Google Trends
Relative search interest index for each company's ticker and primary name, normalized with an anchor term (e.g., 'stock market') to make cross-date comparisons meaningful.
Signal construction
Queries use pytrends with a comparison anchor term so the 0–100 scale is consistent across dates. We batch requests to avoid rate limits and cache aggressively. Each series is converted to a weekly-to-daily interpolated z-score. Data degrades gracefully if the API is blocked.
Limitations & caveats
Unofficial API with no SLA. Sampling noise is documented; two queries for the same term on the same day can return different values. Geographic normalization limited to 'US'.
Evaluation results
Rank IC (Spearman) with Newey-West standard errors, walk-forward with 21-day embargo. Multiple-testing correction: Benjamini-Hochberg at q=0.05 across all signal × horizon × cohort tests.