Measure definitions

See how topic matches become Exposure, Risk, and Sentiment values for each earnings call.

Core measures

The table and example below use sentence counts, with risk and sentiment words evaluated in the matched sentence. Each sentence contributes at most once to each applicable count, even if several terms match.

One call, three sentences

Use the query supply chain for this illustrative three-sentence call:

• “Our supply chain improved this quarter.”

  Matches the query. “Improved” adds a positive count.

• “Our supply chain improved, but the risk of a loss remains.”

  Matches the query. “Improved” adds a positive count, “loss” a negative count, and “risk” a risk count.

• “Demand declined and remains uncertain.”

  Does not match the query, so it contributes nothing to these topic measures.

Call totals

Sentiment is 2 − 1 = 1. The second sentence contributes to both sentiment counts, giving it a net contribution of zero.

Interpret the numbers

Exposure measures discussion found by your query. Risk counts risk-related language within that discussion; it is not the probability of a real-world event or a complete risk assessment.

Sentiment describes the tone of query-matched sentences, not the whole call. A net value of zero can reflect a balance of positive and negative language, rather than an absence of either.

These values are counts and count differences, not percentages, probabilities, or classifier labels. They do not establish causality, forecast outcomes, provide investment recommendations, or make compliance determinations.

For comparisons across calls, consider transcript length and any text restrictions. See normalization guidance.

How query settings change the counts

For queries without weighted dictionaries, the example uses Matched sentence and Whether any appear, the defaults. The scoring options let you change two things:

• What text is scored: The matched sentence, or a word window around each match (5–100 words total including the matched term or phrase; default 22, presets 11, 22, and 44, where 22 equals ±10 words around a two-word phrase). Word windows can cross sentence boundaries.
• What is counted: With Whether any appear, each matched sentence counts at most once toward Risk and each sentiment count. With How many appear, every qualifying dictionary-word occurrence is counted. Exposure still counts matched sentences; Sentiment remains Positive minus Negative.

Overlapping word windows can count the same word more than once. Word counts are not counts of unique words or sentence percentages.

Example: overlapping word windows

Consider risk China risk Brazil risk with the query China OR Brazil. If each match's window covers all five words, How many appear gives Risk 6: three risk words for each of two matches. Whether any appear gives Risk 1, since this is one matched sentence.

Section and speaker filters restrict eligible sentences; see denominator choices. For older datasets, check the saved matching and scoring settings rather than inferring them from an adjacent-sentence label. Keep the engine and settings with your export.

Why sentence filters can show fewer matches

The Snippet Tool's sentiment filters use stricter rules than the counts above: positive shows positive-only sentences, negative shows negative-only sentences, and sentiment shows either one-sided case. Mixed positive-and-negative sentences are excluded.

The second sentence in the example therefore contributes to both exported sentiment counts but would be hidden by these filters.

Dictionary sources

Positive and negative words come from the Loughran-McDonald financial sentiment dictionary. The negative list excludes question and questions.

Risk words come from the deduplicated union of Oxford Thesaurus synonyms for risk, risky, uncertain, and uncertainty, excluding question, questions, and venture.

Overall risk and sentiment

Overall, or unconditional, counts cover language across the call without requiring a topic-query match. They are separate from the core topic measures.

Use them when comparing topic-linked language with overall risk or sentiment discussion. See Overall risk and sentiment for the download and exact column names.

Weighted measures

A weighted dictionary query combines term weights with matching and scoring rules. Weighted values can be fractional and are not plain sentence counts.

The dictionaries are fixed; their terms cannot be edited.

The dictionary provides raw and normalized weights. The scoring settings and dataset methodology determine how they contribute to the output; do not assume every exported value uses normalized weights.

The dictionaries are based on the research behind:

• Firm-Level Political Risk: Measurement and Effects, by Hassan, Hollander, van Lent, and Tahoun.
• Sources and Transmission of Country Risk, by Hassan, Schreger, Schwedeler, and Tahoun.

Cite the relevant source paper. Keep Weighted query terms, which contains the selected dictionary and its raw and normalized weights, with your export.

Weighted values are not comparable to sentence counts or across different dictionaries. Weighted datasets cannot be combined or compared with other datasets.

A weighted dictionary search does not automatically reproduce a dataset from Curated Measures. Curated datasets can use different scoring windows, weighting, and aggregation; review the entry's methodology before treating the results as equivalent.