## 2.1–4.4 Climate news; Country-level and macro data; Official news (sections 2.1, 2.3, 4.4)

## Source details

**Canonical URL:** [2.1–4.4 Climate news; Country-level and macro data; Official news (sections 2.1, 2.3, 4.4)](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025037-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025037-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025037-print-pdf.pdf.json)

---

### Scope, data coverage, and construction of indices
- Coverage:
  - Approximately 30 million news articles published between January 2000 and July 2023 from media outlets in Brazil, Mexico, Chile, Peru, Colombia, Costa Rica, Ecuador, and Uruguay.
  - Official government gazettes examined for 2017-2023 for the eight countries.
  - Media source: Dow Jones Factiva.
  - Each country’s gazette publication frequency ranges from 740 publications on average per month in Chile to 25 in Brazil.
  - Climate Policy Database used to cross-check official news indicators.
- Climate news indices:
  - Daily country-level indicators from January 2000 to July 2023 using climate-related bi-grams from Sautner et al. (2023a).
  - Standardization: frequency standardized by the total number of articles published each month in country j (monthly aggregation for plots).
  - Aggregate formula as reported:
    - Climate News_{j,t} = (1 / N_{j,m}) * sum_{b}^{B} (1[b ∈ C])
      - where b is a bi-gram, 1 is indicator, C is set of bi-grams, N_{j,m} is total number of articles published each month in country j.
  - Climate Risk News index conditions bi-grams to co-occur with "risk", "uncertainty" or synonyms:
    - Climate Risk News_{j,t} = (1 / N_{j,m}) * sum_{b}^{B} (1[b ∈ C]) × 1[b,r ∈ S]
      - where r is a risk-related term, S is set of synonyms.
- Subcategory indices constructed daily for each country:
  1. Regulatory — climate-related regulatory news.
  2. Physical — physical climate events and natural disasters.
  3. Opportunity — climate-related technological developments and opportunities.
- Official-gazette indices:
  - Count bi-grams in official gazettes (2017–2023) for aggregate, opportunity, and physical news; uncertainty terms rare in gazettes so Climate Risk and Regulatory indices not constructed from official gazettes.
  - Major Climate Policy Events binary indicator: manually verified legally binding laws, decrees, regulations; 57 events detected (Table A9).

### Media landscape and linkage to firm-level data
- Media concentration (largest publishing house share of total articles):
  - Brazil: 9 percent
  - Colombia: 9 percent
  - Chile: around 15 percent
  - Mexico: around 15 percent
  - Peru: 20 percent
  - Ecuador: 20 percent
  - Uruguay: 30 percent
  - Costa Rica: 50 percent
- Number of sourced articles by country: exceeds 1/2 a million even in smaller economies.
- Firm-level dataset linkage:
  - 628 publicly listed firms from 2000 (firm financial variables, emissions, ESG data; country-level controls described in sections 2.2 and later).
  - Emissions from Urgentem matched via ISIN; Scope 1 and Scope 2 used (Scope 3 excluded).
  - E-score (environmental pillar of ESG) from Refinitiv Eikon used for brown/green categorizations.

### Key empirical findings (baseline and post-2017)
- Baseline full-sample (January 2000 to June 2023; Table 1; specifications i–v; Observations: 2,035,501; R2: 0.003):
  - Climate Risk coefficient: 0.006 *** (0.001)
  - Aggregate coefficient: 0.001 (0.001)
  - Regulatory coefficient: 0.003 *** (0.001)
  - Opportunity coefficient: -0.001 (0.001)
  - Physical coefficient: 0.001 * (0.001)
  - Interpretation:
    - One standard deviation increase in the Climate Risk News index → 0.6 basis point increase in excess stock returns.
    - Annualized approximations reported: Climate Risk → approximately 1.5% annualized; Regulatory → approximately 0.75% annualized; Physical → approximately 0.25% annualized.
  - Climate risk premium driven by Brazilian, Mexican and Chilean firms, followed by Peru and Colombia (Table A4).
- Post-Paris period (January 2017 to June 2023; Table 2; Observations: 333,247; R2: 0.009):
  - Climate Risk: 0.014 *** (0.002)
  - Aggregate: 0.016 *** (0.003)
  - Regulatory: 0.004 * (0.002)
  - Opportunity: 0.016 *** (0.002)
  - Physical: 0.007 *** (0.002)
  - Emissions Intensity (where included): 0.002 (0.002)
  - Interpretation:
    - One-standard-deviation increase in Climate Risk index → 1.4 basis point increase in equity returns ≡ annualized return of 3.5%.
    - One-standard-deviation increases in Aggregate, Regulatory, Opportunity, Physical indices → 1.4 bps, 0.4 bps, 1.6 bps, 0.7 bps respectively; annualized gains of 4%, 1%, 4%, and 1.8% respectively.
  - Coefficients for carbon intensity positive but not always statistically significant; significance sensitive to controls and sample restrictions (Tables A11, A12).

### Time-varying climate risk premium and factor controls
- Rolling regressions use a 1.5-year window moved by one month.
- Pattern over time:
  - Until Q2 2012: premium mostly negative or insignificant (exception 2008).
  - From Q2 2012 onward: premium turns positive except brief dip in 2014; significant rise before 2015 (Paris Agreement ratification) and a surge stabilizing around 4% annually after 2020.
  - Including carbon emissions yields broadly similar pattern.
- Country-level annualized risk premiums (Climate Risk index):
  - Brazil: 2 to 2.5 percent
  - Mexico: 2 to 2.5 percent
  - Chile: approximately 1 percent
  - Peru: statistically insignificant and comparably small under baseline.
- Time series regressions of climate risk premium ψt on Fama-French factors and MOM (Table 3):
  - Baseline cross-sectional coefficient: 0.006.
  - Cross-sectional post-2017 coefficient: 0.014; time series between 0.012 and 0.013.
  - Residual α0 remains statistically and economically significant in all specifications; climate risk premium not driven by conventional factors.

