Ping An Digital Economic Research Center (DERC)

Ping An - DERC

Ping An Digital Economic Research Center (DERC) utilizes more than 50 TB of high frequency data points, more than 30 years of historical data and more than 1.5 billion data points to drive research on the “AI + Macro Forecast” and provide insights and methods towards precise macroeconomic trends. 

Ping An DERC is a research center for Ping An Insurance (Group) Company of China, Ltd. (“Ping An”) is a world-leading technology-powered retail financial services group. With over 204 million retail customers and 534 million internet users, Ping An is one of the largest financial services companies in the world.

Ping An Digital Economic Research Center

Zheng Zeng

Shanghai Shenzhen CSI 300 Index (SHSZ300 INDEX EQUITY)

<blockquote>
ESG Report Series Issue 3
- AI Driven Indicators, Green Washing Detection, and Firm Characteristics
</blockquote>In our previous report, “Where we stand with climate disclosures and why we need them - Natural Language Processing applications in analysis of sustainability reporting,” (Ping An Digital Economic Research Center, 2020), we applied Natural Language Processing (NLP) to climate risk disclosures by US and Chinese firms to gauge the coverage of climate related and financial impact metrics according to Task Force for Climate-related Financial Disclosures (TCFD) recommendations. The results demonstrated that emissions and energy metrics are in general widely covered, whereas water and land use disclosures are limited and sparse. Similarly, disclosure of financial impact metrics shows considerable heterogeneity across sectors, with some important impact dimensions largely under-reported. In this report, we build on these insights and develop NLP based indicators that allow us to better capture the breadth and depth of companies’ climate risk disclosures. The objective is three-fold: first, we leverage AI tools to better assess a firm's climate risk exposure and its compliance with TCFD reporting guidelines; second, we use AI based indicators to detect potential evidence of greenwashing; third, we explore the relationship between climate risk disclosure and firm characteristics, including financial performance. The analysis presented here focuses on US and Chinese firms (constituents of S&amp;P500 and CSI300) that have disclosed climate risks, although parallel work on European firms shows consistent results. A summary of our findings is as follows:AI driven indicators: AI driven transparency indicators are a valuable complement to ESG analytics tools and ratings currently available in the market. <ul style="list-style-type: square;">
<li>We use NLP to develop climate risk disclosure indicators informing us on the scope and depth of disclosures. Of particular importance are climate risk metrics and financial impact disclosure indicators. We also create composite transparency indicators to carry out automated assessment of TCFD reporting compliance.</li>
<li>As opposed to ratings relying on linear aggregation of particular attributes, which currently dominate the industry<a href="#_ftn1" target="_blank" rel="noopener">[1]</a>, our indicators use a mixture of NLP and probabilistic tools to identify and aggregate relevant attributes.</li>
<li>The indicators developed perform better than existing ESG ratings in the market in differentiating between Green and Brown firms. They also offer insights into the relationship between firm characteristics and climate disclosures.</li>
</ul>Green washing detection: we find patterns of disclosure suggesting possible green washing by high emission firms; also, unsophisticated rating tools may implicitly reward the opaqueness of disclosures.<ul style="list-style-type: square;">
<li>Possible evidence of green washing for high emission firms: <ul style="list-style-type: circle;">
<li>Compliance around disclosure of climate related metrics is higher on average for lower emission firms, thus suggesting underreporting by high emission firms.
</li>
<li>High emission firms score better in disclosing financial impact metrics, but with an important exception: information on capital and financing impact is systematically underreported. As the issue of stranded assets and liabilities<a href="#_ftn2" target="_blank" rel="noopener">[2]</a> is of major importance to high emission sectors, such as oil and coal companies, the results indicate that a number of firms selectively leave this impact dimension undisclosed.</li>
<li>
Disclosure of scope 3 emissions is considerably worse for high emission firms, despite the increasing focus of regulators and investors on this dimension.</li>
</ul>
</li>
<li>
Opaqueness may implicitly be rewarded instead of penalized by unsophisticated rating tools, as some companies engaging in climate risk disclosure appear to be penalized by some ESG rating tools. This gives perverse incentives for selective non-disclosure, leading to greenwashing.</li>
