A proprietary deep learning system analyzes every 10-K and 10-Q for 2,456 US equities and generates daily directional predictions — validated live for 5 years with 200,000+ timestamped trades. No LLM. No hallucination. No API dependency.
A proprietary deep learning architecture processes two information streams to generate daily predictions for each of the next 10 trading days, rolling.
Every 10-K and 10-Q for 2,456 equities. Positive and adverse events and conditions identified but are overlooked in filings that carry predictive power.
Daily market data fused with text features. Cross-modal signal extraction captures what neither source alone reveals.
Return prediction for each of the next 10 trading days on a rolling basis. No hallucination, deterministic inference from identical inputs.
Pre-market via AWS S3. One row per ticker per day. Quintile rank, signal score, recommended side, prediction horizon.
System completed 18 months before large language models entered public awareness. Not an LLM wrapper. Purpose-built deep learning trained specifically on the structure and semantics of SEC filings and market data.
Yale/Goldman Sachs Fin-RATE study (Feb 2026): 17 large language models achieved 43.5% accuracy on filing-based prediction. Our system exceeds 80%. Purpose-built architectures outperform general-purpose models on specialized tasks.
Once the engine runs, every additional product is packaging and positioning. SIGNAL, SELECT, SHIELD, and SHIELD+ are all derived from the same prediction stream. One engine, four products.
Cumulative PnL across 13 overlapping 18-month out-of-sample rolling windows. Every window profitable.
Each product packages the same underlying intelligence for a different institutional buyer.
Daily prediction file delivered pre-market. Continuous signal scores, quintile ranks, and 5 years of historical archive for independent backtesting.
Quarterly Q5 Long and Short baskets with one-page filing attribution summaries. What changed in the filing and why the model believes it matters.
Quarterly list of 40–80 stocks predicted to underperform. No shorting required. Simply exclude from your long portfolio. Based entirely on public SEC filings.
Everything in SHIELD plus Q1 Long names that won't keep up. Two exclusion types — decline risk and underperformance risk — concentrating capital into stronger opportunities.
How the model identified deteriorating disclosure quality long before the market priced it in.
The model detected deteriorating disclosure quality in Werner's SEC filings six consecutive quarters before the stock declined 37.6%. This is precisely the type of early warning that SHIELD and SELECT are designed to surface.
No analyst team in the world manually reads 6,000+ filings per year with this consistency. The model does — every quarter, for every company in the universe.
Every statistic is sourced from live, out-of-sample data or independently validated analysis.
Complete factor attribution using Fama-MacBeth cross-sectional regression, 13 rolling windows.
| Criterion | N10 | N10SF | Detail |
|---|---|---|---|
| Q5 Long Alpha Significant | PASS | PASS | 24.2 bps t=8.13 • 38.0 bps t=7.59 |
| Q5 Short Alpha Significant | PASS | PASS | 22.0 bps t=4.08 • 29.1 bps t=4.71 |
| Alpha Positive All 13 Windows | PASS | PASS | Both sides, both strategies: 13/13 |
| Quintile Monotonicity Post-Adjustment | PASS | PASS | Q5–Q1: 41.9 bps (L) • 39.8 bps (S) |
| Survives Bear Market Regime | PASS | PASS | 17.0 / 30.8 bps, t>12 |
| Survives Bull Market (Long) | PASS | PASS | 20.5 / 29.9 bps |
| Market Neutrality | PASS | PASS | Near-zero MKT beta for Long |
| Size Neutrality | PASS | PASS | SMB loading insignificant in all Q5 specs |
| Idiosyncratic Component | PASS | PASS | R² = 0.22–0.45 — substantial unexplained variation |
63 trials in pipeline across the world's most sophisticated institutional investors, built organically in 5 months.
Sid built Deals Intelligence at LSEG/Refinitiv — the platform used by the top 50 global institutions powering $3T+ in annual transactions. That experience gave him direct relationships with the exact buyer personas now in the pipeline: heads of data, quant teams, and CIOs at the world's largest funds.
He co-authored the Federal AI Maturity Model at the White House and served as Expert Technical Witness for the United States Congress on AI policy — credentials that materially reduce procurement risk for compliance-sensitive buyers like pensions, insurance companies, and sovereign wealth funds.
Sid built the entire Increase Alpha deep learning system from first principles: NLP feature engineering, LSTM architecture, and production signal pipeline. Completed 18 months before ChatGPT launched. 100% proprietary. Zero third-party AI dependency.
SEC filings are the single most information-dense, legally mandated disclosure in the US equity market. Every public company must file. Every filing is public. The edge is in reading all of them, systematically, and extracting the signal that human analysts cannot capture at scale. Evidence that no LLM can extract accurately.
Your team reads 200 filings per quarter. Our system reads 6,000 and tells you which companies are getting better and which are getting worse.
We are evidence-first. Every statistic we cite is sourced from live, out-of-sample data or independently validated analysis. We do not cherry-pick backtest windows. We do not optimize in-sample. We publish the track record as it is.
Backed by Eleven International for media and PR. Conference circuit active. SSRN paper published. Strategic partnership conversations with LSEG, CBOE, S&P Global, and Bloomberg.
Request a trial of any product. We provide the data, the historical archive, and the methodology documentation. You run your own evaluation.
sales@increasealpha.com — We respond within 24 hours.