### Green vs brown firm interactions
- E-score interactions (Table 4):
  - Full sample: E-score × (Regulatory/Opportunity/Physical) interactions negative and significant; one standard deviation increase in E-score reduces required returns by 0.3 bps, 0.3 bps, and 0.2 bps respectively.
  - Post-2017: interaction magnitudes 0.5 to 0.7 bps; annual reduction in required returns of 1.3% to 1.8% for a one standard deviation increase in E-score.
  - Observations: 556,466 (full sample columns); 301,892 (post-2017 columns); R2: 0.004 (full), 0.008 (post-2017).
- Emissions intensity interactions (Table A13):
  - Regulatory exposure × emissions intensity positive and significant: firms emitting one standard deviation more than average → required risk premium increases by 0.3 bps ≡ annualized increase of 0.75%.
  - Other interactions mixed or insignificant.
- Industry clustering:
  - Industry fixed effects (Table A10) leave baseline results consistent, suggesting firm-level pricing.
  - Industry differentials (Table A19; relative to Utilities):
    - Climate Risk × Industrials 0.018 *** (0.005)
    - Climate Risk × Financials 0.011 ** (0.005)
    - Climate Risk × Real 0.029 *** (0.006)
    - Climate Risk × ConsumerDisc 0.023 *** (0.005)
    - Climate Risk × ConsumerStap 0.012 ** (0.005)
    - Climate Risk × BasicMaterials 0.014 ** (0.006)

### Official-gazette news regressions and event study
- Official-news baseline regressions (January 2017 to June 2023; Table 5; Observations: 620,430; R2: 0.004):
  - Aggregate (official): -0.003 * (0.001)
  - Opportunity (official): 0.002 (0.001)
  - Physical (official): -0.016 *** (0.002)
  - Interpretation:
    - Official Aggregate and Physical news associate with negative stock-return effects; Opportunity news positive but statistically insignificant and about one quarter of the media-based premium.
- Event study (panel local projections `à la Jorda (2005)`):
  - Treatment: Major climate legislation published in official gazettes (57 identified events).
  - Brown/green classification:
    - Emissions-based: brown if carbon emissions ≥ 75th percentile within industry and year.
    - E-Score-based: brown if E-Score < 25th percentile within industry and year.
  - Main event-study results:
    - Brown firms’ cumulative differential underperformance: 0.6% after one week (emissions-based classification, panel A, left panel (i)).
    - Low E-Score firms’ relative underperformance: 0.5% after one week (diminishing by two-week horizon).
    - No significant anticipatory effects detected (Table A16).
  - Inference: major climate legislation leads to short-run underperformance of high-emission / low E-Score firms relative to green firms; consistent with portfolio re-pricing and shifts in ESG factor.

### Robustness and validation
- Robustness checks reported include:
  - Inclusion/exclusion of emissions, alternative clustering, firm fixed effects, industry exclusion (Electricity), rolling-window analysis, Newey-West with 6 lags, forward-looking expected return proxies (Martin and Wagner (2019)) on a small subset.
  - Forward-looking proxies (Table A15; 13 firms; Observations: 25,465; R2: 0.145) show positive and significant association between climate risk and forward-looking excess returns; e.g., Climate Risk -0.016 *** (0.003) in that subsample specification.
  - Results robust to exclusion of top carbon-intensive sector (Electricity); risk premia magnitudes higher than baseline.
  - Physical index validated versus EM-DAT for Brazil: months with two or more physical events correspond to spikes in Physical index (Figure A3).
  - Measurement noise reduced by very large news corpus (>30 million articles).
- Summary statistics (Table A7: selected exact figures):
  - Excess Stock Seturn: Mean 0.03, Std 2.69, Min -99.24, 25% -0.28, 50% -0.02, 75% 0.21, Max 99.99, Obs 2,654,720
  - E Score: Mean 43.57, Std 27.53, Min 0.00, 25% 18.93, 50% 46.06, 75% 66.11, Max 97.51, Obs 609,727
  - Emissions (1 + 2): Mean 3.92, Std 1.96, Min 0.03, 25% 2.49, 50% 3.76, 75% 5.55, Max 9.56, Obs 607,287
  - Emissions Intensity (1 + 2): Mean 10.54, Std 3.00, Min 0.08, 25% 8.53, 50% 10.46, 75% 12.67, Max 19.10, Obs 607,287
  - Newspapers Aggregate: Mean 1.70, Std 1.05, Min 0.00, 25% 1.09, 50% 1.61, 75% 2.19, Max 5.39, Obs 62,442
  - Official Aggregate: Mean 0.67, Std 0.99, Min 0.00, 25% 0.00, 50% 0.28, 75% 1.16, Max 5.56, Obs 11,991
  - VIX: Mean 23.12, Std 10.14, Min 9.14, 25% 16.37, 50% 21.33, 75% 27.31, Max 82.69, Obs 15,108

### Main empirical conclusions and implications
- A measurable climate risk premium exists in Latin America’s equity markets:
  - Climate Risk and Regulatory news are associated with higher excess stock returns in both full-sample and post-2017 analyses.
  - Post-2017 coefficients are larger and more statistically significant than full-sample estimates.
  - Official-gazette Aggregate and Physical news associate with negative stock-return effects; Opportunity official news shows smaller positive association relative to media.
- Green (higher E-score / lower emissions) firms experience lower required returns under climate-related uncertainty:
  - E-score interactions indicate lower risk premia for greener firms; emission-intensity interactions point to higher premia for high-emission firms in some specifications.
- Policy timing and market pricing:
  - Daily media-based indices capture earlier-stage policy discussions and opportunities; official gazettes capture legally binding final actions.
  - Distinguishing media vs official-gazette news enables analysis of timing and market pricing of transition and physical risks.
- Industry and country heterogeneity:
  - Climate risk premia vary across industries and countries; Brazil and Mexico show larger annualized premiums (2–2.5 percent), Chile around 1 percent, Peru small/insignificant.
  - Industry-level heterogeneity implies need for granular, sector-specific assessments for financial and policy implications.

*Source: wpiea2025037-print-pdf (sections 2.1, 2.3, 4.4).*

### 2.1    Climate news  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  

### 2.1    Climate news

### Scope and data coverage
- Approximately 30 million news articles published between January 2000 and July 2023 by media outlets in Brazil, Mexico, Chile, Peru, Colombia, Costa Rica, Ecuador, and Uruguay were sourced to build daily indicators of climate news.
- For official climate policy news and major climate legislation, official government gazettes were examined for the eight countries during the period 2017-2023.
- The media source database used: Dow Jones Factiva.
- Each country in the sample publishes laws, decrees, and regulations in its own gazette at varying regularity, ranging from 740 publications on average per month in Chile to 25 in Brazil.
- The Climate Policy Database was used to cross-check and fine-tune official news indicators.

### Media landscape and concentration
- Brazil and Mexico constitute the largest fraction of newspaper articles (see Table A1).
- The number of sourced articles in each country exceeds 1/2 a million, even in smaller economies.
- The largest publishing house shares of total articles: 9 percent in Brazil and Colombia; around 15 percent in Chile and Mexico; 20 percent in Peru and Ecuador; 30 percent in Uruguay; 50 percent in Costa Rica.