<li>Similar to existing literature, we find dispersion between ESG ratings<a href="#_ftn3" target="_blank" rel="noopener">[3]</a> and hence they should be used with caution. Some ESG ratings appear to be positively associated with leverage, after controlling for emissions and other firm characteristics, whereas our AI driven indicators suggest systematically lower leverage for companies engaging in climate risk disclosure.</li>
</ul>Valuations: companies with greater disclosure of metrics on climate risk impact on financial performance have higher valuations and lower leverage, after controlling for carbon emissions and other firm characteristics.<ul style="list-style-type: square;">
<li>Firms with better disclosure of financial impact metrics tend to have higher valuations, lower leverage, and lower cost of capital.</li>
<li>Large cap firms that are TCFD compliant tend to have higher valuations, whereas small and medium cap firms engaging in climate disclosures may still offer considerable opportunities for appreciation.</li>
<li>The results are consistent across a number of specifications and apply after controlling for traditional ESG ratings, which instead appear to have weak or inconclusive bearing.</li>
<li>The results also apply after controlling for carbon emissions, which recent literature has found to be associated with a carbon risk premium and hence lower valuations.<a href="#_ftn4" target="_blank" rel="noopener">[4]</a> Our indicators therefore offer investors and corporates a more granular understanding of valuations beyond carbon emissions.</li>
</ul>Firm value: AI driven indicators suggest that greater disclosure is associated with lower cost of capital, thus boosting firm value. <ul style="list-style-type: square;">
<li>Despite the increasing popularity of ESG criteria,<a href="#_ftn5" target="_blank" rel="noopener">[5]</a> demonstrating whether and how ESG compliance adds value to the firm is notoriously difficult and so far the empirical evidence is mixed or inconclusive.<a href="#_ftn6" target="_blank" rel="noopener">[6]</a>
</li>
<li>We find evidence that companies that are more transparent in disclosing financial impact metrics have lower cost of capital,<a href="#_ftn7" target="_blank" rel="noopener">[7]</a> especially among firms that disclose impact of climate risks on expenditure and assets &amp; liabilities.</li>
<li>Our results on valuations offer a consistent picture in terms of cost of equity, whereas our parallel findings on leverage support consistent implications for the cost of debt. Both these channels represent important components of any corporate valuation framework and hence provide support for climate risk disclosures contributing to firm value. Further research is needed to understand other dimensions, such as revenues and operating margins.</li>
</ul>Summary of AI driven indicators and greenwashing detection <table class="table table-striped"><tbody>
<tr>
<td style="width: 99pt;"> </td>
<td style="width: 6cm;">AI driven indicators</td>
<td style="width: 191.35pt;">Greenwashing detection</td>
</tr>
<tr>
<td style="width: 99pt;">What is it?</td>
<td style="width: 6cm;">AI-based indicators that assess coverage rate of various climate and financial impact related indicators, as well as themes discussed in companies’ climate disclosure reports.</td>
<td style="width: 191.35pt;">Detect anomalies between companies’ disclosures of climate risks and actual exposure.</td>
</tr>
<tr>
<td style="width: 99pt;">How is it done?</td>
<td style="width: 6cm;">
<ul><li>Rule-based matching: For each indicator, construct relevant anchor words or expressions to locate relevant text in disclosure reports and then estimate probability of coverage by assessing sample precision and number of matches found.</li></ul>
<ul><li>Unsupervised learning: extract distinct themes from companies’ disclosure reports and then calculate companies’ relative coverage rate of each theme.</li></ul>
</td>
<td style="width: 191.35pt;"><ul>
<li>Detect anomalies in ESG ratings of companies who engage in climate disclosures vs. those who do not.</li>
<li>Benchmark high vs. low emission firms in terms of disclosure of climate metrics (e.g. emissions, energy use, water use, land use) and financial impact metrics to detect any under-reporting by high emission firms.</li>
<li>When evaluating relationship between disclosures and company performance, detect where traditional ESG ratings and AI driven indicators show inconsistent results.</li>
</ul></td>
</tr>
<tr>
<td style="width: 99pt;">What are the implications?</td>
<td style="width: 6cm;">
Firms with better disclosure of AI driven financial impact indicators have higher valuations, lower leverage, and lower cost of capital. Traditional ESG ratings present either diverging or statistically weaker results.
 
