## 1. Introduction — The Carrot and the Stock: In Search of Stock-Market Incentives for Decarbonization ## Source details **Canonical URL:** [1. Introduction — The Carrot and the Stock: In Search of Stock-Market Incentives for Decarbonization](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022231-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022231-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022231-print-pdf.pdf.json) --- ### Role of the financial sector and policy levers - Annual global investment in the energy sector needs to triple by 2030 to around $4 trillion to reach net zero emissions by 2050 (IEA, 2021). - Two mutually reinforcing policies to reallocate finance toward sustainable projects: - Carbon pricing initiatives (World Bank, 2021) that internalize greenhouse gas emission costs. - Transparency initiatives increasing availability of high-quality, comparable emissions-related data (NGFS, 2022). ### EU context and EU ETS evolution (timeframe of focus: 2013 to 2021) - EU advances on transparency: EU taxonomy, EU green bond standard, Corporate Sustainability Reporting Directive (CSRD). - EU ETS: established 2005, “cap and trade”; phases: - Phase I (2005-2007) — pilot. - Phase II (2008-2012) — free allocation of allowances started to decline. - Phase III (2013-2020) — auctioning became default. - Phase IV (2021-2030) — more ambitious annual decline in emissions allowances. - Coverage: electricity and heat generation, energy-intensive industries, partially aviation. - Limitations: EU ETS addresses transparency only partially — not all sectors/gases included and no forward-looking firm emission paths. ### Research aim and dataset - Research aim: measure stock-market implications of higher carbon prices and increased availability of climate-related data; assess whether stock markets penalize most carbon intensive companies amid rising carbon costs. - Unique dataset: free and paid emissions allowances of 338 publicly traded European companies (2013–2021), compiled by matching EU ETS accounts with Orbis and market data. --- ### 2. Data, stylized facts, and key statistics ### Dataset construction and coverage - Matched almost 5,800 EU ETS account holders to specific firms in Orbis; 2,112 controlled by publicly traded companies. - Initial matched universe after consolidation: 634 publicly traded companies; final sample after exclusions: 338 publicly traded European companies. - Sample composition: - Companies from 24 European countries; UK, Germany, France, Poland and Italy jointly account for 55% of companies. - Sectoral breakdown: energy 12%, chemical producers 19%, mining 8%, transportation 6%, others 55%. - Firm size: about a quarter mid-sized (revenue of less than USD 1bn), three quarters large-sized enterprises. - Commodity and energy price controls: Brent (oil), ICE Dutch TTF one month futures (gas), German electricity futures (electricity) — weekly changes converted into EUR. ### Stylized facts on emissions, allowances, and paid carbon intensity - Verified emissions decreased over the decade for sample firms, especially electricity generators; sharper drop in 2020 due to COVID19. - Purchased vs free allowances: - Electricity: free allocation phased out; share of purchased allowances exceeded 90% since 2019. - Mining and chemistry: purchased allowance coverage still below 50%. - 2019 concentration: 74% of emission allowances purchased by only 1% of account holders. - Paid carbon intensity defined as (carbon price × volume of emissions not covered by free allowances) normalized by yearly revenues. - Sample mean paid carbon intensity ≈ 0.5% of revenue; median near 0. - For many firms paid carbon intensity roughly tripled in 2019-2021. - Electricity firms: paid carbon intensity above 3% of total revenues on average in 2019-2021 versus ≈ 1% in preceding years. - In some cases carbon costs exceeded 10% of revenue. - Country differences: on average paid carbon intensity stayed below 1% of revenue; higher in Greece, Poland and Czech Republic (Greece based on single company). ### Key summary statistics (preserved as presented) - Stock Returns Week-firm | N = 147,171 | Mean = 0.27 | Median = 0.20 | Min = -82.7 | Max = 212 | Stdev = 5.18 - Eurostoxx Return Week | N = 470 | Mean = 0.14 | Median = 0.36 | Min = -18.4 | Max = 7.36 | Stdev = 2.26 - Carbon Price Return Week | N = 470 | Mean = 0.78 | Median = 0.67 | Min = -33.5 | Max = 26.4 | Stdev = 6.70 - Gas Return Week | N = 470 | Mean = 0.55 | Median = -0.21 | Min = -36.3 | Max = 46.5 | Stdev = 7.50 - Oil Return Week | N = 470 | Mean = 0.14 | Median = 0.47 | Min = -26.5 | Max = 35.9 | Stdev = 5.49 - Electricity Return Week | N = 470 | Mean = 0.69 | Median = 0.16 | Min = -47.1 | Max = 47.5 | Stdev = 7.62 - CI (paid) Year-firm | N = 2,894 | Mean = 0.25 | Median = 0.00 | Min = -15.3 | Max = 25.5 | Stdev = 1.39 - CI (free) Year-firm | N = 2,894 | Mean = 0.32 | Median = 0.05 | Min = -0.40 | Max = 39.4 | Stdev = 1.27 --- ### 3. Econometric framework and main empirical findings ### Theoretical decomposition - Stock price = sum over t of discounted future profits (discount factor b_t × profit π_t). - Profit depends on carbon price τ_t, emission rate r_t, free allowances F_t, allowance stock A_{t-1}, demand, and input costs. - Derived decomposition of carbon-price shock impact on valuation into four terms: - [A] Revenue effect from competitors’ output responses. - [B] Additional input cost effects. - [C] Valuation gains on allowance inventory. - [D] Firm’s increased carbon costs (paid emissions net of free allowances). ### Empirical specification - OLS multifactor market-model regressions of firm weekly stock returns r_it on: - Eurostoxx weekly return r_index,t, - Carbon price weekly return r_carbon,t, - Interaction r_carbon,t × firm carbon intensity CI_{i,Y-1}, - Commodity weekly returns vector r_comm (gas, oil, electricity), - Industry and country-month fixed effects. - Main sensitivity of firm i in year Y to carbon price returns = β2 + β3 × CI_{i,Y-1}. - β2: co-movement for carbon intensity = zero. - β3: differential sensitivity with carbon intensity (captures paid carbon intensity effect). ### Full-sample (338 firms, 2013–2021) key estimates (Table 2, column (1) and (2)) - Eurostoxx index coefficient β1 = 0.914. - Carbon price return coefficient β2 = 0.009 (a weekly 1% carbon price increase associated with a 0.009% stock price increase for CI = 0). - Oil and electricity returns: no significant impact; gas returns: positive correlation with stock returns. - Cross-sectional: - β3 < 0 (higher carbon intensity → lower stock return for a given carbon price increase). - Decomposition: - Paid carbon intensity interaction β3_paid is negative and statistically significant. - Free carbon intensity interaction β3_free not statistically significant. - Quantitative thresholds and examples: - Firms spending more than 1.7% of revenue on emission allowance purchases show stock price declines associated with higher carbon price. - Example: firm with paid carbon intensity = 10% → a 1% carbon price increase coincides with a stock price drop of 0.04%, all else equal. - In recent years, a 1% carbon price increase linked to up to 0.21% stock price drop for the most polluting companies (subsample magnitude). ### Robustness and supplementary findings - Results robust to alternative clustering of standard errors, combinations of fixed effects, and normalization by market capitalization (though threshold differs: 7.3% of market capitalization vs 1.7% of revenue). - Paid-intensity significance consistent with markets using current paid carbon intensity as proxy for future paid paths. - No additional explanatory power from firms’ R&D expenditures in cross-sectional variation (A.3.5). --- ### 4. Sub-sample, event, and permanence results ### Industry-specific results - Electricity sector (43 firms): - Sector carbon costs ≈ 3% of revenues in 2019-2021. - Regression: carbon price increases linked to positive and statistically significant stock returns for CI = 0 (β2 > 0); interaction with paid intensity negative (β3paid < 0). - Calibrated threshold: firms with carbon costs above 5.5% of revenues see stock prices decrease when carbon price increases. - Free allowances do not alter main relationship. - Chemicals (67 firms), Mining (26 firms), Other (202 firms): - Carbon costs on average ≤ 0.5% of revenues. - Coefficients for carbon price changes and stock performance generally non-significant in full sample; some positive association in 2013-2017 for non-electricity sectors. ### Sub-periods: 2013-2017 vs 2018-2021 - 2013-2017: ETS price changes positively co-move with stock returns; no significant carbon-intensity heterogeneity. - 2018-2021: rapid carbon price increase period (Market Stability Reserve decided 2018; started Jan 2019; increased EU 2030 target adopted June 2021). - β3paid becomes negative and highly statistically significant; β3free not significant. - During 2018-2021, weekly 1% carbon price increase associated with a 0.08% stock price drop for firm with paid carbon intensity = 10%. - Estimated coefficients: second sub-period β2 = 0, β3paid = −0.830; overall period β2 = 0.009, β3paid = −0.538. - Sensitivity for firm with highest paid carbon intensity in 2018-2021 reaches 0.21%. ### Country groups - “High carbon cost” countries defined by at least one company incurring > 3% revenue carbon costs in any year: Poland, Czechia, Great Britain, Germany, France, Greece, Norway. - In high-cost subsample, relationship significant; carbon price increases associated with stock price decline for companies with relative carbon costs above 2% of revenue. - “Low carbon cost” subsample: no statistically significant relationship. ### Permanence and lags - Regressions with up to four lags of carbon price return: - Contemporaneous carbon price × paid intensity significant; carbon price up to second lag shows positive association with stock returns without interaction. - Overall interpretation: effect as function of firm’s paid carbon intensity is persistent (permanent in this specification). ### Event study with regulatory updates (daily data) - Daily-frequency regressions: Eurostoxx β1 = 0.874; carbon contemporaneous effect β2 = 0.007; paid-intensity interaction negative; free-intensity not significant generally. - Regulatory update events (89 events, 2013–2021): - On regulatory-update days, carbon price increase coincides with stock price decrease β2 = −0.012. - Paid-intensity interaction on update days β3paid = −0.752 (negative and significant). - Disaggregation by update type: - No update days: β3paid = −0.271. - Price-impact updates: β3paid = −0.687. - Free-allocation updates: β3free = −0.912 (free-intensity becomes significant only on these days). - Interpretation: markets incorporate regulatory news; paid intensity is generally priced; free intensity matters on free-allocation regulatory-update days. --- ### 5. Conclusions and policy implications ### Main conclusions - Strongly statistically significant relationship between carbon price changes and stock returns; relationship depends on firms’ paid carbon intensity. - Total emissions matter only through paid carbon intensity; free allowances generally do not affect sensitivity except on specific regulatory-update days. - Threshold and magnitudes: - Firms with paid carbon intensity > 1.7% of revenue decline on average when carbon price increases; firms paying less than 1.7% tend to see stock price rises. - Firm with carbon costs = 10% of revenue → stock price decline of 0.04% on a 1% carbon price increase (full-period example). - In 2018-2021, sensitivity magnitudes larger (up to 0.21% for most polluting firms). ### Policy implications — strengthening decarbonization incentives - Stock price performance creates an incentive channel for shareholders and management of highly polluting firms to decarbonize. - Three levers to increase carbon costs within EU ETS: - Ensure allowance prices continue to rise (e.g., strengthen Market Stability Reserve or introduce explicit price floor). - Phase out free allowances (quickly for domestic air travel; gradually for manufacturing with carbon border adjustment). - Include further sectors in EU ETS (heating, transport, agriculture, waste). - Higher carbon prices expected to have larger impact than phasing out free allowances for some industries (e.g., mining, chemicals remain below 1% of revenue even with zero free allocation at current prices). - Non-linear strengthening: largest effects where carbon costs are highest (electricity sector, recent period, specific countries). ### Transparency versus carbon pricing - High-quality emissions data alone insufficient: markets price only emissions firms must pay for (paid allowances); audited emissions data matter in conjunction with carbon pricing. - Free allocations become relevant for valuations primarily when regulatory updates change free-allocation expectations. ### Financial stability - Aggregated financial stability risk from higher carbon prices appears limited at this stage. - Example metrics: - Firm with carbon costs = 10% of revenues → 1% carbon price increase → stock price decrease = 0.04%. - Carbon prices rose > 300% in 2020-2021. - Model-implied average impact of a 1% carbon price rise on sample firms’ stocks ≈ -0.003%. - Implication: while cumulative effects can produce divergence in share prices, no indication of widespread stock market crash absent major non-linearities; estimates applicable for climate stress tests and transition-risk assessments. *Italic: Source — wpiea2022231-print-pdf (IMF Working Paper — The Carrot and the Stock: In Search of Stock-Market Incentives for Decarbonization).