### Constructed indices and methodology (overview)
- Daily exposure indices were constructed for each country leveraging the news corpus and guided by climate-related bi-grams identified in Sautner et al. (2023a) and methodology building on Baker et al. (2016).
- Indices reflect media coverage of climate events including climate opportunities, regulatory action, and physical risks.
- A climate risk sentiment measure was derived by combining climate events with a notion of uncertainty.
- Official gazette news capture presidential decrees, laws, and regulations with legally binding consequences; official gazettes are expected to publish climate-related news at the end of the legislative process, while newspapers pick up discussions earlier in the process.
- For the event study, a binary series of "major climate legislation" was compiled by extracting identified events via official gazettes and focusing on major, fully legislated climate policies, broadly following Barrett et al. (2024).

### Link to firm-level and other datasets (context)
- Climate news indices were combined with a firm-level dataset covering 628 publicly listed firms from 2000, with firm financial variables, emissions, and ESG data, and selected country-level controls (details in sections 2.2 and later).
- Emissions data underlying some analyses come from Urgentem (matched via ISIN), based on Carbon Disclosure Project (CDP) reporting; Scope 1 and Scope 2 emissions were used (Scope 3 excluded).
- E-score (environmental pillar of ESG) from Refinitiv Eikon is used to construct brown/green categorizations in later analysis.

### Key empirical findings related to climate news (summary from paper)
- A one-standard-deviation increase in the Climate Risk news index is associated with a 1.5% increase in annualized excess stock returns in the full sample.
- Restricting the sample to 2017 onward, the same one-standard-deviation increase is associated with a 3.5% increase in annualized excess stock returns.
- Post-2017, positive effects on stock prices are found from regulatory, opportunity, and physical climate news.
- Latin America’s climate risk premium is increasing over time.
- Results are robust to inclusion of different measures of carbon emissions, various methods of calculating standard errors and clustering, and to country-level time series regressions on conventional Fama-French risk factors.
- Climate risk premiums remain robust when using option-implied forward-looking expected return proxies of Martin and Wagner (2019) in a small-sample study.

### Implications highlighted in the paper (linked to climate news measures)
- Media and official-gazette-based climate news indices allow distinction between early-stage policy discussions (media) and legally binding policy actions (official gazettes), enabling analysis of timing and market pricing of transition and physical risks.
- The constructed daily indices provide a means to capture both upside (climate opportunities) and downside (physical risks, regulatory burden) potential associated with climate news in Latin America.

*Source: wpiea2025037-print-pdf (section 2.1 Climate news) — IMF working paper content as provided.*

### 2.3    Country-level and macro data

### 2.3    Country-level and macro data

### Country-level and macro variables used
- Include VIX to capture financial market movements at both country and global levels over time.
- Incorporate COVID-19 lockdown intensity for each country using the dataset from Hale et al. (2021).
- Include the oil price to capture energy and supply related dynamics.
- For time series regressions, obtain conventional Fama French factor returns from Global factor database of Jensen et al. (2023) for five Latin American countries: Brazil, Mexico, Chile, Peru, and Colombia.  Factor returns data are not available for Costa Rica, Uruguay, and Ecuador.
- Factor returns data include market, value, size, profitability, investment, momentum.
- Summary statistics of all variables are reported in Table A7.

*Construction and aggregation notes*
- Firm-level data are combined with country-level and global variables.
- Factor returns data coverage limitation: Costa Rica, Uruguay, and Ecuador missing.

*Table/series referenced*
- Summary statistics: Table A7.

### Construction of climate news indices (Section 3)

- Daily country-level indicator built from January 2000 to July 2023 reflecting frequency of news articles related to climate action and exposure (emissions reduction, renewable energy, climate regulation, physical shocks).
- Method:
  - Use climate-related bigrams identified by Sautner et al. (2023a); translate them into Portuguese and Spanish.
  - Use Dow Jones Factiva’s newspaper archives to derive a daily count of articles containing these bigrams or combinations.
  - Standardize resulting frequency by the total number of articles in a given month (standardizing by total number of articles per day does not change qualitative results but increases indicator’s volatility).
- Aggregate index formula (as reported):
  - Climate News_{j,t} = (1 / N_{j,m}) * sum_{b}^{B} (1[b ∈ C])
    - where b is a particular bi-gram, 1 is an indicator taking 1 if the bi-gram exists in an article published in country j, C is the set of bi-grams, and N_{j,m} is the total number of articles published each month in country j.
- Indices plotted as monthly averages of daily indicators for readability.

### Climate news subcategories (topic-based indices)
- Construct daily indices for each country for:
  1. Regulatory — climate-related regulatory news.
  2. Physical — news related to physical climate events and natural disasters (floods, drought, sea level rise, etc.).
  3. Opportunity — news about climate-related technological developments and opportunities (wind power, lithium batteries, solar parks).
- Observed patterns:
  - Mildly positive correlation between country indices across the sample (Table A8); driven by major regional/global events (e.g., George W. Bush’s withdrawal from the Kyoto protocol in 2001 and the Paris agreement in 2015).
  - Media reporting primarily driven by country-specific events → relatively heterogeneous patterns across indices.
  - More variation in indices for countries with fewer news sources (Costa Rica and Uruguay).
  - Post-pandemic uptick in Opportunity news index across countries; particularly notable in Brazil (sustainable agriculture, deforestation, carbon credits, biofuels, rapid deployment of wind and solar power).
  - Chile: upward trend associated with rare earths, particularly lithium.
  - Peru: steady decline in Opportunity news index throughout the 2010s reflective of lack of investments in mining and relatively low business confidence.
  - Increased law-making after the pandemic in Brazil, Chile, Colombia, Mexico and Uruguay (Regulatory index).
  - Physical news index does not show a clear upward trend across countries over the past decade; uptick in late 2000s reversed in the 2010s; moderate pick-up from 2020 onward in Brazil, Chile, Costa Rica; Uruguay spike in 2023 related to two El Niño events.
  - Validation: compare Physical index to actual physical climate events in Brazil from the International Disaster Database (EM-DAT); occurrence of two or more events tends to correspond closely to spikes in the Physical index. Exclude biological, geophysical (earthquakes) and extra-terrestrial events.

### Climate sentiment / risk-conditioned index
- Construct Climate Risk News index by conditioning each bi-gram of the aggregate set to be in the same sentence with the words "risk", "uncertainty" or synonyms (specification similar to Hassan et al. (2019)):
  - Climate Risk News_{j,t} = (1 / N_{j,m}) * sum_{b}^{B} (1[b ∈ C]) × 1[b,r ∈ S]
    - where r is a respective risk-related term, S is the set of synonyms of risk; other terms as in the aggregate formula.
- Summary statistics of various indices: Table A7.
- Example illustration: Monthly averages of Climate Risk News index for Brazil from January 2000 to June 2023 capture international and domestic climate policy events (e.g., US withdrawal from Kyoto, Kyoto protocol effective date, UN climate conferences in Copenhagen and Doha, signing of Paris agreement, US withdrawal from Paris under Trump; domestic presidential decrees, laws on bio-fuels, financing climate-related infrastructure projects, fighting deforestation in the Amazon).
- Comparison with Global Climate Policy Uncertainty index (Gavriilidis (2021)): some overlap but different country-specific patterns; constructed to capture country-specific news not necessarily visible in global aggregates.
- Measurement noise considerations:
  - Indices likely measured with some noise; assumed correlated with true signal plus random noise.
  - Laudati and Pesaran (2023) argue increased number of news articles reduces noise and yields more consistent estimates.
  - Constructed exposure measures use more than 30 million news articles, reducing noise substantially.

