Prediction accuracy leaderboard

Ranked by walk-forward prediction accuracy (AUC): how well the model calls this investor's next new buy, out of sample. This is not a return ranking. AUC 0.50 is a coin flip; higher is better. Samples are small and every per-investor edge is noisy and not statistically significant. Research, not advice.

Every other smart-money tracker ranks realised returns. This ranks something only a prediction model can: how well the next new buy is called before it files. As of 2026-09-30. 26 followed investors.

AUC here grades the model on each investor’s own disclosed book, which is a narrower question than the live board answers. The number that describes the product is measured instead: across 2,030 investor-quarters of walk-forward holdout, the top-ranked name was the actual next new buy 1.7% of the time against 0.62% at random, which is 2.77x chance. Full scoring detail on how it works.

#	Investor	Predict AUC	Ahead of index	Quarters	Top predicted buy
1	Renaissance Technologies Renaissance Technologies	0.796	38%	34	INTU
2	Tom Russo Gardner Russo & Quinn	0.789	38%	34	SUNB
3	Prem Watsa Fairfax	0.757	35%	34	RRC
4	Chris Bloomstran Semper Augustus	0.754	41%	34	SUNB
5	Howard Marks Oaktree	0.739	44%	34	BKLN
6	Bruce Berkowitz Fairholme	0.709	41%	34	RRC
7	Nelson Peltz Trian	0.705	44%	34	RRC
8	Philippe Laffont Coatue	0.697	41%	34	INTC
9	Chuck Akre Akre Capital	0.687	56%	34	RRC
10	David Tepper Appaloosa	0.679	53%	34	BKLN
11	Chase Coleman Tiger Global	0.678	44%	34	PAYP
12	George Soros Soros Fund Mgmt	0.674	50%	34	COO
13	Larry Robbins Glenview	0.659	32%	34	COO
14	David Abrams Abrams Capital	0.657	26%	34	RRC
15	Carl Icahn Icahn Capital	0.650	38%	34	RRC
16	Joel Greenblatt Gotham	0.649	32%	34	BEPC
17	Andreas Halvorsen Viking Global	0.637	62%	34	COO
18	Lee Ainslie Maverick	0.634	41%	34	DT
19	David Einhorn Greenlight	0.632	42%	26	RRC
20	Two Sigma Two Sigma	0.620	53%	34	RRC
21	Daniel Loeb Third Point	0.618	47%	34	RRC
22	Brad Gerstner Altimeter	0.617	35%	34	RRC
23	Stephen Mandel Lone Pine	0.612	38%	34	RRC
24	Stanley Druckenmiller Duquesne	0.604	44%	34	DHI
25	Francois Rochon Giverny	0.575	53%	34	WSO
26	Ray Dalio Bridgewater	0.571	35%	34	DIDIY

Below the line: not followed

Three separate reasons put an investor below the line. Four high-turnover market-makers (Citadel, Millennium, D. E. Shaw, Point72) score high AUC because their 13Fs are huge churning books that are easy to predict but say little about conviction, so mirroring them was never the idea. Three more (Buffett, Klarman, Oakmark) left the followed consensus on 14 August 2026 for the opposite reason: we ran an event study on their own new entries against matched controls, and after adjusting for market exposure the entries underperformed by 3 to 5 percentage points with a confidence interval below zero. The rule that demoted them was written before the result was seen. Their per-investor models and pages stay published, and the test covers 40% of new-entry rows and fits a single market beta, so read it as what this instrument measured rather than a verdict on the manager. Gate: runs/gates/q5_leader_skill.json. Two more (Bill Ackman 0.487, Mohnish Pabrai 0.357) are below the line because the model ranks their next buy no better than a coin flip out of sample. Neither number is statistically distinguishable from 0.50 at this sample size. We hold the floor anyway: an investor our own model cannot call is one we do not put in the followed consensus.

Investor	Predict AUC	Quarters
Ken Griffin Citadel	0.913	34
D. E. Shaw D. E. Shaw	0.900	34
Izzy Englander Millennium	0.887	34
Bill Nygren Oakmark	0.735	34
Warren Buffett Berkshire Hathaway	0.716	34
Seth Klarman Baupost	0.709	34
Bill Ackman Pershing Square	0.651	34
Steve Cohen Point72	0.597	34
Mohnish Pabrai Dalal Street	0.467	24

AUC is out-of-sample walk-forward: each quarter is predicted using only earlier quarters. 0.50 is a coin flip. Ahead-of-index is the share of backtested quarters in which the investor’s predicted top-10 finished ahead of the S&P 500 (gross, before costs). Samples are small and none of these per-investor edges is statistically significant. See how it works, calibration, and the methodology. Research, not investment advice.