Measuring Price Effects of Multilingual Global News with Large Language Models
A new study finds that language barriers slow financial markets’ response to global news, even when reports are nearly identical in Japanese and English. Using 15 years of Bloomberg news data and advanced language models, researchers show that English news predicts Japanese stock returns better than Japanese news alone. Japanese investors often miss foreign-language coverage, especially in smaller firms. The effect is weaker in the U.S. The study’s method separates shared from language-specific news content.
What it examines
This paper uses multilingual large language models to study how language barriers affect how quickly financial markets react to global news. By analyzing Bloomberg news in Japanese and English for Japanese and U.S. stocks, it measures how well markets incorporate news across languages and identifies sources of inefficiency.
What it concludes
The study finds that language boundaries slow market reactions to news, creating opportunities for investors who process news in multiple languages. This has practical use in building trading strategies and signals. The results suggest markets are not fully efficient and highlight the value of multilingual news analysis for finance.
Evidence objects
Language barriers slow financial market reactions to global news, even when content is nearly identical. Investors in Japan and the U.S. do not fully combine information from both Japanese and English sources.
key_findings bullet 1 · key_findings · validation V0
Surprisingly, English-only news predicts Japanese stock returns better than Japanese-only news, suggesting Japanese investors often miss or discount foreign-language coverage. In the U.S., English news dominates, but Japanese news still adds value.
key_findings bullet 2 · key_findings · validation V0
Researchers used advanced multilingual models and a novel sentence-level alignment method to separate shared and language-specific news. Inefficiencies are strongest in smaller, domestically owned firms and when readership is low.
key_findings bullet 3 · key_findings · validation V0
This paper uniquely investigates how multilingual LLMs reveal inefficiencies in price discovery caused by language segmentation in global news. Employing advanced NLP methods like multilingual sentence embeddings and optimal transport, it empirically demonstrates language-induced return predictability, offering novel insights with significant implications for AI-driven trading, international finance, and market efficiency.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
In Japan, both Japanese and English news predict next-day returns across individual stocks. Quantitative Methods in Investing & Financial Statement Analysis
Source row: 1321 · abstract type: snippet