Tobin’s Q), firm-specific time trends, firm-quarter dummies, firm-region dummies, and region-specific time trends confirm main findings.
- Forward periods of temperature volatility included (following Ciminelli et al. (2022)) do not change results.

### C. Alternative economic outcomes (sales, employment, productivity)
- Investment ratio (dependent variable instead of log investment): a 1-degree Celsius increase in temperature volatility causes a reduction in the investment ratio of about 1 percentage point after 20 quarters.
- Firms’ sales: a 1-degree Celsius increase in temperature volatility reduces sales by about 0.1 percent in the first year and about 0.5 percent 20 quarters after.
- Employment and labor productivity (annual Compustat-based measures):
  - A 1-degree Celsius increase in temperature volatility causes a medium-term reduction in employment of about 0.7 percent after 5 years.
  - A 1-degree Celsius increase in temperature volatility causes a medium-term reduction in productivity of about 1.2 percent after 5 years.
- Annual-frequency replication:
  - Using annual data, magnitudes are larger: about 5 percent for investment and 2 percent for sales after 5 years.

### D. Broader labor-market and sample-coverage evidence
- Limitations of Compustat acknowledged:
  - Covers only publicly listed firms (may underestimate economy-wide effect if small firms are more affected).
  - Uses firm headquarters locations (may understate effects if corporate offices face different volatility).
- Metropolitan-area (MSA) analysis using monthly employment and unemployment data:
  - A 1-degree Celsius increase in temperature volatility causes a medium-term reduction in monthly employment of about 0.06 percent after 60 months.
  - The same shock causes an increase in the unemployment rate of about 0.02 percentage point after 60 months.
  - Using annual MSA data, a 1-degree Celsius increase reduces employment by about 1.7 percent and increases the unemployment rate by about 0.6 percentage point after 5 years.
  - Employment effect from MSAs is more than double that from Compustat, indicating Compustat-based estimates likely under-estimate true negative effects.

### E. Potential non-linearities and spatial heterogeneity
- Interaction with volatility quartiles:
  - Regression allowing for temperature volatility interactions with quartile dummies (<=25th; 25th–50th; 50th–75th; >75th) shows point estimates typically higher when volatility is initially high (4th quartile), but effects are not statistically different across quartiles.
- Heterogeneity by subnational increases in volatility:
  - Constructing dummies based on the evolution of temperature volatility between 1990 and 2018 (quartiles of ∆TeV) finds no statistically significant differences across coefficients.

### F. Identified channels through which temperature volatility affects firms
- Channel 1 — Increased firm-level economic uncertainty:
  - Using annual firm realized volatility (annualized 12-month standard deviation of daily CRSP returns) and yearly temperature volatility, a 1-degree Celsius rise in temperature volatility increases firms’ uncertainty by approximately 1.5 percent.
  - Quarterly measures using Hassan et al. (2019) indexes show temperature volatility increases firm-level exposure to environmental risks by +1.3 percent and to non-political risks by +1.2 percent (both highly statistically significant).
  - Sautner et al. (2023) firm-level climate exposure indexes: temperature uncertainty positively affects both exposure and associated potential risks with similar magnitudes.
  - Overall: temperature uncertainty increases firms’ riskiness, which is negatively correlated with firms’ performance, particularly investment.
- Channel 2 — Infrastructure damages and power outages:
  - County-level monthly analysis using customers affected by power outages shows that a 1-degree Celsius rise in temperature volatility increases the number of power outages by 9 percent per month.
- Channel 3 — Worker sickness and absenteeism:
  - The study proceeds to test this channel using a dataset on yearly number of sickness days for public school teachers from the Illinois State Board of Education covering 2013- (analysis continued beyond supplied excerpt).

*Source: IMF Working Paper — The Economic Costs of Temperature Uncertainty*

### 2020. First, we collapse teacher-level data at the school-district level at yearly frequency, for a total of 849 school-

### The Economic Costs of Temperature Uncertainty

### Methodology: teacher- and school-district-level analysis
- Collapsed teacher-level data at the school-district level at yearly frequency for a total of 849 school-districts.
- Matched school-district/year observations with yearly temperatures (average and fluctuations) using school-district shapefiles.
- Estimated the specification (Equation (6)) with school-district (훼훼_sd) and year (훼훼_t) fixed effects, controlling for lagged temperature volatility (TTV_sd,t−1), lagged average temperature, and the number of contracted days for each full-time employed teacher entering vector X_sd,t−1.
- Dependent variable: percentage point change in the number of sick days per contracted days for each school-district/year (푦_sd,t − 푦_sd,t−1).

### Key finding: temperature volatility and sick days
- A 1-degree Celsius increase in temperature volatility is associated with an increase in the number of sick days per contracted days of about 8 percentage points (Table 2, Column VII).

