Earnings call data sharpens euro area job vacancy forecasts
A new ECB working paper shows that extracting text-based labour demand indicators from corporate earnings calls significantly improves short-term forecasts of euro area job vacancy rates. The high-frequency indicator provides timely signals ahead of official statistical releases.
High-frequency signals from corporate transcripts
Analyzing nearly 38,000 earnings call transcripts of euro area firms from 2002 to 2025, researchers Agostino Consolo, Claudia Foroni, Claudio Lissona, and Christofer Schroeder constructed a monthly labour demand index based on specific keyword counts.
When integrated into a mixed-frequency Bayesian VAR alongside traditional indicators, the text-based metric yields a 3 percent improvement in nowcast accuracy for euro area job vacancy rates.
The analysis highlights marked sectoral differences: qualitative signals from the manufacturing sector prove significantly more informative than those from services.
Furthermore, hard economic indicators like the unemployment rate add little predictive power once soft, forward-looking survey metrics are included.
Overcoming the publication lag
Official job vacancy statistics in the euro area suffer from a two-month publication lag, frequent revisions, and a short historical record, creating serious blind spots for monetary policy.
While survey measures like factors limiting production offer valuable proxies, high-frequency corporate earnings calls fill a vital gap by delivering unrevised data updated biweekly.
This alternative data source provides central bankers with real-time visibility into emerging labour market tightness long before traditional macro statistics are finalized.
A welcome edge, but no silver bullet
The study convincingly demonstrates how alternative text data can refine short-term economic forecasting.
Yet, an over-reliance on earnings calls risks slanting policy analysis toward large listed industrial firms.
Technological innovation enhances policy toolkits, but central bankers must remain mindful of structural sample biases.