Are analyst accuracy grades meaningful?

Across the ranked firms, no. The spread of 30-day direction accuracy between firms matches what sampling noise produces at their call counts, and a firm's accuracy in one half-year has no rank correlation with the next. dataphil prints a letter only where a record differs from a coin flip at two sigma, and ranks firms on 90-day price-target hit rate, the one trait that persisted.

How many calls does it take to prove an analyst is better than a coin flip?

About 1,200 scored calls to separate a 55% firm from 50% at two sigma; the median ranked firm has 77.

dataphil › Scorecard › Methodology

How the Street gets graded here

How every sell-side call on the site is priced, graded and checked, and why the firm leaderboard no longer hands out letter grades on direction: tested against the whole book on 2026-09-21, the spread of firm accuracy is what coin-flip noise would produce at these call counts. Today: 51 ranked firms, 2 beyond two sigma, chance predicts 3.

5908 calls on 100 names from 51 ranked firms, graded through 2026-09-21. The noise figures on this page are recomputed from the live data each morning; the split-half persistence test was run on 2026-09-21 on that day's seed.

What gets graded

Every sell-side rating change on a covered name from the last 365 days is priced the day it lands and checked 30, 60 and 90 days later. Two things are scored:

Direction at 30 days

A bullish call is right if the stock is higher 30 days later, a bearish one if it is lower. Buy and Sell ratings call their own direction. A downgrade to Hold is a bearish call, an upgrade to Hold a bullish one, and a maintained Hold whose target sits 10%+ from the price leans that way. A true Hold, target near the price, is counted and not graded.

Price target at 90 days

Every call with a target is checked 90 days later: hit if the stock reached it or came within 5%, close if it covered at least half the distance, missed otherwise.

The test: are the firm grades real?

Until 2026-09-21 every firm with 10+ scored calls carried a letter, A to F, on its 30-day direction accuracy. That letter was a claim: this firm is better (or worse) than the others. So the whole book was tested two ways.

Pooled accuracy

51.3%

all ranked firms together

Spread between firms

7.2 pts

observed sd of firm accuracy

Spread from noise alone

7.5 pts

if every firm were an identical coin

Firms beyond two sigma

2 of 51

chance predicts 2.6

1. The spread is noise. If every one of the 51 ranked firms flipped the same coin, sampling alone would scatter their accuracies with a standard deviation of about 7.5 points at their call counts. The real scatter is 7.2. 2 firms sit more than two sigma from the pool; with 51 firms, chance alone puts 2.6 there. To separate a genuine 55% firm from 50% at two sigma takes about 1,200 scored calls. The median ranked firm has 77.

2. It doesn't persist. Split the year at its midpoint, rank every firm on the first half, and its rank tells you nothing about the second: the rank correlation of 30-day direction accuracy across the two halves was −0.07 on 43 firms with 10+ calls in each. A firm that graded A in the first six months was no likelier than any other to grade A in the next. Following the top-ranked firms' calls in the half they weren't ranked on beat following the bottom-ranked firms' calls by nothing that survived a 20 bp round trip, in either direction of the split, at 30, 60 or 90 days. The one trait that did persist was the 90-day price-target hit rate, rank correlation +0.36.

What changed

Before 2026-09-21	Now
Every ranked firm carried a letter, A–F, on direction accuracy.	A letter prints only where the record differs from a coin flip at 1.96 sigma (two-sided test against 50%). Today that is 3 firms: Susquehanna B (62.1% on 103), TD Cowen C (58.3% on 163), Roth Capital F (29.2% on 24).
Leaderboard ordered by direction accuracy, shrunk toward the pool.	Ordered by 90-day target-hit rate, shrunk toward the pool by 40 phantom calls. Direction accuracy still prints, next to its 95% interval.
A percentage stood alone.	Every firm and name shows the 95% range its true rate could sit in. When that range includes 50%, the page says so.
Monthly report cards ranked firms on 8+ calls with no caveat.	The monthly board stays — it is a scoreboard for the month — but says on the page how wide a coin flip's band is at that call count.

What the site still claims

The Street as a whole runs about a coin flip on direction at 30 days — 51.3% across the year — and the monthly report cards show it month by month.
Misses lean one way. Most wrong calls were too optimistic; the ticker pages show which way each firm was wrong.
Calls chase the tape. Target raises land after the stock has already run; "run before the call" is on every table so you can see it.
Target accuracy is a real, persistent difference between firms. It orders the board.

Floors and fine print

A firm's accuracy on a single name prints only after 4 scored calls; a firm gets a page after 10 scored calls across every covered name. No firm gets a letter on a single name, ever — the samples are far too small. The interval on every page is a Wilson 95% interval. "Run before the call" is the stock's move over the 30 calendar days before the rating landed. Closed months' report cards are frozen and never edited. Data: Yahoo Finance rating history and daily closes, rebuilt each morning. Nothing here is advice.

Firm leaderboard → Monthly report cards →

How the Street gets graded here

What gets graded

The test: are the firm grades real?

What changed

What the site still claims

Floors and fine print

Get the calls before they're graded