### Heterogeneity across sectors and firms — sectoral analysis
- Split sample between firms operating in manufacturing and services.
- Firms operating in manufacturing and services represent approximately the 30% and the 10% of the sample, respectively.
- Findings:
  - Temperature uncertainty has statistically insignificant effects for firms in the service sector.
  - Temperature uncertainty has negative, statistically significant, and persistent effects for manufacturing firms (Figure 6, Panels A and B).
- Heat-sensitive industries (per Graff-Zivin and Neidell, 2014) identified; dummy HS = 1 for heat-sensitive sectors (NHS = 1 − HS for non-heat-sensitive).
  - Heat-sensitive sectors include NAICS: food and tobacco (311 and 3122), paper manufacturing (322), motor vehicle manufacturing (3361), metal and mining (2122), transport (48), construction (23), utilities (221).
- Results (Figure 7):
  - Temperature volatility has statistically significant effects on firms’ investment in both heat-sensitive and non-sensitive sectors, but the effect is larger and more precisely estimated for heat-sensitive sectors.

### Firm-level analysis: financial constraints amplify effects
- Considered five firm characteristics proxying financial constraints: age, size, share of liabilities in short-term maturities, share of liquid assets, and Tobin’s Q.
- Modeled nonlinear responses via a smooth transition function F(z_f) with γ=1.5 and F(z_f)=0.5 as the cut-off between low and high regimes; normalization of z uses the average firm characteristic over the entire sample to reduce endogeneity.
- Interpretation:
  - β_L refers to the low-regime case (young, small, low liquidity, low maturity, low Tobin’s Q)—when F(z) ≈ 1.
  - β_H refers to the high-regime case (old, large, high liquidity, high maturity, high Tobin’s Q)—when (1 − F(z)) ≈ 1.
- Quantitative effects:
  - A 1-degree Celsius increase in temperature volatility leads to medium-term investment losses of about 4 percent for firms with a low Tobin’s Q.
  - The same 1-degree Celsius increase leads to medium-term investment losses of about 2½ for small and young firms, and for firms characterized by lower liquidity and with higher share of short-term debt maturity (Figure 8).
  - The difference between regimes is statistically significant for all considered time horizons.

### Robustness checks and additional specification details
- Robustness exercises:
  - Used initial sample value for each firm characteristic in normalization to further reduce endogeneity.
  - Constructed dummy variables equal to 1 if firm-level characteristics exceed the sample average and interacted them with temperature volatility (Equation (9)); results confirm previous findings though regime differences are less clear-cut for some specifications (Figure A23-A24).
- Alternative specification including simultaneous interactions of smooth transition functions for each firm characteristic:
  - Considered the regime expected to amplify effects (low age, low size, low liquidity, low maturity, low Tobin’s Q) with concrete thresholds:
    - Younger firms: age lower than 4 quarters.
    - Smaller firms: (log of) total assets lower than 2.
    - Low-liquidity firms: liquidity lower than 0.03.
    - Short-term maturity firms: ratio of current liabilities to total liabilities lower than 0.2.
    - Low Tobin’s Q firms: Tobin’s Q lower than 1.
  - Results reported in Table A2 indicate Tobin’s Q and size are the most statistically significant firm characteristics in shaping investment losses due to temperature volatility.

### Mechanisms and broader findings
- Identified channels through which temperature volatility reduces economic activity:
  - Increasing power/energy disruptions.
  - Reducing labor productivity.
  - Adding to the riskiness and economic uncertainty firms face.
- Firms facing financial constraints are more exposed to temperature volatility, consistent with the financial-amplification literature.

### Conclusions
- Existing literature focusing on average temperature changes or extreme events may underestimate true economic costs because climate variability (temperature volatility) is an important channel.
- Two contributions of the analysis:
  - The negative impact of climate change on firms’ economic activity in the US is caused by temperature fluctuations rather than average temperature changes (though average temperature effects may be significant in developing economies).
  - Temperature volatility reduces economic activity through power/energy disruptions, lower labor productivity, and higher uncertainty; financially constrained firms suffer disproportionately.

