NLP-derived signals from SEC filings and earnings call transcripts across three independent studies. Financial sector (37 companies), healthcare (34 companies), and diversified market (185 companies). Updated quarterly.
Companies whose forward-looking language in management discussion diverges from peer group tone show predictable underperformance. The peer delta is essential — absolute sentiment carries no signal.
When management tone on earnings calls diverges from the language in the corresponding SEC filing, the inconsistency predicts negative forward returns. The gap captures what executives choose to emphasize verbally vs. what they commit to in writing.
Sequential drift in MD&A sentiment — measuring how a company's filing tone shifts quarter-over-quarter relative to its peer group — predicts forward returns across the full 37-company financial universe.
| Company | Subgroup | Direction | Score |
|---|---|---|---|
| MFC | Life Insurance | Bearish | −2.35 |
| BNS | Canadian Bank | Bearish | −1.89 |
| TD | Canadian Bank | Bearish | −1.55 |
| DB | Capital Markets | Bearish | −1.38 |
| JPM | US Bank | Bearish | −1.26 |
| RY | Canadian Bank | Bearish | −1.01 |
| CINF | P&C Insurance | Bearish | −0.87 |
| UNM | Life Insurance | Bullish | +1.79 |
| FITB | US Regional Bank | Bullish | +1.37 |
| MKL | P&C Insurance | Bullish | +1.24 |
| PGR | P&C Insurance | Bullish | +1.21 |
| MS | Capital Markets | Bullish | +1.20 |
| Subgroup (N) | H2 IC | H9 IC | H11 IC | Coverage |
|---|---|---|---|---|
| Retail Banking (17) | +0.008 | −0.445★ | −0.193★ | H9 + H11 |
| Capital Markets (3) | +0.032 | −0.089 | −0.111 | Insufficient N† |
| P&C Insurance (11) | +0.085 | n/a‡ | n/a‡ | H2 only |
| Life Insurance (8) | +0.057 | n/a‡ | n/a‡ | H2 only |
| Canadian Banks (6) | +0.121★ | −0.315 | −0.170 | H2 + H9 + H11 |
Pooled IC across 232 observations, 17 banking companies, 2018–2025. No look-ahead bias. Signal anchored to actual filing and transcript dates.
| Pooled IC (180d) | −0.115 p=0.079 |
| Pooled IC (90d) | −0.125 p=0.056 |
| Walk-fwd mean IC | −0.193 |
| Correct direction | 6 / 8 yrs |
| 2023 annual IC | −0.336 N=14 |
| 2024 annual IC | −0.422+ N=17 |
| 2025 annual IC | −0.357 N=13 |
| L/S cumulative | +12.9% 19 qtrs |
| H1 Risk Specificity | 4.8× ↑ |
| H11 Transcript Gap | 1.7× ↑ |
| H9 Outlook Divergence | 1.3× ↑ |
| H2 Sentiment Drift | ~0× ↓ |
LM sentiment replicates at IC=-0.100 OOS (p=0.008, N=2,009). Medical devices is the strongest subgroup (IC=-0.295 OOS). Large pharma forward-looking ratio is a standalone signal (IC=+0.350 OOS). Managed care signals flip direction and are excluded from the standard product.
| LM full universe | IC = -0.100 |
| LM medical devices | IC = -0.295 |
| FLR large pharma | IC = +0.350 |
| H11 peer-rel (earnings) | IC = +0.162 |
| Haiku ROA | IC = +0.295 |
| llm_confidence | IC = +0.289 |
| llm_net_sentiment | IC = +0.331 |
| llm_risk_tone | IC = -0.375 |
| llm_forward_ratio | IC = -0.367 |
LM sentiment survives BH correction at IC=-0.090 (p<0.001, N=4,650). Financials within Dow30 is the strongest sector (IC=-0.249). Consumer Discretionary has the largest quintile spread (-30.40%). Panel fixed effects fail cross-sector (p=0.987), confirming sector homogeneity matters.
| Consumer Discretionary | +0.390 |
| Communication Services | +0.298 |
| Materials | +0.199 |
| Financials | +0.185 |
| Information Technology | +0.131 |
| Full universe | +0.190 |
| Consumer Discretionary | -30.40% |
| Financials | -19.93% |
| Information Technology | -16.07% |
| Consumer Staples | -12.90% |
| Health Care | -8.19% |
Signal scores are peer-relative z-scores computed from the most recent filing. Updated quarterly. Positive scores indicate bullish language relative to peer group history. Negative scores indicate bearish. These are research outputs, not trading recommendations.
Financial sector scores use composite z-scores (H2 + H9 + H11 where available). Healthcare and diversified study scores use LM sentiment z-scores. Haiku confidence scores available for 155 Dow30 tickers, 35 financial sector tickers, and 34 healthcare tickers. ROA prediction validated OOS across all three studies.
This dashboard presents outputs from an ongoing independent research initiative into NLP-based signals from corporate disclosures across three independent studies (financials 37 companies, healthcare 34, diversified 185). Nothing herein constitutes investment advice, a solicitation, or a recommendation to buy or sell any security. Signal scores are preliminary research outputs, not trading recommendations.
All signals are subject to continued out-of-sample testing and academic peer review. Past signal performance does not guarantee future results. Walk-forward validation covers 2021-2025 (5 OOS test years). The combined signal was discovered out of sample, not pre-specified. In-sample both LM and Haiku signals are contrarian and cancel (IS IC=-0.007). Healthcare Haiku trend weakens 2022-2025.
Financial sector: H9 and H11 validated for banking (17 companies) only. Three Canadian insurers (IFC, FFH, GWO) contribute filing-based signals only. Healthcare: managed care signals flip direction. Diversified: panel fixed effects fail cross-sector (p=0.987). ROA prediction survives macro controls in financials only (t=+4.77). Leverage and NIM predictions do not survive macro controls.
ContextQuant is an independent research initiative. Findings have not been subjected to peer review or independent replication. No formal backtest of the composite signal has been conducted specifically on the sector-specific universes. The performance metrics presented reflect individual signal out-of-sample IC and hit rates from walk-forward validation, not simulated portfolio returns.