On measuring climate risks using attention search and testing the clean energy-climate hypothesis

ABSTRACT

We measure physical and regulatory climate risks as innovations in several attention to climate change indexes that we construct using search volume data in Google Trends. The intention is to use the constructed risk indexes to test the empirical validity of the clean energy-climate hypothesis, which posits that clean energy prices fall (rise) following a drop (rise) in attention to climate risks. We test the empirical validity of this hypothesis at the macro (market level) using regime switching models and at the micro (firm level) using time-series regressions across clean energy sub-sectors. The macro analysis reveals that our hypothesis is only valid for climate pledges and physical climate risks. When attention to climate pledges drops or when physical climate risks are low, investors become reluctant to hold clean energy assets and vice-versa. Climate policy and climate solution risks have no impact on the nonlinear dynamic behaviour of clean energy prices. These results mean that investors and the public believe that climate policies lack credibility and climate solutions are not effective. The firm-level analysis confirms the findings of the macro analysis and reveals the heterogeneity of climate risks across clean energy sub-sectors.

KEYWORDS:

• Physical climate risk
• regulatory climate risk
• investor attention
• clean energy prices
• smooth transition regression

JEL CLASSIFICATION:

• C24
• C58
• G41
• Q54

Acknowledgements

I would like to thank the Editor, Mark Taylor, and three anonymous referees for their excellent comments. I would also like to thank Richard Dal Monte, Sean Cleary, and the participants at the Canadian Sustainable Finance Network (CSFN) webinar at Queen’s University for helpful comments. The outstanding research assistance of Mulham Akharas is highly appreciated. This paper is financially supported by a research grant (PRD-2022) from Royal Roads University. All errors remain mine.

Data statement

The data set created in this article is available for free download via the following link: https://data.mendeley.com/datasets/bz3cpxvywr/1.

Please cite the data as: Fahmy, H. (2024), ‘Climate Risk Indexes from Attention Search’, Mendeley Data, V1, doi: 10.17632/bz3cpxvywr.1

Data: https://data.mendeley.com/datasets/bz3cpxvywr/1.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/00036846.2024.2382387.

Notes

¹ It is worth noting that carbon is usually viewed in the literature as transition climate risk. In this paper, however, we follow the EPA climate indicators and classify carbon and GHG emissions physical climate risks whereas their regulatory policies, e.g. carbon credit and feed-in-tariff as transition/regulatory climate risks.

² The constructed climate risk indexes are freely available here: https://data.mendeley.com/datasets/bz3cpxvywr/1.

³ Recall that, for any variables X, Y, Z, the joint percentage change is the sum of the individual changes; that is, $log\left(\mathrm{XYZ}\right)=log\left(\mathrm{X}\right)+log\left(\mathrm{Y}\right)+log\left(\mathrm{Z}\right)$.

⁴ Source: https://www.nationalgeographic.com/science/article/how-trump-is-changing-science-environment.

⁵ Source: https://www.ishares.com.; https://www.invesco.com.

⁶ The only difference between $\mathrm{\Phi }$ and $\mathrm{\Omega }$ is that the former only includes the current and one period lag of the variables in the latter. We restrict the lags to only one in the threshold set following the results of the Schwarz information criterion.

⁷ We have also constructed innovation indexes that capture the joint impacts of oil, technology, oceans, health, and ecosystems. The results were very similar to those obtained from using the $I^{\mathrm{WTI}+\mathrm{PSE}+\mathrm{PCC}}$ index. For space limitation, we choose to only report the estimation retults of the latter index.

⁸ The transition function, original and fitted series, and standardized residuals plots for $r^{\mathit{PBD}}$ and $r^{\mathit{PBW}}$ display the same dynamic as that of $r^{\mathit{ECO}}$. We choose not to include these graphs to conserve space.

⁹ Log-linear transformation of WTI and PSE are more suitable for linear time series regressions.

Additional information

Funding

This work was supported by Royal Roads University under Grant RPD 2022.