News Sentiment
News sentiment about the asset over its last lookbackDays daily news snapshots (default 1, the latest). Headlines come from the GDELT global news feed, and a fine-tuned model scores every headline that names the company: whether it is really about the company, and whether it is positive, neutral or negative news for it (a headline it calls neutral scores 0; others score from -1.0 to 1.0). Each trading day has one snapshot at 09:15 New York, and a headline counts toward the first session whose 09:15 cutoff is at or after the time GDELT published it. Backtests and live trading read the same rows. lookbackDays counts trading sessions whether or not the asset had news in them; a single session is noisy, and about 20 sessions is the usual window for a sentiment rule. Only headlines the model judges to be about the company count, and copies of one story within 7 days count once per session. measure (default Mean): Mean is the average sentiment from -1.0 to 1.0, Median is the middle score (the average of the two middle scores for an even count), NetCount is positive minus negative headlines, ArticleCount is the number of qualifying headlines. Coverage begins on 2019-09-23: a window reaching earlier is measured over the sessions since then, and the backtest warns that it held less history than configured. Every measure has no value across sessions with no published snapshot, or when the latest snapshot due has not been published. Mean and Median also have no value when the window holds no qualifying headline; the counts are then 0. Headlines are matched to the issuer, so every listed share class reads the same news and META reads Facebook-era headlines.
News Sentiment
Definition
News sentiment about the asset over its last lookbackDays daily news snapshots (default 1, the latest). Headlines come from the GDELT global news feed, and a fine-tuned model scores every headline that names the company: whether it is really about the company, and whether it is positive, neutral or negative news for it (a headline it calls neutral scores 0; others score from -1.0 to 1.0). Each trading day has one snapshot at 09:15 New York, and a headline counts toward the first session whose 09:15 cutoff is at or after the time GDELT published it. Backtests and live trading read the same rows. lookbackDays counts trading sessions whether or not the asset had news in them; a single session is noisy, and about 20 sessions is the usual window for a sentiment rule. Only headlines the model judges to be about the company count, and copies of one story within 7 days count once per session. measure (default Mean): Mean is the average sentiment from -1.0 to 1.0, Median is the middle score (the average of the two middle scores for an even count), NetCount is positive minus negative headlines, ArticleCount is the number of qualifying headlines. Coverage begins on 2019-09-23: a window reaching earlier is measured over the sessions since then, and the backtest warns that it held less history than configured. Every measure has no value across sessions with no published snapshot, or when the latest snapshot due has not been published. Mean and Median also have no value when the window holds no qualifying headline; the counts are then 0. Headlines are matched to the issuer, so every listed share class reads the same news and META reads Facebook-era headlines.
Parameters
Asset, Measure (Mean, Median, NetCount, ArticleCount; default Mean), Lookback (whole trading sessions, 1 to 252; default 1, the latest 09:15 snapshot)
Output range: Mean/Median: -1 to 1; NetCount and ArticleCount: whole numbers. No value for any measure before 2019-09-23, across missing snapshots or while the due snapshot is unpublished; Mean/Median also has none when the window holds no qualifying news, where the counts are 0
Example
`News Sentiment(asset, Mean, 5) > 0.3` (news about the stock over the last 5 sessions has been clearly positive), or `News Sentiment(asset, ArticleCount, 1) >= 3` (at least three articles about it in today's snapshot)
Full indicator reference