### Climate policy news in official gazettes (Section 3.2)
- Distinguish official gazettes (government bulletins/official publication outlets) from general news outlets to differentiate regulatory initiation vs final approval timing.
- Expectation:
  - Official gazettes publish climate-related news at the end of the legislative process.
  - Media outlets pick up discussions at the beginning of the process and upon law approval.
- Time horizon for official gazette analysis limited to 2017 to 2023 due to publication history across sample.
- Count bi-grams related to aggregate, opportunity, and physical news in official gazettes and construct corresponding indices; also perform textual analysis including uncertainty notion for robustness.
- Observed: very few entries in official gazettes (as expected for regulatory publications).
- Comparative dynamics (media vs official gazettes, last five years aggregated monthly):
  - Brazil and, to some extent, Mexico show surprising synchronization between media-based and official-gazette indices.
  - Colombia, Peru and Uruguay show divergent paths more recently — possible explanations:
    - New legislation might not have been necessary to address climate risks floated in the media.
    - News outlets picking up global or regional climate discourse that domestic policies do not address.
- Construction of Major Climate Policy Events binary indicator:
  - Uses aggregate climate news index from official gazettes and focuses on most important legislation.
  - Manual verification that publications refer to law, decree or regulation on climate action that is fully legislated.
  - Crosscheck with Climate Policy Database for Latin American policies.
  - Include only legally binding policies; exclude government plans or projections not yet implemented.
  - Detect 57 events in total (identified events listed in Table A9).

### Notes on figures and tables referenced
- Figure 1: Climate news from media outlets (monthly averages plotted).
- Figure A3: Physical index vs actual events in Brazil (monthly averages).
- Figure 2: Climate Risk News index for Brazil with annotated international and national climate policies.
- Figure A4: Comparison to Global Climate Policy Uncertainty index.
- Table A3: Official gazettes examined by country.
- Table A7: Summary statistics of indices and variables.
- Table A8: Correlations across country indices.
- Table A9: Identified Major Climate Policy Events.
- Figure A2: Number of firms providing emissions data over time.

### Data coverage and validation remarks
- Bi-grams derived from Sautner et al. (2023a), which used quarterly earnings call transcripts of more than 10,000 firms across 34 countries over 2002–2020.
- Constructed indicators are empirically derived and proven to predict real outcomes related to the net-zero transition.
- More than 30 million news articles used in construction, reducing measurement noise.

### Empirical specification and baseline estimates (Section 4 and 4.1)

- Baseline cross-sectional regression (as outlined by Bolton and Kacperczyk (2021); Aswani et al. (2024)):
  - r_{i,j,t} = β_{0} + β_{1} Climate News_{j,t} + β_{2} Firm Controls_{i,j,t−1} + β_{3} Macro Controls_{j,t} + μ_{t} + γ_{j} + ε_{i,j,t}
    - r_{i,j,t} = excess stock return of firm i in country j at time t.
    - Climate News_{j,t} = set of measures of exposure to climate change derived from newspapers for country j.
    - Firm Controls_{i,j,t−1} include time-varying market betas, log-total assets, debt-to-assets, log-market capitalization, stock price volatility, book-to-market ratio, PPE-to-assets, EBIT-to-assets, capital expenditures-to-assets.
    - Macro Controls include (i) oil price changes, (ii) volatility measures (VIX), (iii) lockdown intensity.
    - μ_{t} and γ_{j} represent time and country fixed effects respectively.
    - Variables are winsorized (Table A7) and standardized to zero mean and unit standard deviation in each estimation.

- Baseline sample and dependent variable:
  - Sample spans January 2000 to June 2023.
  - Dependent variable: excess stock returns.
  - All specifications include country, year-month-day fixed effects.

### Baseline regression results — Full sample (Table 1)
- Specifications (models i to v) separately include each climate news index:
  - Climate Risk coefficient: 0.006 *** (0.001)
  - Aggregate coefficient: 0.001 (0.001)
  - Regulatory coefficient: 0.003 *** (0.001)
  - Opportunity coefficient: -0.001 (0.001)
  - Physical coefficient: 0.001 * (0.001)
- Controls: Firm Controls Yes; Country Controls Yes.
- Fixed-effects: Country FE Yes; Year-Month-Day FE Yes.
- Statistics:
  - Observations: 2,035,501 (for each model)
  - R^2: 0.003 (for each model)
- Notes on significance stars: ***: 0.01, **: 0.05, *: 0.1
- Interpretation:
  - One standard deviation increase in the Climate Risk News index is associated with a 0.6 basis point increase in excess stock returns.
  - Impact magnitudes relative to Climate Risk News index:
    - Regulatory ≈ half the magnitude of Climate Risk News.
    - Physical ≈ one sixth the magnitude of Climate Risk News.
  - Annualized approximations reported in text (derived from coefficients):
    - Climate Risk News → approximately 1.5% annualized impact on excess returns.
    - Regulatory → approximately 0.75% annualized impact.
    - Physical → approximately 0.25% annualized impact.
- Conclusion from full-sample baseline:
  - Positive and statistically significant relationship between climate news and excess stock returns, except for aggregate and opportunity-based climate news.
  - Evidence supports existence of a climate risk premium in Latin America’s largest economies (investors demand positive risk premium to hold stocks exposed to climate-related uncertainty and transition risks).
  - Climate risk premium driven by Brazilian, Mexican and Chilean firms, followed by Peru and Colombia (Table A4).

### Robustness: Post-Paris period and emissions (Table 2)
- Sample restricted to post-Paris agreement period (January 2017 to June 2023) where at least 100 firms provide emissions data (Figure A2).
- Baseline specification estimated with and without carbon emissions intensity.
- Table 2 reported results (columns i–v without emissions; vi–x with emissions). Select coefficients reported:
  - Climate Risk: 0.014 *** (0.002) (columns i–ii)
  - Aggregate: 0.016 *** (0.003) (columns iii–iv)
  - Regulatory: 0.004 * (0.002)
  - Opportunity: 0.016 *** (0.002)
  - Physical: 0.007 *** (0.002)
  - Emissions Intensity: 0.002 (0.002) (in columns where included)
- Controls: Firm Controls Yes; Country Controls Yes.
- Fixed-effects: Country FE Yes; Year-Month-Day FE Yes.
- Statistics:
  - Observations: 333,247 (for each specification)
  - R^2: 0.009 (for each specification)
- Interpretation and remarks:
  - Impact of climate-related news remains consistently positive and statistically significant regardless of inclusion of carbon emissions.
  - Coefficients for carbon intensity remain positive but lack statistical significance in the reported specifications.
  - Role of emissions influenced by correlations among covariates and missing observations.
  - When certain controls excluded (e.g., EBIT-to-assets, stock volatility, lockdown intensity), emissions intensity can become statistically significant and positive.
  - Restricting sample to the last three years — when most firms began reporting emissions data — yields statistically significant coefficients for carbon emissions intensity (Table A12); this result may reflect sample selection or increased climate awareness in recent years.
- Note: Aswani et al. (2024) motivate use of emissions intensity; qualitative and quantitative results similar using level of emissions (Table A11).