* ### 1. Introduction ........................................................................................................ ### 1. Introduction ### Role of the financial sector and policy levers - Annual global investment in the energy sector needs to triple by 2030 to around $4 trillion to reach net zero emissions by 2050 (IEA, 2021). - Funds flow to investments that provide the highest return for a given level of risk; policies that lower returns of highly emissive ventures or provide transparency on higher risks will reallocate funding toward sustainable projects. - Two key policies: - Carbon pricing initiatives (World Bank, 2021) that internalize greenhouse gas emission costs and reduce relative profitability of highly emissive firms. - Transparency initiatives that increase availability of high-quality, comparable emissions-related data (NGFS, 2022), allowing markets to adjust prices to reflect climate-related risks. - Carbon pricing and data disclosure are mutually reinforcing in steering financial markets toward a low-carbon transition. ### The EU context: transparency and carbon pricing - The EU advanced on both fronts: publication of the EU taxonomy, EU green bond standard, and Corporate Sustainability Reporting Directive (CSRD) standardized green financing and reduced greenwashing. - Rising price of carbon within the EU Emissions Trading System (EU ETS) has increased carbon costs for polluting firms. - The EU ETS functions as both a carbon pricing mechanism and a transparency tool by providing audited, comparable information on emissions, free allocation of allowances, and transactions—partially addressing the lack of reliable and comparable climate-related information. - Timeframe of focus: 2013 to 2021, when the system had matured into a well-functioning emissions market. ### The EU ETS: design and evolution - The EU ETS is a “cap and trade” scheme established in 2005. - Trading phases and dates: - Phase I (2005-2007) — pilot. - Phase II (2008-2012) — free allocation of allowances started to decline. - Phase III (2013-2020) — auctioning became the default method for allocating allowances. - Phase IV (2021-2030) — more ambitious annual decline in emissions allowances. - Coverage: electricity and heat generation, energy-intensive industries, and partially the aviation sector; installations must surrender allowances annually for emissions. - Allowance allocation: - A fraction of allowances are allocated for free; the residual must be purchased in public auctions or on the secondary market. - The free allocation fraction varies by sector and typically increases with the risk of carbon leakage; since 2013 the fraction is very low for the electricity sector. - Limitations: the EU ETS addresses the transparency problem only partially because (i) not all sectors and all greenhouse gases are included and (ii) no forward-looking data on firms’ future emission paths is available. ### Research question, dataset, and empirical scope - Research aim: measure stock-market implications of higher carbon prices and increased availability of climate-related data, and assess whether stock markets penalize the most carbon intensive companies amid rising carbon costs. - Unique dataset compiled on free and paid emissions allowances of 338 publicly traded European companies. - Analytical focus: extent to which stock markets consider carbon price dynamics and firms’ carbon intensities; implications for a potential stock market incentive channel to induce decarbonization. ### Key empirical findings (as reported) - A strongly statistically significant relationship exists between carbon price changes and stock returns; the relationship depends on firms’ carbon intensity. - Investors price in the costs of purchased emissions allowances; free allowances do not impact the relationship between carbon prices and stock returns. - Transparent information on emissions levels alone does not appear sufficient for investors to consider carbon price changes in stock valuation; markets react when policies imply current or future financial costs for firms. - Quantitative thresholds and magnitudes: - Companies that spend more than 1.7% of their revenue on emission allowance purchases show stock price declines associated with higher carbon price. - In recent years, a 1% carbon price increase is linked to a stock price drop of up to 0.21% for the most polluting companies. - Robustness: the magnitude varies across subsamples, but the relationship's nature remains unchanged. ### Positioning within the literature - Prior literature has not reached consensus on sign and magnitude of the carbon price–stock price relationship; most papers find the magnitude fairly small (1% change in the carbon price tends to coincide with a stock price change of only a few basis points). - Two main empirical approaches in the literature: - Multifactor market model (MMM) approach: controls for overall market returns and energy price dynamics; finds that relationship evolved over EU ETS phases, often switching from positive to negative as allocation shifted from free to auctioning. - Capital Asset Pricing Model (CAPM) approach: used to estimate a “carbon premium” (excess return of dirty firms over clean ones); evidence of an unstable premium over time linked to evolving EU ETS rules. - Literature highlights: - Positive relationship between carbon price changes and stock returns during periods with larger free allocations; negative or insignificant relationships when auctioning increased. - Stock returns of more carbon-intensive companies are more likely to be negatively correlated with carbon price increases. ### Paper organization - Section 2: related literature. - Section 3: data collection and final dataset. - Section 4: econometric framework and estimates. - Section 5: results summary and policy conclusions. *IMF Working Paper — The Carrot and the Stock: In Search of Stock-Market Incentives for Decarbonization* ### 2012. Oestreich and Tsiakas (2015) find a highly statistically significant carbon premium in 2003-2009 and also ### The Carrot and the Stock: In Search of Stock-Market Incentives for Decarbonization ### Evidence on carbon premium and stock-return effects - Oestreich and Tsiakas (2015) find a highly statistically significant carbon premium in 2003-2009 and link it to the free allocation of allowances; the carbon premium dissipates after 2009 (one year into Phase II of the EU ETS). - Ryszka (2021) employs the CAPM approach on data for 900 European companies and fails to identify a statistically significant carbon premium after controlling for firm-specific characteristics. - Bolton and Kacperczyk (2021), using US firm-level data, find a positive and statistically significant effect of carbon emissions on returns, interpreted as a “carbon premium” reflecting investors’ demand for higher returns due to exposure to carbon risk; they argue the carbon premium is directly related to the total level (and not the intensity) of emissions. - Gorgen et al. (2020) construct a firm “brownness-greenness” indicator and find that (i) brown firms out-perform green firms, on average, and (ii) firms becoming “browner” relative to the preceding year experience negative returns; the two effects are found to have similar magnitudes, revealing an ambiguous