Global vs. Local News and Future Equity Market Returns

Is relevant news sentiment from other countries additive to local news in predicting country stock market returns? In their August 2025 paper entitled “Global News Networks and Return Predictability”, Gustavo Freire, Ali Moin, Alberto Quaini and Amar Soebhag compare the effects of local versus global news sentiment, organized into 260 major themes, on country stock market returns. They apply random forest machine learning to news sentiment variables to construct out-of-sample market timing strategies based on local versus global sentiment measures. Specifically, they:

• Fit random forests on the initial 18 months of local or global data to select base models.
• Use the next six months to select optimal hyperparameters.
• Retrain the model every 20 trading days with an expanding window, so that each new fit uses all data up to the end of the previous month.
• Each day for each country market, go long (short) that market for that day if its projected return is positive (negative).

Using Global Knowledge Graph Database news sentiment data for 520 million articles from 14 developed countries and associated country market returns computed from futures contract prices for 14 equity indexes during February 2015 through December 2023, they find that:

(more…)