</td>
<td style="width: 191.35pt;">Allows for Automatic detection of under-reporting by high-emission firms, and inconsistency between ESG ratings, which allows investors to use ESG ratings with caution when integrating them in investment analysis.</td>
</tr>
</tbody></table><div>
 <hr align="left" size="1" width="33%">
<div id="ftn1" style="text-align: left;"><a href="#_ftnref1" target="_blank" rel="noopener">[1]</a> See, for example, Berg et al. (2019).</div>
<div id="ftn2" style="text-align: left;"><a href="#_ftnref2" target="_blank" rel="noopener">[2]</a> Stranded assets are assets that have become uneconomic (firms are unable to recover their investment cost) or turned into liabilities, due to transition risks such as taxation of fossil fuel supply or consumption (e.g., Caldecott et al. 2015). Stranded liabilities refer to the cost of retiring uneconomical long-lived assets, such as oil and gas infrastructure.</div>
<div id="ftn3" style="text-align: left;"><a href="#_ftnref3" target="_blank" rel="noopener">[3]</a> See Berg et al. (2019), for example.</div>
<div id="ftn4" style="text-align: left;"><a href="#_ftnref4" target="_blank" rel="noopener">[4]</a> See Bolton and Kacperczyc (2020).</div>
<div id="ftn5" style="text-align: left;"><a href="#_ftnref5" target="_blank" rel="noopener">[5]</a> See Amel-Zadeh and Serafeim (2020).</div>
<div id="ftn6" style="text-align: left;"><a href="#_ftnref6" target="_blank" rel="noopener">[6]</a> See Cornell and Damodaran (2020) for an overview.</div>
<div id="ftn7">
<a href="#_ftnref7" target="_blank" rel="noopener">[7]</a> See Goss and Roberts (2011) for similar evidence on bank loans and corporate social responsibility. Using US data, they find that socially responsible firms pay between 7 and 18 basis points less than other firms, unsecured loans being the most sensitive to social responsibility concerns.
 
</div>
</div>


ESG Report Series Issue 3
- AI Driven Indicators, Green Washing Detection, and Firm Characteristics
In our previous report, “Where we stand with climate disclosures and why we need them - Natural Language Processing applications in analysis of sustainability reporting,” (Ping An Digital Economic Research Center, 2020), we applied Natural Language Processing (NLP) to climate risk disclosures by US and Chinese firms to gauge the coverage of climate related and financial impact metrics according to Task Force for Climate-related Financial Disclosures (TCFD) recommendations. The results demonstrated that emissions and energy metrics are in general widely covered, whereas water and land use disclosures are limited and sparse. Similarly, disclosure of financial impact metrics shows considerable heterogeneity across sectors, with some important impact dimensions largely under-reported. In this report, we build on these insights and develop NLP based indicators that allow us to better capture the breadth and depth of companies’ climate risk disclosures. The objective is three-fold: first, we leverage AI tools to better assess a firm's climate risk exposure and its compliance with TCFD reporting guidelines; second, we use AI based indicators to detect potential evidence of greenwashing; third, we explore the relationship between climate risk disclosure and firm characteristics, including financial performance. The analysis presented here focuses on US and Chinese firms (constituents of S&amp;P500 and CSI300) that have disclosed climate risks, although parallel work on European firms shows consistent results. A summary of our findings is as follows:AI driven indicators: AI driven transparency indicators are a valuable complement to ESG analytics tools and ratings currently available in the market. 
We use NLP to develop climate risk disclosure indicators informing us on the scope and depth of disclosures. Of particular importance are climate risk metrics and financial impact disclosure indicators. We also create composite transparency indicators to carry out automated assessment of TCFD reporting compliance.
As opposed to ratings relying on linear aggregation of particular attributes, which currently dominate the industry[1], our indicators use a mixture of NLP and probabilistic tools to identify and aggregate relevant attributes.
The indicators developed perform better than existing ESG ratings in the market in differentiating between Green and Brown firms. They also offer insights into the relationship between firm characteristics and climate disclosures.
Green washing detection: we find patterns of disclosure suggesting possible green washing by high emission firms; also, unsophisticated rating tools may implicitly reward the opaqueness of disclosures.
Possible evidence of green washing for high emission firms: 
Compliance around disclosure of climate related metrics is higher on average for lower emission firms, thus suggesting underreporting by high emission firms.

High emission firms score better in disclosing financial impact metrics, but with an important exception: information on capital and financing impact is systematically underreported. As the issue of stranded assets and liabilities[2] is of major importance to high emission sectors, such as oil and coal companies, the results indicate that a number of firms selectively leave this impact dimension undisclosed.

Disclosure of scope 3 emissions is considerably worse for high emission firms, despite the increasing focus of regulators and investors on this dimension.