effect of carbon risk on stock returns. - CAPM vs MMM: CAPM focuses on excess returns of highly-emitting over comparable clean firms and interprets excess returns as compensation for undiversifiable “carbon risk”; CAPM estimates depend heavily on constructing portfolios of stocks with very similar firm characteristics. MMM explicitly measures co-movement between carbon price changes and stock prices, allowing a direct link from stock performance to carbon price changes, which is the focus of this paper. ### Identifying clean versus carbon-intensive firms - Classification approaches used in the literature: - Tian et al. (2016): companies that generate 50% or more of their electricity from fossil fuels are carbon-intensive. - Da Silva et al. (2016): use corporate information on the use of renewable and non-renewable energy sources. - Oestreich and Tsiakas (2015): classify as “dirty” firms that received annually more than one million free allowances during the initial two phases of the EU ETS. - Witkowski et al. (2021): develop a carbon risk exposure ratio = (actual emissions − free allowances) normalized by companies’ total assets. - Note: Carbon intensity classifications based on EU ETS data have not yet been used in studies following the MMM approach (to the authors’ knowledge). ### Alternative analysis techniques and recent textual approaches - VAR-GARCH models: Dutta et al. (2018) find that carbon emissions prices transmit their volatility to stock performance of European electricity companies. - Nonlinear ARDL: Wen et al. (2020) find asymmetry in China—an increase in the carbon price affects stock prices stronger than a decrease. - Text-based metrics: Sautner et al. (2020) build text-based carbon intensity metrics from earnings conference calls; Deng et al. (2022) using text-based metrics show that until late-2021 stocks of US companies more vulnerable to a low-carbon transition performed better, likely reflecting investor expectations of a slowdown in US transition policies; in Europe effects were the opposite. - Faccini et al. (2021) find that the climate policy factor is priced in the U.S. stock market, with investors demanding positive risk premia for companies exposed to US climate risk policy. ### Contribution of this paper - Extends MMM literature by covering Phase III of the EU ETS (2013-2020) and the start of Phase IV (launched in 2021); previous MMM studies were limited to 2005-2017 when free allocation remained high and allowance prices stayed low. - Expands sample size and sectoral coverage beyond electricity: final dataset consists of 338 publicly traded companies affected by the EU ETS (sample construction details below); previous MMM studies had much smaller samples (maximum 65 companies in Zhu et al. 2018). - Introduces a quantitative measure of corporate carbon costs based on EU ETS data for 2012-2021: - Constructed indicator reflects annual costs related to purchase of emission allowances. - Normalization uses revenue and market capitalization (preferred over total assets, broadly similar to Witkowski et al. (2021) but with different normalizers). - Matching methodology draws from Abrell et al. (2021) and Jaraite et al. (2013) when linking EU ETS accounts with ORBIS corporate data. ### Data: dataset construction and coverage - Sources used: - EU ETS data on emissions, emission allowances and transactions. - Orbis company databases (Bureau van Dijk). - Yahoo Finance financial market indicators. - Abrell et al. (2021) EUTL-derived dataset. - Matching challenges and results: - EU ETS data provided at disaggregated installation level and not directly linked to publicly traded companies; tasks: match installations to parent companies, keep parent companies quoted on exchanges, consolidate installations belonging to same parent. - Heterogenous company identification numbers in EUTL required substantial manual work. - Matched almost 5,800 account holders within the EU ETS with specific firms from Orbis; 2,112 are controlled by publicly traded companies. - After consolidation, initial matched universe: 634 publicly traded companies. - The matched 5,800 account holders accounted for around 95% of purchased emission allowances within the EU ETS in any given year since 2013; two thirds of these purchased allowances belong to publicly traded companies. - Final sample narrowed to European traded companies; excluding non-European traded companies (not more than 2.5% of allowances in any year) and removing firms with missing data resulted in final dataset of 338 publicly traded companies. - Sample composition: - Companies from 24 European countries; UK, Germany, France, Poland and Italy jointly account for 55% of companies in the dataset. - Sectoral breakdown: energy 12%, chemical producers 19%, mining 8%, transportation 6%, plus numerous manufacturers from other sectors. - Firm size: about a quarter mid-sized (revenue of less than USD 1bn), three quarters large-sized enterprises. - Commodity and energy price controls: - Weekly changes of oil, electricity and gas prices converted into EUR used as control variables: Brent price for oil; ICE Dutch TTF one month futures for gas; German electricity futures prices from Europe Exchange AG for electricity. ### Stylized facts on emissions, allowances, and paid carbon intensity - Verified emissions: - Companies in the sample, particularly electricity generators, decreased verified emissions over the latest decade. - Sharper drop in 2020, pronounced in electricity generation and aviation, triggered by the COVID19 pandemic. - Purchased vs free allowances: - Shares of purchased allowances in total allowances differ across industries. - Electricity sector: free allocation phased out; share of purchased allowances exceeded 90% since 2019. - Mining and chemistry: coverage by purchased allowances still below 50%. - Negative share of purchased allowances in 2012 reflects free allocation exceeding actual emissions during Phase II, allowing accumulation of allowances. - Concentration: in 2019, 74% of emission allowances were purchased by only 1% of account holders. - Paid carbon intensity (relative carbon costs): - Calculated as (carbon price × volume of emissions not covered by free allowances) normalized by firms’ yearly revenues; assumptions: purchases spread evenly throughout the year and approximated by mean annual price; alternative normalization metrics used in robustness checks (Annex A.3.3). - For the majority of sample firms, paid carbon intensity roughly tripled in 2019-2021 compared to earlier years due to sharp increase in carbon price since 2018. - Sample mean paid carbon intensity remained around 0.5% of revenue; median has not deviated much from zero. - In a few cases carbon costs exceeded 10% of revenue. - Sectoral differences: - Electricity generating firms: paid carbon intensity above 3% of total revenues on average in 2019-2021 compared to around 1% in preceding years. - Other industries: average paid carbon intensity does not