Climate change has started to significantly affect many firms, but properly quantifying firm-level exposure to climate change risks and opportunities is a challenge for investors, regulators and policy makers. Complications arise because:
- the effects of climate change on firms are highly uncertain;
- the effects of climate change are likely to be heterogeneous across firms, even in the same industry; and
- there is no common understanding of how to reliably quantify firm-level climate change exposure.
While firms’ voluntarily disclosed carbon emissions are gaining some traction as an exposure measure, this data is limited. Furthermore, disclosed emissions reflect firms’ historic, rather than future, business models, and do not distinguish between good and bad emissions.
These challenges are severe and can impede the reallocation of resources from brown to green firms. Furthermore, the lack of a firm-level exposure measure may contribute to the potential mispricing of climate risks and opportunities, and it complicates the development of financial instruments that allow market participants to hedge the effects of climate change.
In our paper, we use earnings call transcripts to construct time-varying measures of firm-level exposure to climate change. To construct our measures, we identify word combinations – bigrams – that signal climate change conversation in conference calls. Our method adapts the machine learning keyword discovery algorithm proposed by King et al. (2017) to produce four related sets of climate change bigrams: the first captures broadly-defined climate change aspects, while the remaining three cover shocks related to specific topics – opportunity, physical (e.g. sea level rises, natural disasters), and regulatory shocks (e.g. carbon taxes, cap-and-trade markets).
These measures count the frequency with which certain climate change bigrams occur in relation to the total number of bigrams in the transcript. We interpret these as representing the occurrence of climate change events or shocks at a firm. We also construct measures of the first and second moment associated with these shocks: whether the events represent good or bad news (based on expectation) and whether they are uncertain.
For the first moment, we construct sentiment measures which count the relative frequency of climate change bigrams that occur near words that are positive or negative in tone. For the second moment – risk measures – we count the relative frequency of climate change bigrams mentioned in the same sentence as the words risk or uncertainty (or their synonyms). Our sample contains more than 80,000 annual observations originating from more than 10,000 unique firms in 34 countries between 2002 to 2019.
Top bigrams associated with exposure to climate change opportunities refer to new (green) technologies, such as electric vehicles. Top regulatory bigrams are related to regulatory and/or governmental interventions associated with climate change and the goal to reduce carbon emissions. Top bigrams linked to physical shock exposures are related to hurricanes, desalination, or droughts. We validate our approach by examining individual text fragments taken from the point in the transcripts identified by our algorithm as the moment when participants discuss climate change, and verify that the call fragments are indeed centred on salient climate issues.
The data for the broadly-defined exposure measure (Figure 1) reveals that discussions of climate change issues increase remarkably over time until around 2011, starting in the mid-2000s. There is a modest decline up to the largely unsuccessful 2012 Doha Climate Summit, with a levelling off at a high level (compared to the years before 2011) in the subsequent years. We observe a renewed increase in climate change exposure since the Paris Agreement in 2015 and the 2016 election of US President Donald Trump. Climate change exposure reaches its highest overall level at the end of the sample in 2019.
Figure 1: Time-series evolution of climate change exposure
Utilities and firms in construction and coal mining have the highest overall exposure to climate change. Utilities top the exposure ranking for opportunity and regulatory shocks, which signifies that they face opportunities (e.g. renewable energy) and regulatory risks (e.g. carbon taxes) related to climate change. Physical climate change exposure is highest for firms operating in the areas of paper and allied products, heavy construction, and insurance.
For all the measures, we find large intra-industry variation, indicating that firms within the same industry will benefit or suffer in various degrees from climate change. Furthermore, exposure to climate change varies substantially across countries (Figure 2), and we document links between our exposure measures and proxies for the regulatory and physical impacts of climate change.
Figure 2: Cross-country distribution of climate change exposure
Most variations in our exposure measures play out at the firm level (rather than country or industry level, or over time). Only half of this firm-level variation is constant, suggesting that over time, within an industry, different firms are exposed to climate change.
We then compare the results of this analysis with a similar exercise for a firm’s carbon intensity (emissions scaled by assets) as well as its ISS Carbon Risk Rating. The firm-level variation for carbon intensities and the ISS measures are substantially smaller, especially compared to our topics-based measures. Two-thirds of the variation in the ISS ratings is constant. Carbon intensities, which are increasingly used in finance literature, are driven mostly by industry effects.
Our climate change exposure measures, on the one hand, and the carbon intensity and ISS measures, on the other hand, overlap to some extent – as expected given that they all aim to capture dimensions of firms’ climate change exposure. Carbon intensities appear to correlate mostly with our measures of opportunity and regulatory shocks. The ISS rating reflects our measures of opportunities more than those of regulatory or physical events. Overall, our analysis suggests that these alternatives are more specialised than our (more comprehensive) measures.
The role of economic factors
We also explore the role of important economic factors that prior work has identified as potentially being related to firm-level climate change exposure.
Higher climate change attention in the media is associated with a rise in firms’ exposures to regulatory and physical climate shocks, but not with opportunity shocks. A reason for this could be that the media pays more attention to environmental rules and physical threats to economic activity than to the opportunities that climate change might offer to businesses. Conference call participants that follow the media may therefore be more likely to address such topics.
Firm-level institutional ownership is negatively related to climate change exposure. This effect is particularly strong in recent years and originates primarily from a negative association between institutional ownership and exposure to regulatory and opportunity shocks. This is consistent with institutional investors starting to underweight (or divest) from firms with high climate change exposure, apparently without distinguishing much between firms with upside and downside exposures.
Firm exposure to regulatory shocks is negatively associated with valuation changes. We can document such an effect only for the second half of the sample – the years during which climate change exposure is relatively high (since 2011). At the same time, we cannot detect that changes in firm valuations reflect firm-level exposures to opportunity shocks.
Our paper is available here.
This blog is written by academic guest contributors. Our goal is to contribute to the broader debate around topical issues and to help showcase research in support of our signatories and the wider community.
Please note that although you can expect to find some posts here that broadly accord with the PRI’s official views, the blog authors write in their individual capacity and there is no “house view”. Nor do the views and opinions expressed on this blog constitute financial or other professional advice.
If you have any questions, please contact us at email@example.com.
Between 70.4% and 96.8%
King, G., P. Lam, and M. E. Roberts (2017). Computer-assisted keyword and document set discovery from unstructured text. American Journal of Political Science 61 (4), 971-988.