Curated Measures

Country Risk

Firm-level country risk similar to Hassan, Schreger, Schwedeler & Tahoun (REStud, 2023).

Updated Jul 23, 2026 45 series 1,010 viewers 2 downloaders

Sign in to download curated-measure data

Preview

You can browse series and charts without an account. Downloads become available after you sign in, based on your plan.

  • Plans at a glance: Free: selected Curated Measures included. Standard: time-series for every series, firm panels at 1 credit per series. Research: time-series and firm panels for every series.
Create free account Sign in Compare plans

Overview

This page is the home of the maintained, quarterly-updated continuation of the firm-level country risk measure introduced in Hassan, Schreger, Schwedeler, and Tahoun (2023, Review of Economic Studies). The scoring follows closely what is done in the paper; the data extends it from 2003 to the present and is refreshed quarterly. NL Analytics is the authors' vehicle for ongoing updates, and the original country-risk.net release points here for current data.

The measure assigns every firm, in every quarter, a CountryRisk score for each of 45 countries: how much risk the firm's executives and the analysts on its earnings calls associate with that country at that point in time. Because the panel is firm by country by quarter, it supports the cross-border questions the paper is built around: how perceptions of one country's risk surface in firms headquartered elsewhere, and how local and foreign perceptions of the same country diverge.

If you already know the paper, this page is written to answer the two questions you actually have:

  1. Is this the same series?
  2. How do I get the updated version for my own work?

If you arrived from country-risk.net

The public archive ends in 2020-Q4. Everything from 2002-Q1 through 2026-Q2, including the full period covered by the archive, is available here as a single, internally consistent panel. Analysis can therefore use one continuous series rather than splice a free historical file onto a paid update. The historical period is recomputed under the same methodology as the update, avoiding a break in construction where the public data ends.

How closely it matches the published series

Computed against the CountryRisk series in the published firm-country-quarter panel distributed at country-risk.net, on the matched overlap: N = 321,048 matched firm-quarters per country, spanning all 76 quarters from 2002-Q1 through 2020-Q4.

CountryRisk, Pearson correlation Average Lowest Highest
Firm-quarter (within country) 0.933 0.748 (Israel) 0.974 (Hungary)
Aggregate (quarterly) 0.983 0.932 (Israel) 0.995 (Greece)

At the firm level, 43 of the 45 countries exceed a correlation of 0.85. The two exceptions—Israel (0.748) and Taiwan (0.786)—still track the published series closely after quarterly aggregation, with correlations of 0.932 and 0.965, respectively. The firm-level results are most relevant for cross-sectional and panel applications, while the aggregate results show that the reproduced and published series follow very similar patterns through time. We document the remaining differences below; they primarily reflect differences in call-level sample composition and tokenization that become much smaller after aggregation.

The comparison follows the published panel's own conventions: call-level scores are scaled by transcript length, collapsed to quarterly means by firm, and gaps of up to three quarters are carried forward as in the original import pipeline. A no-fill variant is included in the repository outputs as a robustness check. Correlations were computed on the v1.1 vintage of the firm-level files listed on this page.

Full per-country results:

Country Firm-quarter Pearson Quarter Pearson
AR 0.972 0.992
AU 0.944 0.986
BE 0.946 0.989
BR 0.954 0.993
CA 0.898 0.967
CH 0.917 0.981
CL 0.929 0.984
CN 0.917 0.990
CO 0.933 0.989
CZ 0.948 0.984
DE 0.948 0.991
EG 0.955 0.992
ES 0.955 0.988
FR 0.951 0.988
GB 0.944 0.993
GR 0.968 0.995
HK 0.965 0.995
HU 0.974 0.987
ID 0.943 0.987
IE 0.940 0.989
IL 0.748 0.932
IN 0.934 0.979
IR 0.957 0.986
IT 0.966 0.990
JP 0.924 0.991
KR 0.917 0.984
MX 0.867 0.936
MY 0.971 0.988
NG 0.933 0.982
NL 0.929 0.989
NO 0.967 0.960
NZ 0.941 0.982
PH 0.931 0.966
PK 0.967 0.983
PL 0.943 0.987
RU 0.952 0.995
SA 0.938 0.989
SE 0.963 0.986
SG 0.963 0.987
TH 0.908 0.987
TR 0.963 0.991
TW 0.786 0.965
US 0.857 0.973
VE 0.968 0.988
ZA 0.883 0.973
Average 0.933 0.983

> Every number in this table is reproducible: the MIT-licensed script in our public GitHub repository, regenerates the full set from two inputs, the firm-level files from this page and the published firm-country-quarter panel, which is freely downloadable from country-risk.net. Each run writes a manifest recording the row counts and SHA-256 hashes of every input file, so you can verify you are comparing the same data we did.