exceed 0.5% of revenue. - Context: average profit margin for European electricity generating firms in sample is around 8%. - Country differences: - On average paid carbon intensity stayed below 1% of revenue even at end of sample. - Costs noticeably higher only in Greece, Poland and Czech Republic (Greece result based on single company). - Poland and Czech Republic noted among countries with lowest GDP per unit of energy-related CO2 emissions (OECD, 2015). - Interpretation: Low paid carbon intensity for majority of firms suggests limited direct impact on profitability and likely limited influence on stock prices outside electricity sector; this hypothesis is tested in econometric analysis in Section 4. ### Key summary statistics (preserved as presented) - Frequency | N | Mean | Median | Min | Max | Stdev. - Stock Returns Week-firm | 147,171 | 0.27 | 0.20 | -82.7 | 212 | 5.18 - Eurostoxx Return Week | 470 | 0.14 | 0.36 | -18.4 | 7.36 | 2.26 - Carbon Price Return Week | 470 | 0.78 | 0.67 | -33.5 | 26.4 | 6.70 - Gas Return Week | 470 | 0.55 | -0.21 | -36.3 | 46.5 | 7.50 - Oil Return Week | 470 | 0.14 | 0.47 | -26.5 | 35.9 | 5.49 - Electricity Return Week | 470 | 0.69 | 0.16 | -47.1 | 47.5 | 7.62 - CI (paid) Year-firm | 2,894 | 0.25 | 0.00 | -15.3 | 25.5 | 1.39 - CI (free) Year-firm | 2,894 | 0.32 | 0.05 | -0.40 | 39.4 | 1.27 *Source: IMF Working Paper — The Carrot and the Stock: In Search of Stock-Market Incentives for Decarbonization (dataset and analysis summarized from the PDF content).* ### 4. Econometric Analysis ### 4. Econometric Analysis ### 4.1 Theory - Stock price written as the sum of discounted future profits (stock price = sum over t of discount factor b_t times profit π_t). - Firm profit modeled as a function of carbon price, demand price P_t(q_t, q_t^comp), unit cost C(ω), emission rate r_t (tCO2e/unit), carbon price τ_t (per tCO2e), free allowances F_t, and allowance stock A_{t-1}. - Appendix derives impact on stock price of an exogenous carbon price shock producing an expression decomposed into four terms: - [A] Revenue effect if competing firms reduce output in response to higher carbon price. - [B] Additional costs due to carbon price impact on inputs. - [C] Valuation gains on allowance inventory. - [D] Firm’s increased carbon costs (paid emissions net of free allowances). - Implication: impact of a carbon price shock on stock price depends on firm-specific factors beyond immediate carbon cost increase (relative reaction to competitors, input structure sensitivity, allowance holdings, and decarbonization path). ### 4.2 Specification for the empirical analysis - Regression setup: OLS multifactor market-model style regressions of firm weekly stock returns r_it on: - Eurostoxx index weekly return r_index,t, - Carbon price weekly return r_carbon,t, - Interaction of carbon returns with firm carbon intensity CI_{i,Y-1}, - Commodity weekly returns vector r_comm (gas, oil, electricity), - Industry and country-month fixed effects FE_{i,t}. - Main estimated sensitivity of firm i in year Y to carbon price returns: derivative equals β2 + β3 × CI_{i,Y-1}. - β2 captures co-movement for a company with carbon intensity of zero (combined effect of terms [A], [B], [C] from theory). - β3 captures degree to which stock markets treat firms with different carbon intensities differently (reflects sensitivity to paid carbon intensity, term [D]). - Estimation choices: - Time-month interacted fixed effects to account for country-specific business cycles; industry fixed effects for sector heterogeneity. - Errors clustered at the firm level. - Use of CI_{i,Y-1} (latest publicly available yearly statistics) to avoid using information unavailable to market participants. ### 4.3 Full sample estimates - Sample: 338 publicly traded firms over 9 years. Total carbon intensity - Total carbon intensity CI_{i,Y}^{total} computed as (E_{i,Y}^{total} × P̂_Y) / R_{i,Y}, where: - E_{i,Y}^{total} = firm i’s verified ETS emissions in year Y-1, - P̂_Y = mean price of carbon in year Y-1, - R_{i,Y} = total revenue of firm i in year Y-1 (EUR). - Regression results (Table 2, column (1)): - Eurostoxx index coefficient β1 = 0.914 (individual stock returns co-move with stock index in a highly statistically-significant way). - Carbon price return coefficient β2 = 0.009 (a weekly increase in carbon prices of 1% is associated with an increase in stock prices of 0.009%). - Oil and electricity price returns: no significant impact on stock returns. - Gas price returns: correlate positively with stock returns. - Key cross-sectional relationship: β3 < 0 (the higher the carbon intensity, the lower the stock price return for a given increase in carbon prices). - Note: β3 is not significantly different from 0 under some alternative standard error clustering approaches (see Appendix A.3.1). Free and paid carbon intensity - Decomposition: CI_{i,Y}^{total} = CI_{i,Y}^{free} + CI_{i,Y}^{paid}, where: - CI_{i,Y}^{free} uses E_{i,Y}^{free} (free allowances received by firm i in year Y-1) from EU ETS database, - CI_{i,Y}^{paid} uses E_{i,Y}^{paid} = E_{i,Y}^{total} − E_{i,Y}^{free}. - Regression results (Table 2, column (2)): - Co-movements with Eurostoxx, oil, electricity, gas, and β2 unchanged in value and significance relative to total-intensity regression. - Paid carbon intensity effect: β3_paid is negative and statistically significant (higher paid carbon intensity → significantly lower stock returns when carbon prices increase). - Free carbon intensity effect: β3_free is not significantly different from zero (free emissions do not change stock price sensitivity to carbon price returns). - Quantitative interpretation: - Firms with a paid carbon intensity above 1.7% face an association where an increase in carbon prices is linked to a drop in stock price (threshold 1.7%). - Example: for a firm with paid carbon intensity of 10%, a carbon price increase of 1% will coincide with a stock price drop of 0.04%, all else equal. - Uncertainty band width on the plotted sensitivity depends on CI squared via the variance formula: δ = sqrt(Var(β2) + CI^2×Var(β3) + 2×CI×Cov(β2,β3)). - Interpretation and robustness: - Stock markets take carbon prices into account when valuing firms’ stocks; co-movement varies with paid carbon costs. - Markets appear to price only the part of emissions firms need to pay for within the EU ETS; free allowances do not affect sensitivity. - The statistical significance of β3_paid is consistent with markets using today’s paid carbon intensity as a proxy for future paid carbon intensity paths. - No statistically significant additional explanatory power from firms’ R&D expenditures for cross-sectional variation (see Appendix A.4.6). - Results robust to alternative clustering of standard errors, different combinations of fixed effects, and choice of normalization factor (revenues or market capitalization) as shown in Appendix Table 6. - Appendix also examines timing of purchases, allowance inventory build-up, and firm R&D