Climate Risk Indexes

Download Data                    Download Annotated Index

We are pleased to host the climate Physical Risk Index (PRI) and Transition Risk Index (TRI) and the global Climate Risk Vocabularies developed by Lavinia Rognone, Giovanna Bua, Daniel Kapp, and Federico Ramella in "Transition versus physical climate risk pricing in European financial markets: a text-based approach".

Using text-based approaches and authoritative sources, the global physical and transition climate risk vocabularies provide full relevance-ranked phraseologies associated with the two types of risks. The PRI and TRI reflect daily innovation to physical and transition risk, respectively, considering Reuters News articles from January 2005 to the present (updated semi-annually).

Starting from authoritative and scientific texts on climate change, the authors screen and aggregate the content into two documents: a Physical Risk Document (PRD) and a Transition Risk Document (TRD). These documents encompass all the information about climate risks which the authors use to feed the text-based algorithms. PRD and TRD are converted into numerical vectors using the term frequency-inverse document frequency (tf-idf) method. During pre-processing, stop-words are removed and terms are stemmed. The tf for each document's unigrams and bigrams is calculated as the number of times each term appears in the document divided by the total number of terms. The idf score instead reflects how common (rare) a term is in a language (in our case English). Finally, the tf-idf for each term is obtained as tf x idf. As constructed, the tf-idf scores are relevance scores, with high tf-idf scores terms being more representative of the document's topic. These vocabularies are ranked by relevance, globally representative of climate change information, and sourced from authoritative texts rather than arbitrarily chosen by the authors.

The authors then gather a large number of news articles from Reuters News using the Factiva database, starting from January 2005. They create daily news documents and compute a tf-idf vector representation for each of them, similar to the method used for the vocabularies. As a next step, the authors calculate the cosine-similarity between each daily news document and the PRD (TRD). The cosine-similarity is a text-analysis method that evaluates how similar pairs of text are. It works with vectors, as it computes the angular distance between pairs of vectors. The rule is: The more similar the content of two texts is, the closer their tf-idf vectors point, the smaller their vector angular distance is, the higher the cosine-similarity is. The cosine-similarity method generates the "concern" time series which reflect the portion of daily news discussing physical or transition risk. Finally, the PRI and TRI are the residuals of AR1 processes of the concern series, thus representing innovation/shocks to climate change risks.

PRI and TRI spikes on days in which there is an unexpected discussion on climate physical and transition risk, respectively, encompassing a broad range of relevant topics including acute and chronic physical hazards, adaptation and mitigation policies, and net-0 targets.