How to get it

Each country series' firm-level panel can be unlocked individually with an export credit on the Free and Standard ($50/mo) plans, or comes included in full on Research ($2,000/quarter, student pricing on request), which includes panel exports across the whole Curated Measures library and your own custom searches. Time-series for this measure is included on Standard and Research.

Non-commercial academic use of the historical archive remains free on country-risk.net. This page is for the updated, maintained, citable vintage.

Explore the data

CountryRisk: United States (v1.1) over time

Date range: January 1, 2002 to June 30, 2026.

Time-series chart for CountryRisk: United States (v1.1) over time.
Loading figure...

Explore series

Series

45 total Data through Jun 30, 2026

Browse the related series and open any series for its full query, filters, and methodology notes.

No series match your filter.

Methodology and data

What is the same, and what is different

The measure is computed as in the paper. Verified directly against the original implementation, the following are identical:

  • The scoring rule for CountryRisk: every occurrence of a country term is a center, risk words are counted in the ±10-word window around it, and each occurrence contributes the country term's tf-idf weight multiplied by that risk count.
  • The counting window: ±10 words around each country term, with the country term's own tokens included in the count (center-inclusive), as in the paper.
  • The per-country tf-idf keyword libraries and their weights are identical to the ones that generated the published series, as is the risk and uncertainty synonym set.
  • Window geometry for one- and two-word country names, case-sensitive matching of country terms, and per-occurrence accumulation.
  • Special-case masking: mentions inside "Gulf of Mexico" do not count toward Mexico, as in the original.

The differences that remain are peripheral. They affect which text enters the count and how scores are scaled, not how the measure is defined:

  • Tokenization. Word boundaries are drawn slightly differently (punctuation-split, digits retained). Multi-word country names are matched as phrases, and a small number of de-accented institutional terms and rare three-or-more-word entries do not match under the current tokenizer. These are negligible in the earnings-call corpus, and the handling is validated to track the published series.
  • Sample composition. The mapping to gvkey and headquarters country differs from the original release. This is likely the main source of the remaining firm-level discrepancy, and it averages out under aggregation, which is why the aggregate series correlate far more tightly than the firm-level ones.
  • Minor filter and edge-case rules. No minimum-length filter on very short transcripts, and one boilerplate safe-harbor rule is handled slightly differently. The safe-harbor keyword set used to filter out safe-harbor-related snippets matches.
  • Scale. Firm-level files ship raw weighted counts alongside each call's word count; the paper's normalization and quarterly aggregation are applied downstream and documented under Data. Scale choices do not affect correlations or country rankings.

Data

Every row carries identifiers you may already use: gvkey, cusip, isin, ticker, company name, and headquarters country, plus the country the score refers to. Each row is one earnings call in one country series, carrying both the calendar date and start time, so you can aggregate to firm-country-quarter on whichever convention your panel uses and handle multiple calls per quarter explicitly.

Scores are shipped raw: the weighted risk count for each call, together with the call's word count. To reproduce the published panel's convention, multiply by 100,000, divide by the word count, average calls to firm-quarter, and carry gaps of up to four quarters forward as in the original import pipeline. The correlation repository implements exactly these steps, so you inherit our aggregation only if you choose to. Exports are available as CSV.

References

Hassan, T. & Schreger, J. & Schwedeler, M. & Tahoun, A., "Sources and Transmission of Country Risk," The Review of Economic Studies, 91(4):2307-2346 (2023)

NL Analytics. (2025). Country Risk [Data set]. NL Analytics. https://apps.nlanalytics.tech/curated-measures/country-risk/

Back to Curated Measures