costs. *Source: 4. Econometric Analysis, wpiea2022231-print-pdf* ### 4.4 Sub-sample estimates ### 4.4 Sub-sample estimates ### Results by industry - Dataset: companies from electricity (43 firms), chemicals (67 firms), mining (26 firms), and “Other industries” (202 firms). - Electricity sector: - Carbon costs close to 3% of revenues in 2019-2021 for the sector as a whole. - Regression result: carbon price increases linked to positive and statistically significant stock returns (β2 > 0). - Interaction: the relationship varies with firm’s carbon intensity; paid component interaction is negative (β3paid < 0). - Calibrated threshold: on average, firms with carbon costs above 5.5% of revenues see stock prices decrease when carbon price increases (as described in Equation (5)). - Free allowances do not alter this relationship. - Chemicals and mining sectors: - Carbon costs on average do not exceed 0.5% of revenues. - Coefficients for carbon price changes and stock performance are non-significant (see columns 4 to 6 in Table 2). ### Results by sub-periods (2013-2017 vs. 2018-2021) - Sub-period split rationale: 2013-2017 had low and stable carbon prices; 2018-2021 saw a rapid carbon price increase (Market Stability Reserve decided in 2018 and started in January 2019; increased EU 2030 target proposed in 2020 and adopted in June 2021). - Whole-period vs sub-period estimates: - 2013-2017: ETS price changes positively co-move with stock returns (very small effect); no statistically significant difference by firm carbon intensity. - 2018-2021: coefficient on carbon price interacted with paid carbon intensity (β3paid) becomes negative and highly statistically significant; free allocation interaction (β3free) not statistically significant. - Interpretation: in 2018-2021 carbon price becomes inversely correlated with stock returns for companies that must buy allowances (positive paid component). - Contrast with full sample: negative effect in full sample observed only for companies exceeding 1.7% carbon intensity threshold, whereas in 2018-2021 effect applies to all firms with positive paid carbon intensity. - Magnitude example: - During 2018-2021, a weekly increase in carbon prices of 1% was associated with a drop in stock prices of 0.08% for a firm with paid carbon intensity of 10%. - Distributional evidence: - Full period: 2,894 firm-years. - Recent sub-period (2018-2021): 1,335 firm-years; firms in this sub-period have higher paid carbon intensities (0.4% vs. 0.2% in overall sample). - Estimated coefficients: second sub-period (β2 = 0, β3paid = −0.830); overall period (β2 = 0.009, β3paid = −0.538). - Sensitivity for the firm with highest paid carbon intensity in 2018-2021 reaches 0.21%. ### Results by country groups - Country classification: a country is "high carbon cost" if at least one company from the country incurred carbon costs exceeding 3% of revenue in at least one year over the full period. - High carbon cost sub-sample: firms from Poland, Czechia, Great Britain, Germany, France, Greece and Norway. - Some of these countries have higher than EU average share of coal in primary energy (BP, 2021) and low CO2 productivity (OECD, 2015). - Core findings hold: carbon price–stock performance relationship is statistically significant and depends on carbon intensity; paid component is priced by the market. - Magnitude: carbon price increase associated with stock price decline for companies with relative carbon costs above 2% of revenue. - Low carbon cost sub-sample: no statistically significant relationship found. ### 4.5 Permanence of the effect - Method: main regression re-estimated adding up to four lags of carbon price return, both without interaction and interacted with paid and free carbon firm intensity. - Findings: - Carbon price up to the second lag has a positive association with stock returns. - Only the contemporaneous carbon price return interacted with paid carbon intensity is significantly correlated with stock returns, and with the same sign as previous results. - Carbon price return is not significantly associated with stock returns at any lag. - Overall interpretation: the effect of carbon price as a function of the firm’s paid carbon intensity is permanent. ### 4.6 Event study using regulatory updates (daily data and regulatory update events) - Daily-frequency robustness check: - Eurostoxx co-movement highly significant (β1 = 0.874). - Carbon price contemporaneous effect: small but highly statistically significant positive effect on stock prices (β2 = 0.007). - Paid carbon intensity interaction negative (β3paid < 0); free intensity not statistically significant. - Regulatory update events: - Dataset: 89 update events covering 2013-2021; events concern changes to the supply of emission allowances only (Kanzig, 2022 extended to end-2021). - Event-specification estimates: - On regulatory update days, carbon price increase coincides with a small but highly statistically significant decrease in stock prices (β2 = −0.012). - Paid carbon intensity interaction on update days: β3paid = −0.752 (negative and significant). - Free carbon intensity still not statistically significant on generic update days. - Regulatory-update-type disaggregation: - Dummy stype can take three values: “no update”, “free” (updates concerning allocation of free allowances), and “price” (updates impacting supply/demand of auctioned allowances). - Estimates (Table 4) indicate: - Paid-intensity interaction on days with no regulatory update: β3paid = −0.271. - Paid-intensity interaction on days with updates impacting allowance price: β3paid = −0.687. - Free-intensity interaction becomes significant only on days with regulatory updates concerning free allowances: β3free = −0.912. On such days, reductions in free allowances (expected to raise auction demand and carbon price) are associated with larger stock price declines for firms with high free carbon intensities. - Interpretation: markets incorporate regulatory news and price paid and—on specific regulatory-update days—free carbon intensity in firm valuations. ### 5.1 Conclusions - Sample and main result: - Novel dataset: 338 European companies between 2013 and 2021. - Strongly statistically significant relationship between carbon price changes and stock returns; relationship depends on firms’ paid carbon intensity. - Total emissions do not matter beyond paid carbon intensity. - Sensitivity and thresholds: - A firm with carbon costs of 10% of revenue would see its stock price decline by 0.04% on average when carbon prices rise by 1%. - Over the entire observation period, firms with paid carbon intensities exceeding 1.7% decline on average when carbon price increases; firms paying less than 1.7% of revenue for carbon allowances see stock price rises. - Explanation for positive correlation at zero carbon intensity: in markets with inelastic demand, cost pass-through can produce windfall profits for low-carbon firms (references: Fabra and Reguant, 2014; Sijm et al., 