Japan News Signal
ExtractAlpha’s Japan News Signal is a news-based NLP signal for Japanese stocks developed using Nikkei’s Flash News dataset. ExtractAlpha’s research team has developed a unique signal utilizing a hybrid of 1) a sentiment model constructed from Japanese Bidirectional Encoder Representations from Transformers (BERT), and 2) a machine learning model that uses embedded text vectors to predict long-horizon returns. This approach allows our signal to capture sentiment, as well as more persistent and latent predictive features of the news to identify greater alpha at longer horizons.
Nikkei Inc. is a leading financial media group in Japan and Nikkei Flash News is a proprietary dataset of Nikkei’s Japan-focused news articles. The historical data is available from 2011 with 1.5 million articles as of 2023. About one third of the news is related to listed companies, and the unique security coverage is comprehensive at around 3,000 per year.
Our backtest shows that a decile long-short portfolio built from this signal generates an average annual return of 14.3% with a Sharpe ratio of 1.65, and very modest drawdowns in the period March 2011 to February 2024, with positive returns in every year. In addition, we find that the return is not due to factor or industry tilts and does not diminish significantly even if the trading is delayed up to 5 days after the publication of the news.
To view the white paper, please contact us.