Tracking Poland’s Economic Policy Uncertainty Through Newspaper Coverage

Turning newspaper coverage into an indicator

The researchers draw on nearly two million articles from six newspapers: Rzeczpospolita, Parkiet, Gazeta Wyborcza, Dziennik Gazeta Prawna, Gazeta Polska Codziennie and Puls Biznesu. Their monthly series covers January 2000 through June 2026.

Following the newspaper-based framework developed by Baker, Bloom and Davis, the automated filter flags articles containing at least one term from each of three categories: economy, policy and uncertainty. The dictionary adapts these categories to Polish vocabulary and institutions, including European Union bodies.

For each newspaper and month, the researchers divide matching articles by total articles. This prevents a newspaper from contributing more simply because it publishes more material. They then scale each newspaper’s series by its standard deviation during available months of the fixed 2000–2024 baseline, combine the series with equal weights, and normalize the resulting index to a baseline average of 100. That value is a reference point, not a percentage or a threshold separating certainty from uncertainty.

Keeping a changing newspaper panel comparable

Only Rzeczpospolita and Parkiet cover the earliest years. Other titles enter later, with the six-paper panel complete from 2014. Adding a newspaper with a different reporting pattern could otherwise create an artificial jump.

The authors address this by working backward from the full panel and adjusting earlier, smaller panels using periods of overlapping coverage. Importantly, the published series modifies the original Baker–Bloom–Davis splicing procedure: it matches levels through multiplication rather than an additive adjustment.

The researchers report that the additive approach produced three negative monthly readings in the final dictionary specification. Multiplicative splicing eliminated those values and better preserved the broader 2008–2009 financial-crisis increase seen in the original two-paper series. This improves historical comparability, although early observations still depend on fewer newspapers and adjustments estimated from later overlaps.

What human readers confirmed

Three trained annotators reviewed 600 articles without seeing the automated classifications. The sample included 480 flagged articles and 120 near-misses containing economy and uncertainty terms but no policy match.

Reviewers confirmed economic policy uncertainty in 337 of the 480 flagged articles: precision of 70.2%, with a reported 95% confidence interval of 66.0–74.1%. In other words, roughly three in ten flagged articles were not confirmed. Common errors involved uncertainty unrelated to policy or policy vocabulary used outside economic contexts.

Independent reviewers agreed on 87.5% of the 120 double-coded articles. However, the audit was restricted to articles meeting the economy-and-uncertainty keyword conditions, and the researchers do not report a reliable population-level recall estimate. The dictionary was also selected using this sample, limiting its independence as a performance test. Future revisions, the authors acknowledge, require fresh validation.

Which events moved the index?

The researchers assess selected events against local baselines rather than simply comparing raw index heights, because levels rise over the sample. Among those episodes, the June 2016 Brexit referendum produced the largest positive standardized local deviation—not the highest absolute reading in the entire series.

Other pronounced increases accompanied the March 2020 COVID-19 lockdown, the September 2022 energy and inflation crisis, and the December 2023 change of government. Several selected geopolitical or political events did not produce comparable positive local deviations. These comparisons support the authors’ interpretation of a policy-focused measure, but they are not causal tests of individual events.

Monthly percentage changes correlate at approximately 0.26 with US EPU and 0.25 with global EPU. The authors interpret these modest associations as consistent with a predominantly domestic orientation. They do not quantify domestic versus external contributions. Moreover, the annotation guidelines allow foreign-policy uncertainty regardless of country, so domestic orientation is an empirical interpretation, not a strict geographic filter. Similar correlations with financial-volatility indicators also mean the shared international component cannot be identified specifically as foreign policy uncertainty.

Useful measurement, not demonstrated prediction

Higher index changes coincide with weaker Polish equities, zloty depreciation and higher sovereign risk premiums. The analysis does not establish causality or forecasting ability. Reported correlations select the strongest association across four lags, and significance levels are not adjusted for lag selection or multiple testing.

Editorial checks provide qualified reassurance: removing both politically aligned newspapers leaves a monthly-change correlation of 0.80 with the published index. A pooled test finds no systematic partisan amplification, although one title shows weak evidence in the predicted direction.

The monthly data are publicly available through the project website, GitHub and OSF under CC BY 4.0. The paper’s contribution is a documented measure of newspaper coverage that researchers can study alongside other indicators—not a direct reading of everyone’s uncertainty or proof of its economic effects.