2006; Bushnell et al., 2013). - Sub-sample drivers: - Relationship driven by segments with high carbon costs: electricity sector (free allowances phased out in 2013, high paid intensities), recent period (2018-2021), and countries with high paid carbon intensities (including Poland, Greece, Czechia and Germany). - Reasons for absence in many sub-samples: - Carbon costs remain low as share of revenues for most industries (median close to 0% even in 2018-2021). - Corporate carbon cost information published yearly with time lags and not straightforward to compile. - Uncertainty in estimated carbon costs due to timing of allowance purchases, firms’ hedging via derivatives, and omission of indirect costs from purchased inputs (dataset does not capture costs from purchased inputs, such as electricity). - Role of free allocations: - Free emissions nearly never affect the stock–carbon price relationship, except on regulatory-update days concerning free allocations where free intensity becomes relevant. - If phasing out of free allocation continues and regulatory updates on free allocation become more frequent, free carbon intensity may become more important for investors. - Policy and incentive interpretation: - Controlling for market, energy prices, industries and country cycles, stocks of firms with high carbon costs under-perform in weeks where carbon price increases. - Event-study evidence using regulatory updates supports viewing carbon price changes as a driver of stock performance rather than reverse causality. - Implication: stock markets discriminate carbon-intensive firms during carbon price increases, creating a quantifiable incentive channel ("the carrot and the stock") for shareholders and management to decarbonize operations, as lowering carbon costs yields measurable financial upside that can be weighed against decarbonization costs; this also affects management remuneration linked to stock performance. *Italic: Source — wpiea2022231-print-pdf, section 4.4 and adjoining subsections.* ### 5.2 Policy implications ### wpiea2022231-print-pdf - 5.2 Policy implications ### Strengthening the decarbonization incentive - Stock price performance serves as an incentive channel for shareholders of highly polluting companies to decarbonize firms’ operations. - Any policy leading to higher carbon costs for emitting firms will tend to strengthen this incentive. - Three ways to increase carbon costs for firms within the EU ETS: - Ensure the price of allowances continues to rise (examples: strengthening the Market Stability Reserve or introducing an explicit price floor). - Phase out free allowances: - Could be done quickly for domestic air travel without impact on competitiveness. - Could be done gradually for the manufacturing sector alongside the introduction of a carbon border adjustment. - Include further sectors in the EU ETS (currently heating, transport, agriculture and waste management are largely excluded). - Higher carbon prices are expected to play a bigger role than the phase-out of free allowances: - If the volume of emissions and the price of carbon stay at their current level while the free allocation of allowances drops to zero, the relative carbon costs in industries like mining and chemicals will remain below 1% of revenue. - The impact of carbon prices on stock performance was most pronounced where carbon costs were highest; this could strengthen the incentive channel in a non-linear way. ### Mitigation: transparency vs. carbon pricing - High-quality, reliable and comparable data on firms’ carbon intensity is not always sufficient to ensure financial markets contribute to climate change mitigation. - The EU ETS provides audited emissions and carbon costs published since 2005, yet: - In most cases stock markets do not consider firms’ total emissions when determining the impact of carbon price changes on stock prices: only those emissions for which firms need to pay do matter. - The only exception to this pattern is on the days of regulatory changes with regard to free allocation of allowances; on these occasions free carbon intensity makes the relationship between stock price and carbon price changes more negative. - Conclusion: stock markets can help channel private investments to lower-emitting firms using high-quality emissions data, but only in conjunction with a carbon pricing scheme. ### Financial stability - Aggregated financial stability risk for stocks from higher carbon prices appears to be limited at this stage. - Empirical findings: - For a firm with carbon costs representing 10% of revenues, an increase in carbon prices of 1% is linked with a stock price decrease of 0.04%. - Carbon prices increased by over 300% in 2020-2021. - According to the estimated model, the average impact of a 1% rise in carbon prices on stocks of firms in the sample would be no more than -0.003%. - While the impact is robust and, when accrued over time, can lead to share prices diverging, barring a tipping point or major non-linearity, further increases in carbon costs do not appear to indicate a risk of a widespread stock market crash. - Applicability: - The specifications and estimates can be used when modeling the impact of rising carbon prices on stock portfolios, for instance in climate stress tests. - Results are applicable for assessing carbon price-related transition risk, i.e. risks to financial institutions “related to the process of adjustment towards a low-carbon economy” (BCBS, 2021). *Source: wpiea2022231-print-pdf - 5.2 Policy implications* ### References ### References ### Key citations - The References list cites empirical and theoretical studies on corporate behavior in emissions trading systems, carbon pricing, stock-market responses to carbon costs, and related data sources. Notable recurring themes in the citations include the EU ETS, firm-level impacts of CO2 prices, carbon risk in equity markets, and data sources such as the EU Transaction Log, Orbis, ESMA, IEA, NGFS, World Bank, and OECD. - Selected numeric and bibliographic items present in the References (as cited in the source content): - BCBS (2021). Climate-related risk drivers and their transmission channels. - Bolton, P. and Kacperczyk, M. (2021). Journal of financial economics, 142(2):517–549. - BP (2021). Statistical Review of World Energy | Energy economics. - ESMA (2022). Final report emission allowances and associated derivatives. - IEA (2021). Net zero by 2050. - NGFS (2022). Final report on bridging data gaps. - World Bank (2021). State and Trends of Carbon Pricing. - The References include many empirical papers on the EU ETS and stock-market effects (examples include Abrell et al. 2021; Bushnell et al. 2013; Fabra and Reguant 2014; Sijm et al. 2006; Oestreich and Tsiakas 2015; Veith et al. 2009; Wen et al. 2020; Witkowski et al. 2021). ### A.1 Orbis and EU ETS data matching - Matching process steps: - I. Matching installations with specific companies - Aggregate raw EUTL data by account holder (operator holding