Yahoo's epsCurrentYear can freeze at a stale vintage while live consensus moves

Claim

On Nu Holdings, Yahoo's epsCurrentYear read $0.88833 on 2026-07-31 and again on 2026-09-28 — unchanged across two months — while the live FY26 consensus in the same vendor's eps_trend table moved $0.808 → $0.853 (+5.6%).

The 2026-07-31 analysis built its #1 ranked risk on the frozen field: that consensus was ~18% too high because it rested on a tax rate management had disowned, making a "mechanical downgrade cycle" near-certain and dated. The frozen field overstated the consensus being attacked by roughly 9%, which made the gap to the analysis's own tax-normalised estimate look far larger than it was.

The claim was RETRACTED eight weeks later. The downgrade cycle never came: the effective tax rate did rise 8.69% → 14.17%, and NU beat anyway because pre-tax income grew +29.6% QoQ and absorbed it. FY26 estimates were revised up — 8 up, 0 down in 30 days — and no analyst asked a single tax question on the call.

The rule

Never quote a consensus EPS from epsCurrentYear or forwardEps. Use the eps_trend table, which carries analyst counts and a 7/30/60/90-day revision history, so a frozen value is visible as a flat series rather than invisible as a single number.

And before ever predicting a downgrade cycle, read eps_revisions — the up/down counts. A forecast that estimates are about to fall is directly falsifiable by a field showing 8 up and 0 down, and checking it costs one lookup.

Why a single number cannot be trusted here

A point estimate carries no freshness signal. The same $0.88833 reads identically whether it was computed this morning or in March, so there is no way to tell a live field from a dead one without a series to compare it against. That is the whole argument for the eps_trend table: it is the same data with a timestamp attached.

Related

• [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]] — the sibling defect in the adjacent field, and also confirmed still live on NU today: forwardPE 11.03 runs off FY27, understating the FY26 multiple by ~32%. One vendor, two EPS fields, two different lies.
• [[pitfall-multiple-trim-inherits-the-broken-vendor-field]] — what happens downstream when a broken EPS field feeds a rendered price level.

History

• 2026-09-28 — written after the NU re-analysis retracted the July downgrade-cycle thesis. The retraction had two distinct causes and both are worth keeping separate: a reasoning error (the forecast held pre-tax income fixed while the tax rate rose) and this vendor error. Either alone would have weakened the claim; together they made a non-existent risk the file's top-ranked one.