Opaqueness may implicitly be rewarded instead of penalized by unsophisticated rating tools, as some companies engaging in climate risk disclosure appear to be penalized by some ESG rating tools. This gives perverse incentives for selective non-disclosure, leading to greenwashing.
Similar to existing literature, we find dispersion between ESG ratings[3] and hence they should be used with caution. Some ESG ratings appear to be positively associated with leverage, after controlling for emissions and other firm characteristics, whereas our AI driven indicators suggest systematically lower leverage for companies engaging in climate risk disclosure.
Valuations: companies with greater disclosure of metrics on climate risk impact on financial performance have higher valuations and lower leverage, after controlling for carbon emissions and other firm characteristics.
Firms with better disclosure of financial impact metrics tend to have higher valuations, lower leverage, and lower cost of capital.
Large cap firms that are TCFD compliant tend to have higher valuations, whereas small and medium cap firms engaging in climate disclosures may still offer considerable opportunities for appreciation.
The results are consistent across a number of specifications and apply after controlling for traditional ESG ratings, which instead appear to have weak or inconclusive bearing.
The results also apply after controlling for carbon emissions, which recent literature has found to be associated with a carbon risk premium and hence lower valuations.[4] Our indicators therefore offer investors and corporates a more granular understanding of valuations beyond carbon emissions.
Firm value: AI driven indicators suggest that greater disclosure is associated with lower cost of capital, thus boosting firm value. 
Despite the increasing popularity of ESG criteria,[5] demonstrating whether and how ESG compliance adds value to the firm is notoriously difficult and so far the empirical evidence is mixed or inconclusive.[6]

We find evidence that companies that are more transparent in disclosing financial impact metrics have lower cost of capital,[7] especially among firms that disclose impact of climate risks on expenditure and assets &amp; liabilities.
Our results on valuations offer a consistent picture in terms of cost of equity, whereas our parallel findings on leverage support consistent implications for the cost of debt. Both these channels represent important components of any corporate valuation framework and hence provide support for climate risk disclosures contributing to firm value. Further research is needed to understand other dimensions, such as revenues and operating margins.
Summary of AI driven indicators and greenwashing detection 

 
AI driven indicators
Greenwashing detection


What is it?
AI-based indicators that assess coverage rate of various climate and financial impact related indicators, as well as themes discussed in companies’ climate disclosure reports.
Detect anomalies between companies’ disclosures of climate risks and actual exposure.


How is it done?

Rule-based matching: For each indicator, construct relevant anchor words or expressions to locate relevant text in disclosure reports and then estimate probability of coverage by assessing sample precision and number of matches found.
Unsupervised learning: extract distinct themes from companies’ disclosure reports and then calculate companies’ relative coverage rate of each theme.


Detect anomalies in ESG ratings of companies who engage in climate disclosures vs. those who do not.
Benchmark high vs. low emission firms in terms of disclosure of climate metrics (e.g. emissions, energy use, water use, land use) and financial impact metrics to detect any under-reporting by high emission firms.
When evaluating relationship between disclosures and company performance, detect where traditional ESG ratings and AI driven indicators show inconsistent results.



What are the implications?

Firms with better disclosure of AI driven financial impact indicators have higher valuations, lower leverage, and lower cost of capital. Traditional ESG ratings present either diverging or statistically weaker results.
 

Allows for Automatic detection of under-reporting by high-emission firms, and inconsistency between ESG ratings, which allows investors to use ESG ratings with caution when integrating them in investment analysis.



[1] See, for example, Berg et al. (2019).
[2] Stranded assets are assets that have become uneconomic (firms are unable to recover their investment cost) or turned into liabilities, due to transition risks such as taxation of fossil fuel supply or consumption (e.g., Caldecott et al. 2015). Stranded liabilities refer to the cost of retiring uneconomical long-lived assets, such as oil and gas infrastructure.
[3] See Berg et al. (2019), for example.
[4] See Bolton and Kacperczyc (2020).
[5] See Amel-Zadeh and Serafeim (2020).
[6] See Cornell and Damodaran (2020) for an overview.

[7] See Goss and Roberts (2011) for similar evidence on bank loans and corporate social responsibility. Using US data, they find that socially responsible firms pay between 7 and 18 basis points less than other firms, unsecured loans being the most sensitive to social responsibility concerns.
 



Climate Disclosures and Financial Performance