account). - Merge installations by account holder to aggregate emission allowances at account holder level. - Match account holders with specific companies via Orbis using account holder name, company identification number and address. - Outcome: database including EUTL data on allocated and surrendered emission allowances and key financial data from Orbis for each account holder. - II. Identifying publicly quoted parent companies - Identify publicly quoted companies controlling account holders using Orbis. - Exclude account holders not linked to publicly traded companies (no stock prices). - When account holders linked to more than one quoted company, select the quoted company nearest to the account holder in the ownership chain (“lowest traded owners”, LTO). - Stock prices of LTOs were collected using Yahoo Finance. - III. Consolidating account holders belonging to the same parent company - Aggregate EUTL data on purchased emission allowances at the LTO level to obtain a proxy for total purchased emission allowances per LTO. - These steps are summarized in Figure 13 (referenced in source). ### A.2 Theory derivation - Analytical approach: - Start from firm profits and compute derivative with respect to the carbon price. - Evaluate at the profit-maximizing equilibrium. - With equation (12) inserted, the first term in equation (11) equals zero. - Assuming a change in carbon prices today leads to a parallel shift in the carbon futures curve (dτt / dτ0 = 1), the second term is also zero. - Assuming other effects do not depend on the carbon price (dδt / dτ0 = 0), obtain the simplified expression where π⋅ and q⋅ are consistent with profit maximization. - The impact of an exogenous carbon price shock on the stock price is then given by the derived expression (as presented in the source). ### A.3 Industry-subperiod estimates - Main findings by industry and sub-period: - Electricity sector: results for the full sample are confirmed. The relationship between carbon price and electricity companies’ stock performance becomes carbon-intensity dependent in the second sub-period (β3paid < 0). - Chemicals, mining, other sectors: relationship between carbon price changes and stock performance is positive and statistically significant in the first sub-period but not affected by firm carbon intensity. Almost no statistically significant relationships in the second sub-period for these industries. - Additional references to tables and figures for subperiod regressions are present in the source (see section A.3 and related tables). ### A.4 Robustness checks - Controls selection (A.4.1) - Robustness to choice of different controls varying at industry, country, firm and monthly level was checked (see Table 6 referenced). - Clustering of standard errors (A.4.2) - Results robust to clusterings of standard errors at the industry, firm, and country level (see Tables 7, 8, 9). - Footnote 32: whenever standard errors are clustered at the industry level, estimates for the individual industries cannot be obtained. - Timing of purchases (A.3.2) - Carbon intensity computation multiplies number of allowances a firm purchases (or receives for free) by the average carbon price of that year, then normalizes by the firm’s revenue. - Using the average carbon price implicitly assumes firms spread purchases evenly through the year. - Rationale: main channel for allowance purchases are public allowance auctions held every week throughout the year with equal volumes (footnote 33). - Many firms purchase allowances using futures, the most traded being the December expiry contract (ESMA, 2022). - The EU ETS physical allowance transactions list confirms a high share of non-administrative transactions taking place in December, corresponding to delivery of future contracts (Figure 14 referenced). - Given futures market dominance, firm-specific purchase patterns cannot be identified using the EU ETS database. - Normalization variables (A.3.3) - Main specification uses revenue to normalize carbon cost (less biased proxy for firm size than profit). - Reproducing core results using market capitalization for scaling (Table 10) finds: - Significance and sign of coefficients next to commodities returns, Eurostoxx return and carbon price return invariant to normalization selection. - Magnitude of coefficients differs: estimated threshold of carbon intensity above which carbon price increases penalize the stock price is much higher when scaled by market capitalization (7.3% of market capitalization) versus the main specification (1.7% of revenue). - Core result confirmed: statistically significant link between carbon price returns and stock price dynamics; magnitude depends on firm carbon intensity, particularly the paid component, while free allocation is disregarded by investors. Result robust to normalization factor. - Allowance inventories (A.3.4) - Higher firm yearly carbon bill associated with larger stock-price sensitivity to carbon price increases (section 4.2). - Firms can hedge via derivatives or build up inventories of emission allowances. - EU ETS transaction log could in theory reconstruct firm allowance inventories up to April 2019 (footnote 35), but data quality is poor (Mahringer, 2021): - Transferring or receiving account of a transaction often missing. - Some free allocation transactions cannot be linked to firms, producing meaningless (negative) inventories. - Conclusion: cannot use inventory figures in regressions due to data quality issues. - Footnote 35 clarifies derivative transactions appear in the transaction log only if they lead to actual delivery of allowances; logged time corresponds to allowance delivery not derivative trade. - R&D expenditure (A.3.5) - Robustness check conditioning on firm R&D expenditures (Table 11). - Rationale: higher R&D might imply better ability to withstand transition risks; use R&D spending as proxy for innovation/readiness. - Data availability issues: - Only about 44% of companies report positive R&D expenditure. - A third have zero R&D expenditure. - Remaining firms do not report R&D data. - Sub-sample regressions: - Firms with R&D equal to zero (column 1, Table 11), firms with positive R&D (column 2), and (conditional on positive R&D) firms above or below median R&D (columns 3 and 4). - No statistically significant result beyond positive association between stock returns and Eurostoxx and gas returns. - Carbon price no longer statistically significant with stock prices of firms in these sub-samples. - Interpretation: loss of statistical significance likely due to sample changes; only 12 electricity companies and only half of companies headquartered in high carbon intensity countries report positive R&D, so the strong relationship present for highly carbon-intensive firms is less represented. *The Carrot and the Stock: In Search of Stock-Market Incentives for Decarbonization. Working Paper No. WP/2022/231* --- _Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022231-print-pdf.pdf_