*Key dataset counts and sample sizes referenced*
- Full-sample observations: 2,035,501 (Table 1).
- Post-2017 observations: 333,247 (Table 2).
- More than 30 million news articles used to construct indices.

*Main empirical conclusion*
- A measurable climate risk premium exists in Latin America’s equity markets: climate-risk and regulatory news are associated with higher excess stock returns; climate news effects are robust in full-sample and post-Paris analyses even after controlling for firm-level characteristics, macro controls (oil price, VIX, COVID lockdown intensity), and emissions intensity (where available).

_Italic: Source — wpiea2025037-print-pdf - 2.3    Country-level and macro data_

### 2023.  We standardize all variables to have zero mean and unit standard deviation.  The dependent variable is excess sto

### wpiea2025037-print-pdf - 2023

### Methodology and data
- All variables standardized to have zero mean and unit standard deviation.
- Dependent variable: excess stock returns.
- Robust standard errors reported in parentheses.
- All specifications include country, year-month-day fixed effects.
- Winsorize certain variables to remove the effect of extreme observations (provided in Table A7).
- Regressions of excess stock returns include:
  - Time-varying market betas and firm controls: log-total assets, debt-to-assets, stock price volatility, book-to-market ratio, PPE-to-assets, EBIT-to-assets, capital expenditures-to-assets.
  - Additional controls: oil price changes, market volatility, COVID lockdown intensity.
  - Lagged control: log of carbon emissions from the previous year to avoid look-ahead bias.
- Rolling regressions for time-varying risk premium use a sample size of one and a half years, moving the sample window by one month.

### Climate risk premium over time
- Time-varying climate risk premium estimated as sensitivity of stocks to the Climate Risk News index; mean estimate shown with 68% and 90% confidence intervals (robust standard errors).
- Pattern over time:
  - Until the second quarter of 2012, the premium is mostly negative or insignificant, with exception of 2008.
  - From the second quarter of 2012 onward, the risk premium turns positive except for a brief dip in 2014 (taper tantrum).
  - Significant rise shortly before 2015 aligning with ratification of the Paris Agreement in December 2015.
  - Between 2018 and 2020, the premium declines but remains positive and significant, then surges and stabilizes around 4% annually.
  - Including carbon emissions produces a broadly similar pattern.
- Post-2017 (compared to full sample):
  - Climate risk premiums and statistical significance of climate change exposure measures increase.
  - A one-standard-deviation increase in the Climate Risk index corresponds to a 1.4 basis point (bps) increase in equity returns, equivalent to an annualized return of 3.5%.
  - One-standard-deviation increases in the Aggregate, Regulatory, Opportunity, and Physical indices are associated with equity return increases of 1.4 bps, 0.4 bps, 1.6 bps, and 0.7 bps, translating to annualized gains of 4%, 1%, 4%, and 1.8%, respectively.
- Notable observation: pronounced rise in the opportunity risk premium reflecting regional advances in solar energy, lithium batteries, and other green technologies; amplification after 2017 consistent with prior literature.

### Climate risk premium by country
- Country-level estimates for Brazil, Chile, Mexico, and Peru (Climate Risk index):
  - Annualized risk premium of 2 to 2.5 percent in Brazil and Mexico.
  - Approximately 1 percent in Chile.
  - For Peru, climate risk premium statistically insignificant and comparably small under baseline specification.
- Country-level variations highlight interplay between climate risk and investor sentiment across national contexts.
- Excluded countries due to insufficient firm counts: Colombia, Uruguay, Costa Rica, and Ecuador (number of firms per country in Table A4).

### Climate risk premium and conventional factor returns
- Time series regressions of climate risk premium ψt on conventional Fama-French factors and momentum:
  - Regression: ψt = α0 + α1 Ft + εt, where Ft includes market risk premium, SMB, HML, RMW, CMA, and MOM.
  - α0 measures residual climate risk premium; α1 shows sensitivity to the six traditional risk factors.
- Results:
  - Newey-West adjusted heteroskedasticity and autocorrelation robust standard errors reported using 6 lags.
  - Baseline cross-sectional coefficient of climate risk premium: 0.006.
  - Time series coefficients remain between 0.006 and 0.007 for full sample.
  - Cross-sectional post-2017 coefficient: 0.014; time series remains between 0.012 and 0.013.
  - Climate risk premium marginally absorbed by risk factors in post-2017 sample, but residual α0 remains statistically and economically significant in all specifications.
  - Conclusion: estimated climate risk premium not driven by conventional market risk factors; it captures investor pricing of climate risks.
- Table 3 specific results (selected exact figures preserved):
  - Constants range from 0.006 ∗∗∗ to 0.013 ∗∗∗ across models.
  - Market factor coefficients shown (examples): -0.010, -0.007, -0.018, -0.020 ∗∗∗; and positives in post-2017 columns (e.g., 0.009, 0.012, 0.037 ∗∗).
  - Observations: 228 (models i–v) and 72 (models vi–x).
  - R2 values presented: 0.037, 0.051, 0.020, 0.083, 0.141, 0.110, 0.084, 0.068, 0.079 (as shown).
  - Newey-West standard-errors in parentheses; significance notation: ***: 0.01, **: 0.05, *: 0.1.

