Paid DailyPDPaid Daily
Open the desk
QuantInsti Quantitative Learning

How to Trade the News: Turning Noisy Sentiment Data Into a Real Signal (Python)

NowPress play to follow along0:00 / 8:54
Chapters5 segments · tap to seek
ticker lesson

Breakdown

0:00lesson

Filtering news with metadata

  • News filtering requires metadata not keywords
  • LSEG data anchors sentiment analysis
  • News file and scores file tabs structure data
  • Each row represents a news item with timestamp, headline, language, source, and subjects
1:49lesson

Sentiment scoring for companies

  • News items stamp probabilities for companies
  • Three-class sentiment: -1/0/1 scores
  • Micro sentiments capture term-specific signals
  • Commodity production volume sentiment is critical
3:40lesson

Price overlay with sentiment

  • Sentiment data is high-frequency and noisy
  • Kalman filter smoothes volatile signals
  • Alibaba example shows daily sentiment summary
  • Raw sentiment fluctuates but patterns emerge
5:29lesson

State space models for filtering

  • Kalman filter grounds high-frequency data
  • State space models reduce noise systematically
  • Sentiment becomes a model-informed signal
  • Filtering adds predictive value to raw data
7:24lesson

Multi-dimensional sentiment correlations

  • Price correlates with volatility and supply-demand
  • Trust sentiment and price forecasts align with price
  • Correlation plots reveal factor relationships
  • Multi-dimensional analysis provides richer signals

Informational only — this is QuantInsti Quantitative Learning’s content, decoded by Plutus. Not Paid Daily’s advice or a recommendation. The outline, timestamps, and claims are extracted from what the creator said; verify before acting.