*Source: IMF Working Paper — The Economic Costs of Temperature Uncertainty*

### References

### References (Content Unit)

### Major themes and topical coverage
- Climate impacts on economic activity and production:
  - "The Effects of Weather Shocks on Economic Activity: What are the Channels of Impact?" Acevedo et al., 2020.
  - "Global non-linear effect of temperature on economic production." Burke, Hsiang, and Miguel, 2015. Nature, 527, 235-239.
  - "Temperature impacts on economic growth warrant stringent mitigation policy." Moore and Diaz, 2015. Nature Climate Change 5, 127.
  - "Estimating economic damage from climate change in the United States." Hsiang et al., 2017. Science, 356(6345), 1362-1369.
  - "The impact of climate conditions on economic production. Evidence from a global panel of regions." Kalkuhl and Wenz, 2020. J. Environ. Econ. Manage. 103, 102360.
- Temperature volatility, variability, and extreme temperatures:
  - "Temperature variability implies greater economic damages from climate change." Calel et al., 2020. Nature Communication 11, 5028.
  - "Day-to-day temperature variability reduces economic growth." Kotz et al., 2021. Nature Climate Change 11, 319–325.
  - "Temperature fluctuations in a changing climate: an ensemble-based experimental approach." Vincze, Borcia, and Harlander, 2017. Sci Rep 7, 254.
  - "Influence of extreme weather and climate change on the resilience of power systems: Impacts and possible mitigation strategies." Panteli and Mancarella, 2015. Electric Power Systems Research, Volume 127, 259-270.
- Firm-level and microeconomic responses to climate and temperature shocks:
  - "Adapting to climate change: Long-term effects of drought on local labor markets." Bastos, Busso, and Miller, 2013. IDB Working Paper Series IDB-WP-466.
  - "Taken by storm: business financing and survival in the aftermath of Hurricane Katrina." Basker and Miranda, 2018. Journal of Economic Geography 18, 1285–1313.
  - "Assessing the impact of energy prices on plant-level environmental and economic performance: Evidence from Indonesian manufacturers." Brucal and Dechezleprêtre, 2021. OECD.
  - "Firm-level climate change exposure." Sautner et al., 2023. Journal of Finance, 78(3):1449-1498.
  - "Firm-Level Political Risk: Measurement and Effects." Hassan et al., 2019. The Quarterly Journal of Economics, 134(4).
- Uncertainty, volatility, and macro/financial channels:
  - "The Impact of Uncertainty Shocks." Bloom, 2009. Econometrica, 77.
  - "Fluctuations in Uncertainty." Bloom, 2014. The Journal of Economic Perspectives, 63.
  - "The Finance Uncertainty Multiplier." Alfaro, Bloom, and Lin, 2022. NBER Working Paper w24571.
  - "Financial uncertainty and real activity: The good, the bad, and the ugly." Caggiano et al., 2021. European Economic Review. Volume 136.
  - "Risk matters: the real effects of volatility shocks." Fernández-Villaverde et al., 2011. Am. Econ. Rev. 101, 2530–2561.
  - "The Macroeconomic Cost of Climate Volatility." Mumtaz and Alessandri, 2021. Available at SSRN.
- Adaptation, sectoral reallocation, and technological responses:
  - "Adaptation to Cyclone Risk: Evidence from the Global Cross-Section." Hsiang and Narita, 2012.
  - "The Marginal Product of Climate." Deryugina and Hsiang, 2017. NBER Working Paper 24072.
  - "Contribution of air conditioning adoption to future energy use under global warming." Davis and Gertler, 2015. Proceedings of the National Academy of Sciences 112, 5962–5967.
  - "Corporate climate risk: measurements and responses." Li et al., 2020. SSRN Electron J.
- Methods and econometric approaches relevant to the study:
  - "Estimation and inference of impulse responses by local projections." Jordà, 2005. American Economic Review, 95(1), 161–182.
  - "The Methods of Growth Econometrics." Durlauf, Johnson, and Temple, Palgrave Handbook of Econometrics, 2009.
  - "Modelling Non-Linear Economic Relationships." Granger & Terasvirta, 1993. OUP Catalogue.
  - "Is Economic recovery a Myth? Robust Estimation of Impulse Response." Teulings and Zubanov, 2013. Journal of Applied Econometrics, 29(3): 497-514.
  - Monte Carlo and robustness checks: Hauk and Wacziarg, 2009. "A Monte Carlo Study of Growth Regressions," Journal of Economic Growth 14 (2009), 103-147.
- Policy and climate economics overviews:
  - "How to Mitigate Climate Change." IMF, 2019. Fiscal Monitor. Washington, October.
  - "The economics of climate change." Stern, 2008. American Economic Review, 98, 2, 1-37.
  - "Climate change: The ultimate challenge for economics." Nordhaus, 2019. American Economic Review, 109, 6, 1991-2014.
  - "What’s the cost of Net Zero? Net Zero Financing Roadmaps." UNFCCC, November 3, 2021.

### Notes on the structure of the cited literature (as reflected in the source)
- The reference list includes peer-reviewed journal articles, NBER Working Papers, IMF and UNFCCC policy documents, IDB and OECD working papers, SSRN preprints, and methodological texts.
- Several references focus explicitly on temperature volatility, temperature variability, and non-linear temperature effects on production and growth (explicit phrases preserved from titles and journal entries).
- Firm-level studies cited address investment, financing, capital destruction, labor markets, and firm risk exposures in response to climatic shocks and energy price changes.
- Methodological references emphasize impulse response estimation, local projections, non-linear modelling, Monte Carlo studies, and robustness checks.

*Italic source: wpiea2025026-print-pdf - References (extracted content provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025026-print-pdf.pdf_