Spoken Alpha
We listen to earnings calls so you don't have to.
Spoken Alpha scores per-speaker language deviation against each executive's own call history — flagging tonal shifts before consensus catches them. Hold execs and analysts accountable for what they said, and how those deviations played out in the next print.
The pattern
When an executive's tone breaks from their own baseline, results usually follow.
We don't score against industry averages, sentiment dictionaries, or cross-company corpora. We score each speaker against their ownprior earnings calls — same speaker, same company, strict-prior dates. That's the only comparison that captures real tonal change.
Speaker
Damon Audia — Chief Financial Officer
What we flagged
Audia hedged +47% more than his trailing-4-call baseline — a sustained shift in qualifier density across forward-looking answers, not a one-line slip. The deviation methodology flags this as a Tier-1 signal: a meaningful break from the speaker's own prior style, no industry comparison required.
Excerpt — analyst Q on margin guidance
“I think it's probably fair to say we'd expect some pressure there in the near term, though obviously a lot depends on how the second half plays out — there are a few moving pieces.”
Trailing baseline density of qualifiers (“probably”, “obviously”, “a few”, “a lot depends”) is visibly elevated. Sample excerpt
Deviation z-score
Sample data+2.4σ
Hedge density Δ vs baseline
Sample data+47%
5-day price reaction
Sample data−8.2%
This is what every signal page on Spoken Alpha looks like — a Tier rating, the underlying deviation math, the speaker excerpts that drove it, and the realized outcome to grade ourselves with. Numbers here are illustrative until the per-speaker deviation pipeline and EODHD price-return joins are live.
The trade
Trade BEFORE the next print, not after the call.
The signal comes from a speaker's prior calls — no look-ahead. We use it to position before that speaker's next earnings event, then exit through the live-day move.
Entry
T-1 close
TWAP-sliced through the day before earnings. Slice count scales with cap bucket (mega 4 → small 12).
Hold
One trading day
Captures the overnight gap and the live-day open. Avoids multi-week squeeze risk that has burned earlier longer-hold variants.
Exit
Day B ~2pm ET
TWAP out before late-session noise. Live-day timing is grounded in the SEC 8-K Item 2.02 release timestamp, not a vendor calendar.
Backtest results: pending. The 1-day-hold harness is being rebuilt against authoritative 8-K release timestamps; an earlier 60-day-hold variant was shelved. We do not publish a Sharpe number until the harness clears walk-forward, deflated-Sharpe, and bootstrap-CI checks. See methodology for what we will and won't claim.
What you get
A longitudinal accountability layer for execs and analysts.
Executive track records
Each executive carries a longitudinal scorecard: deviation history, guidance-vs-print outcomes, skepticism received from analysts, and the trade signals their calls generated. Hold the speaker accountable across cycles, not just one print.
Analyst track records
Skepticism Accuracy Rating per analyst — when this analyst pushed back hard, how often did the print disappoint? Question style, skeptical exhibits, z-score trajectory. Know which sell-side voices to weight up.
Per-call deviation scores
Tier 1 / 1.5 / 2 / 3 ratings, qualifier density deltas, scored Q&A exchanges with interpretive labels (hedging, evasion, scripted, candid). Drill into the excerpt that moved the score.
Trade-ready alerts
D-7, D-3, and D-1 alerts route every Tier-1 setup through the same blotter your PMs already work in. Cap-bucket-aware sizing notes are surfaced inline with the alert.
Methodology
Honest about what's built and what's still being built.
Per-speaker longitudinal deviation
For a target call on date T, we compute the speaker's baseline across that speaker's prior calls only — same speaker, same company, strict-prior dates. Anything closer than strict-less-than leaks future information. The signal is a modifier on the LLM scorer's tier output, not a standalone label.
Live-day timing
The 1-day hold spans T-1 close → Day B ~2pm. Live-day timing is grounded in SEC 8-K Item 2.02 release timestamps (BMO before 09:30 ET, AMC after 16:00 ET), not a vendor calendar that can drift.
What we will not claim
We do not publish a backtest Sharpe number until the harness clears walk-forward with deflated-Sharpe and bootstrap-CI checks. We do not claim the strategy survives slippage and borrow costs in mid- and small-cap until we've modeled them explicitly. An earlier 60-day-hold variant was killed; that result does not speak to the 1-day mechanic and we won't conflate them in conversation.
What's still in flight
Per-speaker deviation v1 (multi-week build), 1-day-hold backtest harness, full-history 8-K Item 2.02 backfill, EODHD price-return joins for realized-outcome scoring. Any number on this page tagged Sample data is illustrative and will recalibrate when those pipelines are live.
Built on open research
The phenomena we score — hedging, conditionality, ownership detachment, vocal affect — are public-domain constructs from two decades of peer-reviewed accounting and forensic-linguistics research. Our contribution is the engineering on top. See the reading list →
Early access
We're onboarding a small set of funds for the first calibration cycle.
Tell us about your mandate and we'll set up a 30-minute call. We won't hand-wave the methodology, and we won't pitch you on results we can't back up yet.