WHO CALLED IT RIGHT
How this works

Methodology

“Probability is the very guide of life.”

— Bishop Joseph Butler, 1736[1]
§ 1

Data sources

  • SEC EDGAR 13F-HR filings. Quarterly position disclosures required of institutional managers with $100M+ AUM. Filings are due 45 days after quarter end. Q1 2026 filings posted around 2026-05-15.
  • Public statements. Verifiable quotes from CNBC, Bloomberg, podcasts, X posts, and quarterly letters (Oaktree memos, GMO letters, ARK commentary).
  • Price data. End-of-day closes from public market data providers, snapshot at the time of build.
§ 2

From 13F to a “call”

A 13F shows quarter-end positions, not intra-quarter trades. We translate each delta into an implied directional view:

  • NEW or INCREASED → BULLISH
  • EXITED or DECREASED → BEARISH
  • HELD → NEUTRAL (not scored in v0)
§ 3

Brier scoring rule

Brier (1950)[2] = (forecast − outcome)². We treat each directional call as a confidence-1 forecast and score against the realized 30-day move.

  • Bullish call: outcome = 1 if price moved up (≥ 0%), else 0.
  • Bearish call: outcome = 1 if price moved down (≤ 0%), else 0.
  • Magnitude weight scales the penalty: a 2% wrong move is less bad than a 20% wrong move. Floor 0, cap 1.

Lower Brier = better calibration. 0.00 means perfect; 1.00 means maximally wrong.

§ 4

Color tiers

  • green ≤ 0.20 — well-calibrated
  • yellow 0.21 to 0.40 — mixed
  • red> 0.40 — poorly calibrated for this set
§ 5

What v0 skips

  • Position sizing. A 0.1% portfolio nibble and a 5% conviction add count the same. A production version would weight by % of fund AUM.
  • Risk-adjusted returns. Raw price move only. No beta adjustment, no sector hedge attribution.
  • Probabilistic forecasts.Real Brier expects a stated probability (e.g., “65% chance NVDA up by year-end”). 13F filings don't disclose probabilities. We use confidence=1 as a coarse approximation.[3]
  • Selection bias. We track 5 of 19 funds with real Q1 data; the other 14 are stubs. Roster will fill in as 13Fs are parsed.
  • Survivorship bias.We only score positions disclosed. Closed positions we don't see don't penalize.
§ 6

Update cadence

Data lives in github.com/alex-jb/whocalleditright-data. The site re-fetches every 30 minutes (ISR). Quarterly 13Fs land in Feb / May / Aug / Nov; we re-score the prior quarter when new data arrives. Public statement calls update continuously.

§ 7

Not financial advice

This site is a research and calibration audit. It is not investment advice, an offer to sell, or a solicitation. The author holds no positions in any name shown. Past calibration does not predict future calibration.

References
  1. [1] Butler, J. (1736). The Analogy of Religion, Natural and Revealed. London.
  2. [2] Brier, G. W. (1950). “Verification of forecasts expressed in terms of probability.” Monthly Weather Review, 78(1), 1–3.
  3. [3] Gneiting, T., & Raftery, A. E. (2007). “Strictly proper scoring rules, prediction, and estimation.” Journal of the American Statistical Association, 102(477), 359–378.