Not logged in.
Quick Search - Contribution
Contribution Details
| Type | Master's Thesis |
| Scope | Discipline-based scholarship |
| Title | Using Natural Language Processing to Estimate Climate Risk |
| Organization Unit | |
| Authors |
|
| Supervisors |
|
| Language |
|
| Institution | University of Zurich |
| Faculty | Faculty of Business, Economics and Informatics |
| Number of Pages | 124 |
| Date | 2022 |
| Abstract Text | This paper examines if exposure towards climate risk reported in regulatory disclosures impacts asset returns. Doing so, it uses machine learning algorithms to quantify 8-K, 10-K as well as Management Call scripts. Training the algorithms to distinguish between opportunities as well as regulatory and physical climate risks, the thesis finds that an increased disclosure is positively associated with asset returns in the time-series. Also, opportunities and regulatory threats show an economically and statistically significant effect on the cross-sectional variation of asset returns, while physical exposure is only marginally priced. Based on these findings, the paper aims to further challenge standard approaches in empirical asset pricing that assume a sparse model configuration and faultless variable selection. Testing the results in a high-dimensional regression setting, a model-selection method that accounts for selection mistakes is proposed. This approach is argued to consistently evaluate the marginal variation of climate risk exposure relative to a large pool of factors. Consistent with earlier results, explanatory power is monotonically reduced for each specification due to the larger absorption of the underlying asset variation based on the covariate structure. However, opportunity-based exposure is shown to maintain statistical significance beyond the explanatory power of the factor pool. |
| Export |
BibTeX
|