Extreme Weather Event Impacts: Data and Models

This is the model and data description for the paper "What Firms Actually Lose (and Gain) from Extreme Weather Event Impacts".

Data

The simplest form of the project is the 13,277 firm-event impacts. This is created by analyzing over 1.7 million filings (3.5 billion paragraphs) of all publicly listed firms in the US. We analyse filing types: event-based Form 8-K, quarterly reported Form 10-Q, and annually reported Form 10-K. We upload all paragraphs that contain at least a mention of an extreme weather event / physical risk in a more fine-grained dataset.

Event-Impact Data

This is the final data, including the 13,277 firm-event impacts (name: event_impact_data). It has the following structure:

• event_id: ID of the event in the NOAA billion-dollar dataset
• cik: unique company identifier [If this combination is present, then there was a firm-event impact]
• asset, economic_flows, none: classification dimensions of the impact channel classifier; shows whether at least one document indicated an asset impact (asset==1), a pure economic flow impact, i.e. at least one document was only addressing an impact through economic flows and NOT through assets as well (economic_flows==1), or none if at least one document mentioned a none impact (neither asset, nor economic flows; none==1)
• neutral, negative, positive, reimbursement: classification dimensions of the impact dimensionality classifier; shows whether at least one document indicated a neutral, negative, or positive impact, or a reimbursement
• name: event name according to NOAA
• event_type: disaster event type according to NOAA
• begin_date: event begin date according to NOAA
• cpi_adjusted_cost: CPI-adjusted cost of the event according to NOAA

As a complementary dataset, we also upload the NOAA extreme weather event dataset (name: NOAA_event_with_summary), obtained here:

• Name: event name according to NOAA
• Disaster: disaster event type according to NOAA
• Begin Date: event begin date according to NOAA
• End Date: event end date according to NOAA
• CPI-Adjusted Cost: CPI-adjusted cost of the event according to NOAA
• Unadjusted Cost: non-adjusted cost of the event according to NOAA
• Deaths: deaths caused by the event according to NOAA
• Event Duration: event duration according to NOAA
• Event ID: ID of the event in the NOAA billion-dollar dataset
• Summary: event description / summary provided by NOAA

File-based Data

We classify over 1.7 million files and 3.5 paragraphs with our models. We analyse event-based Form 8-K, quarterly reported Form 10-Q, and annually reported Form 10-K. We upload paragraphs per file classified at least as extreme weather relevant. If a file didn't contain at least one extreme-weather-related paragraph, we leave the "paragraph" column empty (names: classified_data_8K, classified_data_10Q, classified_data_10K):

• path: filepath and unique identifier of the file
• company: unique company identifier
• year: year of the filing
• filename: filename of the 8-K, 10-K, 10-Q given by us in the downloading process
• date: date of the filing
• paragraph: text paragraph that is analyzed with the LLMs
• num_paragraphs: number of paragraphs that the entire filing contained
• num_words: number of words that the entire filing contained
• Storm, Flood, Heatwave, Drought, Wildfire, Coldwave, physical risk: Indicates 1 if the text mentions any of the above categories
• item1a, item7, item8: indicates 1 if the text was in the item mentioned; Form 8-K filings contain item7 and item8, Form 10-K filings contain item1a (not used in the analysis)
• impact: indicates “Yes” if the text mentions that the company was impacted to any extreme weather event
• impact_YY: indicates “Yes” if the text mentions that the company was impacted by the specific extreme weather event with the event_id YY in the NOAA data
• impact_directionality: indicates “Neutral”, “Negative”, “Positive”, or “Reimbursement” according to the impact directionality classification
• impact_channel: indicates “asset”, “only_economic_flows”, “none” according to the impact channel classification; “only_economic_flows” means that there was no asset impact co-detected with economic flows

Models

We upload all models, and training data in this repository. Model usage is described in the corresponding model pages.

Repository under construction

The repository is currently being constructed. If you have any questions, please reach out to tobias.schimanski@df.uzh.ch.

Paper

If you make use of the data, please cite the paper:

@article{schimanski2026extremeweatherimpacts,
  title        = {What Firms Actually Lose (and Gain) from Extreme Weather Event Impacts},
  author       = {Schimanski, Tobias and Gostlow, Glen and Toetzke, Malte and Leippold, Markus},
  url          = {https://ssrn.com/abstract=6035794},
  doi          = {10.2139/ssrn.6035794}
}