Methodology
Avarieux surfaces public events with their sources cited and leaves the read to you. This page is the disclosure behind that: the academic literature that establishes why each public-record source carries research value, the principles any internal evaluation is held to, and the things we explicitly will not claim. There is no performance scorecard here, by design.
Every public-record source Avarieux surfaces has a peer-reviewed paper establishing why it carries research value. We didn't invent these findings; we make the underlying source legible at consumer pricing. The papers below are why the source is worth showing — not a claim about what any individual event will do.
Finding
Year-over-year changes in 10-K and 10-Q language are associated with subsequent stock returns. Companies that materially modify their filings underperformed by ~30 bps/month vs companies with stable language.
Why it matters here
The basis for surfacing filing-language changes as a documented public event. We show the section-level diff and cite the filing; the interpretation is yours.
Finding
Domain-specific (not general-purpose) sentiment lexicons are required for finance text. Generic NLP libraries misclassify ~75% of negative words in 10-K filings.
Why it matters here
Why we don't pipe SEC text through off-the-shelf sentiment APIs. Filing-language work requires finance-domain vocabulary or it returns noise.
Finding
Insider trades (Form 4) are associated with future returns when filtered for the right insiders and routine-ness. Non-routine trades by senior insiders were linked to ~7% abnormal annual returns.
Why it matters here
Why Form 4 filings are a documented public event worth surfacing — with the routine vs non-routine distinction shown, not editorialized.
Finding
Search-frequency spikes (Google Trends) preceded price moves on small-cap and IPO stocks in the sample studied.
Why it matters here
The academic basis for treating public attention across sources as a documented event worth recording — not a prediction.
Finding
Media tone (pessimism in WSJ columns) was associated with short-term price movements that reverted within a week in the sample.
Why it matters here
Why we treat a news event as a short-window fact with its timestamp and source, and never as a standing signal.
Finding
Prediction markets aggregated dispersed information efficiently for binary, time-bounded outcomes across the 50+ markets studied.
Why it matters here
Why a prediction-market price is a documented public data point worth surfacing alongside filings — shown as a quoted number with its source.
Finding
Beta alone does not explain the cross-section of returns. Size and value factors capture most of what beta misses.
Why it matters here
Why, where we report risk context, we report several factors rather than a single number.
Finding
Mispricing can persist longer than arbitrageurs can stay solvent. Information edges decay over hours-to-days, not seconds.
Why it matters here
Why Avarieux is a research-and-discovery tool, not an execution product. We surface what's public; we don't trade.
We do not publish a results scorecard on this site. What we will commit to publicly is the discipline any internal evaluation is held to — so that if we ever do report a number, you know how it was produced.
Any internal evaluation samples (ticker, date) pairs at random, never starting from the events we want to look good. Selecting on the dependent variable is the most common way backtests lie.
Evaluation uses only what was knowable at the timestamp in question — no look-ahead, no restated data substituted for what was actually published that day.
Where validation is done, it trains on one window and tests on the next, sliding forward — so nothing is tuned on the data it's then judged against.
Small samples are called small. Coverage gaps are named. We would rather under-claim and be right than over-claim and be quoted back to a user who lost money.
Every honest research product makes its limits clear, because the alternative is being wrong publicly. These are the things Avarieux will never tell you, regardless of how compelling the data looks.
Open any US-listed ticker — every event shown carries its source and timestamp. No sign-up required.