LIVE EARNINGS ANALYSIS

Fetches the latest SEC 8-K filing, runs ProsusAI/FinBERT sentiment analysis, overlays the current MacroScope macro regime, and generates analyst insights with Claude.

38 earnings calls12 companies2022-10-25 — 2024-08-01Updated 2026-06-27

AVG SENTIMENT SURPRISE BY REGIME

Expansionbaseline +0.1500 · avg surprise +0.1415
Late-Cyclebaseline +0.0500 · avg surprise +0.1980
Recoverybaseline +0.1000 · avg surprise +0.0000
Contractionbaseline -0.2000 · avg surprise +0.3508

SENTIMENT SURPRISE FORMULA

sentiment_score = pos − neg
surprise = score − regime_baseline
▲ beats if surprise > +0.05
▼ misses if surprise < −0.05
● in line otherwise
Beats expectations32 / 38
Misses expectations2 / 38
In line4 / 38

COMPANY EXPLORER · HISTORICAL DATASET

Select a company above to view historical earnings sentiment analysis

REGIME-CONDITIONAL EARNINGS SENTIMENT

Expansion10 calls
avg surprise +0.1415
sentiment +0.292baseline +0.15
TOP SURPRISES
NVDA May 24+0.3300
MSFT Apr 24+0.2950
META Apr 24+0.2000
Late-Cycle15 calls
avg surprise +0.1980
sentiment +0.248baseline +0.05
TOP SURPRISES
NVDA Aug 23+0.6100
MSFT Oct 23+0.3800
AMZN Aug 24+0.2700
Recovery0 calls
avg surprise +0.0000

No calls in dataset

Contraction13 calls
avg surprise +0.3508
sentiment +0.151baseline -0.20
TOP SURPRISES
NVDA Feb 23+0.7200
MSFT Apr 23+0.5800
MSFT Oct 22+0.5300

METHODOLOGY

How EarningsEdge works ↓

DATA SOURCE — SEC EDGAR 8-K FILINGS

Earnings call transcripts are sourced directly from SEC EDGAR via the sec-edgar-downloader library. 8-K filings filed within 4 days of each quarterly earnings release are retrieved for 20 S&P 500 companies spanning Technology, Financials, Energy, Consumer, and Healthcare sectors. HTML is stripped and keyword filters ensure only genuine earnings call documents are processed.

SENTIMENT MODEL — ProsusAI/FinBERT

Each transcript is scored using ProsusAI/FinBERT, a BERT model fine-tuned on 10,000+ financial news sentences. The model outputs three probabilities (positive, negative, neutral) that sum to 1. Long transcripts are split into 400-word chunks (up to 8 chunks) and scores are averaged. The final sentiment_score = positive − negative, ranging from −1 (maximally bearish) to +1 (maximally bullish).

MACRO REGIME — MacroScope HMM

Each earnings call is stamped with the macroeconomic regime in effect at the time of the report, sourced from MacroScope — a Hidden Markov Model trained on FRED macroeconomic indicators (unemployment, yield curve, credit spreads, industrial production). The four regimes are Expansion, Late-Cycle, Recovery, and Contraction.

THE HYPOTHESIS — SENTIMENT SURPRISE

Corporate earnings calls are not delivered in a vacuum — management tone is partly a function of the macro environment. EarningsEdge quantifies whether a company's earnings tone was better or worse than the prevailing macro regime would predict. This “sentiment surprise” is the core analytical claim.

sentiment_surprise = sentiment_score − regime_baseline
beats_expectations if surprise > +0.05
misses_expectations if surprise < −0.05
in_line otherwise

REGIME BASELINES

Expansion+0.15
Late-Cycle+0.05
Recovery+0.10
Contraction−0.20

Baselines represent the empirical average FinBERT sentiment score observed historically during each regime across S&P 500 earnings calls.

NOT INVESTMENT ADVICE. EarningsEdge is an academic research project. Sentiment scores and regime annotations are derived from historical data and NLP models and should not be used as the sole basis for any investment decision. Past sentiment patterns do not predict future returns.