Interesting read. Do you think it’s possible that with a “news_lag” of only 1 you may be taking into account news that is actually occurring after your prediction time-index due to different timezones being used from your news data source and your stock price data source?