jev-alpha-bench
Study of whether Jev's reading of financial news predicts stock returns: strong same-day signal, no tradeable alpha.
Overview
jev-alpha-bench asks whether Jev can predict stock returns from financial news, built on the author's jev-go client. Nasdaq-100 headlines from January 2022 to October 2023, one per stock per day, each get one request with three questions: a Choice over bullish, bearish or neutral, a Score for how big a move it justifies, and a Noul for whether it is already priced in. The pre-registered signal is P(bullish) minus P(bearish), with returns measured in excess of QQQ and confidence intervals from a bootstrap over dates.
Run on 16 September 2026 against jev-1.13.0 over 5,000 events and 15,000 requests in three arms, the result is that the signal has a rank IC of +0.24 against the publication-day move but -0.008 one full close later, when a trade is actually possible, and a long-short book loses 18 bps per trade after costs. The README carefully explains the dataset's timing offsets and a second study giving Jev price candles without news, concluding it reads news well, cannot read a chart, and there is no money in either.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Is this headline bullish, bearish or neutral for the stock? | bullish · bearish · neutral |
| Score | How big a move does this headline justify? | — |
| Yes / no | Is this news already priced in? | yes · no |
How it works
- Collect one headline per stock per day
- Ask Jev three questions per headline
- Compute P(bullish) - P(bearish) signal
- Measure rank IC and long-short returns across timing windows
- Bootstrap by date for confidence intervals
Strengths
- Pre-registered signal and careful timing analysis
- Separates reading the news from tradeable prediction
- Raw per-event output published
Caveats
- Negative result: no tradeable alpha
- Historical period and single news corpus
At a glance
- Type
- Benchmark
- Status
- research
- Language
- Go
- Runtime
- benchmark scripts
- Has code
- yes
Use cases and integrations
How curated lists describe it
Does Jev predict stock returns from news? It reads the news well; there is no tradeable alpha. Three arms separate reading from recall.
hellogumbo/awesome-jev
Frequently asked questions
Does Jev predict stock returns?
It reads news well (IC +0.24 same day) but the signal vanishes by the time you can trade (IC -0.008).
Can it read price charts?
The second study concludes it cannot.
Is the result reproducible?
Raw per-event outputs are published in results/.
Related projects
PyTorch factor-research stack with 213 factors and cost-aware backtests, plus a preview typed portfolio-decision adapter.
ProjectbetaPythonCrypto terminal that condenses up to 1,000 posts plus market data into a Jev buy, sell, hold or take-profit card; no trading.
ProjectprototypePythonTrade Arena proof of concept comparing trading strategies on 13 US stock funds, grown out of a pre-registered Jev study.
ProjectprototypePythonResearch wiring LAYA/Jev typed decisions into a quant stack with Claude fallback: good news reading, no market-beating edge.
ProjectresearchPythonReproducible test of viral Jev HFT claims: direction not calibrated, never under 100 ms, and no edge after trading fees.
ServiceresearchPythonPoint-in-time O'Neil momentum backtest where Jev picks entries and judges exits, with a jackknife showing no resolvable edge over rules.
ProjectresearchPython