### A lower climate risk premium for "green" firms (E-score and emissions interactions)
- Hypothesis: greener firms perceived as less risky and may require lower returns, particularly under higher climate-related uncertainty.
- Interaction approach: climate news indices × environmental (E) pillar of ESG scores and × carbon emissions intensity.
- Results with E-score (Table 4):
  - Full sample (columns i–v): news indices’ main coefficients positive and statistically significant; interactions for Regulatory, Opportunity, and Physical indices yield negative and statistically significant coefficients.
    - A one standard deviation increase in the E-score corresponds to reductions in required returns of 0.3 bps, 0.3 bps, and 0.2 bps for regulatory, opportunity, and physical related news, respectively.
  - Post-2017 (columns vi–x): all interaction effects statistically significant, magnitudes ranging from 0.5 to 0.7 basis points, translating into an annual reduction in required returns of 1.3% to 1.8% for a one standard deviation increase in the E-score.
  - Estimations include firm controls, country controls, country FE, year-month-day FE.
  - Observations: 556,466 for full sample columns; 301,892 for post-2017 columns.
  - R2: 0.004 (full sample columns) and 0.008 (post-2017 columns).
- Results with emissions intensity (Table A13 summary):
  - Main effects of climate exposure measures remain positive and statistically significant.
  - Interaction terms: only regulatory exposure × emissions intensity is positive and statistically significant.
    - For firms emitting one standard deviation more than the average, required risk premium increases by 0.3 basis points, equivalent to an annualized increase in required returns of 0.75%.
  - Opportunity and Physical interactions positive but statistically insignificant; Climate Risk and Aggregate interactions negative but statistically insignificant.
- Interpretation: E-scores and emissions provide complementary evidence that greener firms face lower climate risk premia.

### Industry-level heterogeneity
- E-scores and emissions cluster by industry; analysis assesses whether investors price climate risk at firm vs. industry level.
- Including industry fixed effects (Table A10) leaves baseline results highly consistent, suggesting firm-level pricing of climate exposure.
- Industry differential effects (Table A19, Climate Risk index; relative to utilities):
  - Pronounced climate risk premiums in real estate, consumer discretionary, and basic materials.
  - Less consistent or insignificant effects in financials, energy, and technology.
  - Note: energy sector highly heterogeneous in climate exposure (from traditional power plants to renewables).
- Interaction effects by industry (Figure 6 visualization):
  - Across industries, climate risk premiums consistently positive; interaction effects predominantly negative (risk premiums decrease for greener firms within an industry), except for the technology sector.
  - Opportunity measure exhibits strongest effects in: basic materials, consumer discretionary, consumer staples, and energy.
  - Climate Risk exposure drives responses in financials, industrials, and real estate.
  - Industrials, real estate, and consumer discretionary display positive and significant premiums across exposure types.
  - Other industries price only specific types of risks, indicating variability in investor perceptions.
- Conclusion: climate risk pricing varies significantly across industries and exposure types; granular, industry-specific assessments are necessary to evaluate financial and policy implications.

*Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025037-print-pdf.pdf*

### 4.4    Official news

### 4.4    Official news

### Methodology
- Replicate baseline regressions using official news indices based purely on textual analysis, focusing on the Aggregate, Opportunity and Physical news indices and excluding the Climate Risk and Regulatory news indices.
- Rationale for exclusions: instances of the term "risk" are rarely observed in official publications and official news is primarily regulatory announcements.
- Expectation: official gazettes publish climate-related news at the end of the legislative process (as opposed to media outlets), reducing concern of reverse causality given time to prepare, debate and pass laws.
- All variables standardized to zero mean and unit standard deviation.
- Dependent variable: excess stock returns.
- Controls and design:
  - Firm Controls: included.
  - Country Controls: included.
  - Fixed effects: Country FE and Year-Month-Day FE included.
  - Robust standard errors clustered at firm level; Newey-West estimator with 6 lags used for some local-projection confidence intervals.

### Regression results (Table 5)
- Sample period: January 2017 to June 2023.
- Key coefficient estimates (standard errors in parentheses; significance codes as reported):
  - Aggregate: -0.003 * (0.001)
  - Opportunity: 0.002 (0.001)
  - Physical: -0.016 *** (0.002)
- Statistics:
  - Observations: 620,430
  - R2: 0.004
- Notes and interpretation:
  - Official news on Aggregate and Physical climate risks feature negative coefficients.
  - Laws/decrees may signal emergency aid or negative shocks from large natural disasters, or public investments in adaptation lowering climate-related uncertainty.
  - Legislation on opportunity-related climate policies is associated with a positive but insignificant climate risk premium, about one quarter of the premium estimated using media news.
  - Additional covariates in regressions: log-total assets, debt-to-assets, stock price volatility, book-to-market ratio, PPE-to-assets, EBIT-to-assets, capital expenditures-to-assets, oil price changes, market volatility, and covid lockdown intensity.
  - Certain variables winsorized to remove extreme observations (provided in Table A7).

### Event study: major climate legislation (local projections)
- Specification: panel local projections `à la Jorda (2005)` at firm level (equation (5.1) provided).
- Treatment indicator: I(Leg_{i,t}) = 1 if major climate legislation published in official gazettes on that day.
- Heterogeneity captured by Category_{i,t} (brown = 1, else 0).
- Brown/green classification approaches:
  - Emissions-based: brown if carbon emissions ≥ 75th percentile within industry group and year; green otherwise.
  - E-Score-based: brown if environmental pillar of ESG score (E-Score) < 25th percentile within industry and year.
  - Robustness: similar results when using median thresholds; emissions-based classification more resilient to green-washing; E-Score has longer data history and more reporting firms.
- Inference:
  - Standard errors clustered at firm level; Newey-West with 6 lags (or 12) yields almost identical results.
  - No significant anticipatory effects found across multiple time horizons (results in Table A16).

- Main event-study findings:
  - High-emission (brown) firms’ stock prices significantly underperform relative to green firms:
    - Cumulative differential of 0.6% after one week (emissions-based classification, left panel (i)).
    - Underperformance persists over the two-week window, with reduced statistical significance.
  - Low E-Score firms experience relative underperformance of 0.5%, which diminishes by the end of the two-week period.
  - Marginal effect on brown firms shows a significant decline in stock prices within the first week under both classification schemes (Figure A5).

- Interpretation:
  - Results consistent with Pástor et al. (2021): brown firms require higher expected returns in equilibrium, but when ESG factor shifts, green firms outperform brown ones.
  - Major climate legislation may signal shifts in the ESG factor and/or reduce climate-related uncertainty by charting responses (e.g., support for renewables, electricity sector reform, carbon market frameworks).

### Robustness checks
- Investor awareness and rolling-window analysis:
  - Rolling window length: 1.5 years, rolled by one month.
  - Climate risk premium shows a steady increase on average and remains predominantly positive after the second quarter of 2012, with a marked increase before and after the Paris Agreement; prior to 2012 the premium often turned negative or declined.
  - Including carbon emissions in estimations yields a climate risk premium that remains positive and aligns closely with baseline but is more variable.
- Sector influence:
  - Re-estimated baseline excluding the top carbon-intensive sector (Electricity) (results in Table A14).
  - Findings remain qualitatively similar; risk premia magnitudes are higher than baseline, indicating results are not driven solely by a single carbon-intensive sector.
- Forward-looking expected return proxies:
  - Calculated options-implied forward-looking expected return proxy of Martin and Wagner (2019) on a small subset of the sample.
  - Results mostly show a positive, statistically significant, and economically meaningful association between climate risk and forward-looking excess stock returns (Table A15).
  - Coefficient estimates larger in magnitude compared to baseline, but sample size limitations prevent generalization to the full population.

