Paper
From official language to market-facing signal
War-Related Diplomatic Sentiment Observatory
This site presents WDSI (War-Related Diplomatic Sentiment Index) series built from official diplomatic texts, with cross-country comparison, 7-day and 30-day smoothed views, raw publication-day values, and CSV downloads.
Loading latest build time...
Citation
Please cite the paper when using this dataset
APA Style
Yang, C., Zhang, S., Kyriakou, I., & Papapostolou, N. C. (2026). When words move money: Diplomatic sentiment and international capital flows. SSRN. https://ssrn.com/abstract=6437669
Chicago Style
Yang, Cunyi, Shuchi Zhang, Ioannis Kyriakou, and Nikos C. Papapostolou. 2026. "When Words Move Money: Diplomatic Sentiment and International Capital Flows." SSRN. https://ssrn.com/abstract=6437669
Current selection
Loading
--
Loading summary data...
30-day smoothing
--
Loading summary data...
30-day change
--
Compared with the 30-day smoothed value 30 days earlier
Latest publication
--
--
Sample coverage
--
--
Countries / regions
--
Earliest observation
--
Latest observation
--
Overview
Cross-Country WDSI Signal Board
Start with the full cross-country picture, then move into each country series. The headline number on each card is the latest 7-day smoothed value: the more negative it is, the more tense the war-related diplomatic tone.
Explorer
Index Explorer
The default view shows the 7-day smoothed WDSI. You can also switch to the 30-day smoothed series or the raw publication-day mean to see the more discrete jumps on release days.
Loading chart...
Method
How Diplomatic Text Becomes WDSI
WDSI is neither keyword counting nor one-shot sentiment scoring. It comes from cleaning official diplomatic texts, identifying war-related sentiment in stages, validating the results through consensus checks, and aggregating them into a daily high-frequency index.
Collect and clean official texts
The paper starts from official releases by multiple foreign ministries, including press-conference transcripts, statements, briefings, and written responses. These texts are segmented, stripped of greetings and procedural language, and reorganized into comparable daily series.
Identify sentiment in three stages
The paper uses a multi-stage LLM pipeline that first screens whether a text is about war, armed conflict, or military security, then classifies sentiment direction, and finally assigns an intensity score on a discrete scale from -3 to 3.
Run consensus checks and manual audits
To reduce misclassification, the paper repeats each judgement multiple times. Only matching results pass directly; disagreements are reviewed before moving on. The pipeline is also audited against manually labelled samples to keep the classifications close to researcher judgement.
Construct the daily index series
The paper first forms a raw score on each publication day. When multiple war-related passages appear on the same day, they are combined into one conference- or statement-level score rather than simply averaged. Missing days are then forward-filled and smoothed with a 7-day rolling average to form the baseline WDSI.
Why It Matters
What We Have Already Found in This Index
Our paper shows that WDSI is not just a description of diplomatic tone. It is a high-frequency signal that links geopolitical expectations, cross-border capital flows, and market responses.
It moves before conventional risk measures
Our paper finds that war-related diplomatic wording often shifts before standard geopolitical risk indicators do, making it useful for detecting changes in market expectations around escalation or de-escalation earlier.
It changes the direction of cross-border flows
In the China sample, more negative diplomatic sentiment is associated with stronger capital outflow pressure. Our estimates show that a one-unit decline in China's WDSI corresponds to roughly a 35.6 million USD shift in daily net flows toward outflows.
It carries both safe-haven and spillover effects
Our paper also finds that more negative U.S. diplomatic sentiment is associated with stronger safe-haven demand for U.S. equities. Negative sentiment from the United Kingdom, Japan, and South Korea also spills over into the Chinese market, with clearer effects on returns and trading activity.
More value is still waiting to be found
The site exposes the raw series as well as the 7-day and 30-day smoothed versions, making it easier to extend the dataset into event studies, cross-country comparisons, asset-pricing work, and capital-flow tests.
Download
CSV Downloads
This section keeps the most direct download links for the full dataset and country-level CSV files, including raw publication-day values together with the 7-day and 30-day smoothed series.