101Music Equities
¿Puede la atención en Wikipedia anticipar los ingresos de las empresas musicales antes de que se publiquen? No: fallaron todos los criterios fijados de antemano salvo el placebo.
- Año
- 2026
- Estado
- Activo
- Python
- pandas
- scikit-learn
- SEC XBRL
- Wikipedia pageviews
- walk-forward
- pre-registration
Esta página solo está disponible en inglés.
The question
Can public attention data forecast a music company's quarterly revenue before it is reported? A quarter closes, every show and pageview is already public, and the revenue number arrives about 35 days later. I nowcast at quarter-end plus 10 trading days, on average 22 days before the print.
What I did
Wrote and froze the plan before any model ran: nine US-listed companies, including Sonos as a placebo, an anchored walk-forward over 2019–2023, three naive baselines to beat, a shuffled-label null, COVID years excluded from the headline in advance, and a lockbox from 2024-07-01 onward. Data came from SEC filings (first-filed values only, 64 restatements caught), 8-K earnings dates, Wikipedia pageviews and prices.
What came out of it
No edge. The nowcast beat 1 of 3 baselines, sat at p = 0.68 against the shuffled-label null, and was positive in 0 of 5 folds. Wikipedia pages for Photosynthesis, Chess and Coffee scored −0.78, better than the curated music map at −2.36. Adding attention to plain persistence raised the error from 0.2032 to 0.3435. The lockbox was never opened.
What I'd try next
Model per-show economics instead of per-quarter revenue, which turns roughly 200 observations into roughly 50,000, and start collecting forward ticket inventory now, because it can never be backfilled.