### Conclusion (from this section)
- Official-gazette climate-related Aggregate and Physical news associate with negative stock-return effects; Opportunity news shows a smaller, insignificant positive association relative to media-based estimates.
- Major climate legislation announcements lead to short-run underperformance of high-emission / low E-Score (brown) firms relative to green firms (cumulative differentials: 0.6% after one week by emissions; 0.5% by E-Score).
- Robustness analyses (rolling windows, sector exclusion, forward-looking return proxies) support the main findings and suggest the climate risk premium is generally positive, growing over time, and not driven by a single sector.

*Source: 4.4 Official news — wpiea2025037-print-pdf*

### References

### References

### Cited literature (selected)
- Aswani, J., Raghunandan, A., and Rajgopal, S. (2024). Are carbon emissions associated with stock returns? Review of Finance, 28(1):75–106.
- Baker, S. R., Bloom, N., and Davis, S. J. (2016). Measuring Economic Policy Uncertainty*. The Quarterly Journal of Economics, 131(4):1593–1636.
- Barnett, M., Brock, W., and Hansen, L. P. (2020). Pricing uncertainty induced by climate change. The Review of Financial Studies, 33(3):1024–1066.
- Bolton, P. and Kacperczyk, M. (2021). Do investors care about carbon risk? Journal of Financial Economics, 142(2):517–549.
- Fama, E. F. and French, K. R. (2015). A five-factor asset pricing model. Journal of financial economics, 116(1):1–22.
- Giglio, S., Kelly, B., and Stroebel, J. (2021). Climate finance. Annual Review of Financial Economics, 13(1):15–36.
- Ilhan, E., Sautner, Z., and Vilkov, G. (2021). Carbon tail risk. The Review of Financial Studies, 34(3):1540–1571.
- Jorda, O. (2005). Estimation and inference of impulse responses by local projections. American Economic Review, 95(1):161–182.
- Martin, I. W. and Wagner, C. (2019). What is the expected return on a stock? The Journal of Finance, 74(4):1887–1929.
- Painter, M. (2020). An inconvenient cost: The effects of climate change on municipal bonds. Journal of Financial Economics, 135(2):468–482.
- Pastor, L., Stambaugh, R. F., and Taylor, L. A. (2021). Sustainable investing in equilibrium. Journal of Financial Economics, 142(2):550–571.
- Sautner, Z., van Lent, L., Vilkov, G., and Zhang, R. (2023a). Firm-level climate change exposure. The Journal of Finance.
- Sautner, Z., Van Lent, L., Vilkov, G., and Zhang, R. (2023b). Pricing climate change exposure. Management Science.

### Data sources and databases referenced
- Factiva (noted as Source for Table A1).
- EM-DAT database (used for physical climate event occurrences; Figure A3).
- Climate policy uncertainty index by Gavriilidis (2021) (referenced in Figure A4).

### Appendix listings (tables of counts and shares)
- Table A1: Share of articles by country
  - Brazil 38.90 1.60
  - Mexico 37.01 21.79
  - Chile 8.62 47.21
  - Peru 5.68 27.75
  - Colombia 4.38 0.14
  - Costa Rica 1.58 0.16
  - Uruguay 2.53 0.61
  - Ecuador 1.31 0.74
- Table A4: Number of firms by country
  - Brazil 2634 41.88
  - Mexico 1362 21.66
  - Chile 1251 19.90
  - Peru 65 10.35
  - Colombia 34 5.41
  - Ecuador 3 0.48
  - Uruguay 1 0.16
  - Costa Rica 1 0.16
- Table A5: Number of firms by industry (top entries)
  - Financials 1111 17.68
  - Consumer Discretionary 1041 16.56
  - Industrials 951 15.13
  - Basic Materials 731 11.62
  - Consumer Staples 701 11.15
  - Utilities 671 10.67
- Table A6: Number of firms by sector (selected entries)
  - Electricity 51 8.12
  - Investment Banking and Brokerage Services 45 7.17
  - Banks 42 6.69
  - Food Producers 39 6.21
  - Industrial Metals and Mining 39 6.21

### Summary statistics (Table A7: selected variables and exact figures)
- Excess Stock Seturn: Mean 0.03, Std 2.69, Min -99.24, 25% -0.28, 50% -0.02, 75% 0.21, Max 99.99, Obs 2,654,720
- E Score: Mean 43.57, Std 27.53, Min 0.00, 25% 18.93, 50% 46.06, 75% 66.11, Max 97.51, Obs 609,727
- ESG Score: Mean 47.47, Std 20.93, Min 1.25, 25% 32.22, 50% 48.84, 75% 63.25, Max 93.86, Obs 609,727
- Debt / Assets: Mean 26.68, Std 18.42, Min 0.00, 25% 12.54, 50% 26.11, 75% 38.05, Max 82.17, Obs 2,326,366
- Emissions (1 + 2): Mean 3.92, Std 1.96, Min 0.03, 25% 2.49, 50% 3.76, 75% 5.55, Max 9.56, Obs 607,287
- Emissions Intensity (1 + 2): Mean 10.54, Std 3.00, Min 0.08, 25% 8.53, 50% 10.46, 75% 12.67, Max 19.10, Obs 607,287
- Newspapers Aggregate: Mean 1.70, Std 1.05, Min 0.00, 25% 1.09, 50% 1.61, 75% 2.19, Max 5.39, Obs 62,442
- Official Aggregate: Mean 0.67, Std 0.99, Min 0.00, 25% 0.00, 50% 0.28, 75% 1.16, Max 5.56, Obs 11,991
- Market (time series variable): Mean 0.05, Std 1.56, Min -17.55, 25% -0.68, 50% 0.08, 75% 0.82, Max 16.28, Obs 29,778
- VIX: Mean 23.12, Std 10.14, Min 9.14, 25% 16.37, 50% 21.33, 75% 27.31, Max 82.69, Obs 15,108

### List of official gazettes (Table A3)
- Brazil: Díario Oficial da Unĩao (DOU) — Portuguese
- Mexico: Diario Oficial de la Federación — Spanish
- Chile: Diario Oficial de la República de Chile — Spanish
- Colombia: Diario Oficial — Spanish
- Peru: El Peruano — Spanish
- Uruguay: Diario Oficial — Spanish
- Costa Rica: La Gaceta Diario Oficial — Spanish
- Ecuador: Registro Oficial del Ecuador — Spanish

### Identified climate-related events (Table A9: selected items with dates)
- Brazil 2017-06-27: Establishing the Brazilian Forum on Climate Change
- Brazil 2018-03-16: RenovaBio
- Brazil 2018-07-05: Procedures and conditions for carrying out electric vehicle recharging activities
- Brazil 2022-01-06: Law on a just energy transition
- Brazil 2023-10-27: Nationally Determined Contribution
- Chile 2022-06-13: Climate Change Framework Law
- Colombia 2021-12-22: Law on low carbon development
- Costa Rica 2018-01-25: Law on incentives and promotion of electric transportation
- Mexico 2021-03-08: Emissions Trading System Mexico
- Peru 2018-04-18: Framework Law on Climate Change

### Empirical results — selected table coefficients (exact values preserved)
- Table A10: Stock returns and climate change exposure (Adding industry fixed effects)
  - Climate Risk 0.006 ∗∗∗ (0.001)
  - Aggregate 0.001 (0.001)
  - Regulatory 0.003 ∗∗∗ (0.001)
  - Opportunity -0.001 (0.001)
  - Physical 0.001 ∗ (0.001)
  - Observations 2,035,501; R2 0.003
- Table A11: Subsample (2017) with emissions level
  - Climate Risk 0.014 ∗∗∗ (0.002)
  - Aggregate 0.016 ∗∗∗ (0.003)
  - Regulatory 0.004 ∗ (0.002)
  - Opportunity 0.016 ∗∗∗ (0.002)
  - Physical 0.007 ∗∗∗ (0.002)
  - Emissions Level 0.003 (0.002)
  - Observations 333,247; R2 0.009
- Table A12: Sample 2019-2021
  - Climate Risk 0.014 ∗∗∗ (0.002)
  - Aggregate 0.020 ∗∗∗ (0.003)
  - Regulatory 0.006 ∗∗ (0.003)
  - Opportunity 0.016 ∗∗∗ (0.003)
  - Physical 0.005 (0.003)
  - Emissions Intensity 0.004 ∗∗ (0.002) in all columns reported
  - Observations 277,954; R2 0.011
- Table A13: Carbon Emissions (selected columns)
  - Full sample (i) Climate Risk 0.017 ∗∗∗ (0.002); Aggregate 0.010 ∗∗∗ (0.002); Opportunity 0.011 ∗∗∗ (0.002)
  - Subsample starting from 2019 (vi–x) Climate Risk 0.020 ∗∗∗ (0.003); Aggregate 0.018 ∗∗∗ (0.003); Opportunity 0.016 ∗∗∗ (0.003)
  - Interaction: Climate Risk × Emissions Intensity -0.003 (0.002) in one specification
  - Observations 401,371 (full); 235,739 (subsample)
- Table A15: Forward-looking expected return proxies and climate change exposure (13 firms)
  - Climate Risk -0.016 ∗∗∗ (0.003)
  - Aggregate 0.048 ∗∗∗ (0.008)
  - Regulatory 0.030 ∗∗∗ (0.006)
  - Opportunity 0.029 ∗∗∗ (0.007)
  - Physical 0.010 ∗∗ (0.003)
  - Observations 25,465; R2 0.145
- Table A16: Anticipation effects (Jan 2017 to Jun 2023)
  - Event(t+1) coefficients range from -0.001 to -0.002 with standard errors (0.001)–(0.002)
  - Event(t+7), Event(t+14), Event(t+21), Event(t+28) reported; coefficients mostly 0.000 to 0.001 with standard errors (0.001)–(0.002)
  - Observations range from 823,863 down to 790,021 across columns; R2 0.001
- Table A17: Stock returns and climate change exposure (Firm Fixed effects; Jan 2017–Jun 2023)
  - Climate Risk 0.014 ∗∗∗ (0.002)
  - Aggregate 0.015 ∗∗∗ (0.003)
  - Regulatory 0.009 ∗∗∗ (0.002)
  - Opportunity 0.017 ∗∗∗ (0.003)
  - Physical 0.004 ∗ (0.002)
  - Observations 333,247; R2 0.013
- Table A18: Clustering standard errors by firm-year-month
  - Climate Risk 0.014 ∗∗∗ (0.002)
  - Aggregate 0.016 ∗∗∗ (0.003)
  - Regulatory 0.004 ∗ (0.002)
  - Opportunity 0.016 ∗∗∗ (0.002)
  - Physical 0.007 ∗∗∗ (0.002)
  - Observations 333,247; R2 0.009
- Table A19: Climate risk premium by industry (relative to Utilities; sample from 2017 onwards)
  - Climate Risk overall -0.005 (0.004)
  - Climate Risk × Industrials 0.018 ∗∗∗ (0.005)
  - Climate Risk × Financials 0.011 ∗∗ (0.005)
  - Climate Risk × Real 0.029 ∗∗∗ (0.006)
  - Climate Risk × ConsumerDisc 0.023 ∗∗∗ (0.005)
  - Climate Risk × ConsumerStap 0.012 ∗∗ (0.005)
  - Climate Risk × BasicMaterials 0.014 ∗∗ (0.006)
  - Observations 777,248; R2 0.004

### Figures and notes (selected)
- Figure A1: Median carbon intensity by sector (tCO2 / mn$revenue).
- Figure A2: Number of firms reporting carbon emissions and E-Score per year.
- Figure A3: Physical climate events in Brazil: News and materialized events. Notes: vertical orange lines represent the number of physical climate events reported per month in Brazil (EM-DAT).
- Figure A4: Country level climate risks vs global climate policy uncertainty. Notes: compares country-level climate risk indicator to climate policy uncertainty by Gavriilidis (2021).
- Figure A5: The effect of major climate legislation on green firms. Notes: shows impact on green firms (Total impact) using local projections.

### Methodological notes and data treatment (from Appendix notes)
- Sample periods vary by specification, including January 2000 to December 2023; January 2017 to June 2023; January 2019 to December 2023.
- All specifications standardize variables to have zero mean and unit standard deviation.
- Dependent variable in main regressions: excess stock returns.
- Controls commonly include: time-varying market betas, log-total assets, debt-to-assets, stock price volatility, book-to-market ratio, PPE-to-assets, EBIT-to-assets, capex-to-assets, oil price changes, market volatility (VIX), and COVID lockdown intensity.
- Fixed effects used across specifications: country, year-month-day, industry, and in some specifications firm fixed effects.
- Winsorization applied at 1% for Book to market, Debt to assets, EBIT to assets, and Market beta; excess stock returns larger than 100% are dropped.
- Standard errors clustered by firm or by firm-year-month depending on specification.
- Statistical significance notation: ***: 0.01, **: 0.05, *: 0.1.

### Acknowledgments (Appendix B)
- Acknowledgement of benefits from Bergé (2018) and https://www.tidy-finance.org for high-dimensional fixed effects estimations and financial calculations.

*Source: wpiea2025037-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025037-print-pdf.pdf_