Filing Intelligence for Institutional Investors

We Read Every SEC Filing.
You Get the Alpha.

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.

70%
Combined Win Rate
1.36
Sharpe Ratio
+0.029
Information Coefficient
5yr
Live Track Record
IC +0.029***
13/13 Windows Profitable
Beta −0.14
Momentum Exposure 0.00
18M Long Win Rate 80.2%
Q5 Short Accuracy 83.0%
Universe 2,456 equities
FF5 Alpha +24.2 bps***
IC +0.029***
13/13 Windows Profitable
Beta −0.14
Momentum Exposure 0.00
18M Long Win Rate 80.2%
Q5 Short Accuracy 83.0%
Universe 2,456 equities
FF5 Alpha +24.2 bps***

100% Proprietary

Zero dependency on any third-party AI. Built 18 months before ChatGPT launched.

Zero Momentum Exposure

βMom = 0.00 at all horizons. Genuinely orthogonal to your existing book.

Glass Box Architecture

Every prediction traceable to specific filing features. Deterministic, reproducible outputs.

Institutional Validation

Fama-French 5-Factor, Bonferroni correction, 12-point robustness battery.

The Engine

One Engine. Four Products.

A proprietary deep learning architecture processes two information streams to generate daily predictions for each of the next 10 trading days, rolling.

01

SEC Filing Ingestion

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.

02

OHLCV Integration

Daily market data fused with text features. Cross-modal signal extraction captures what neither source alone reveals.

03

LSTM Prediction

Return prediction for each of the next 10 trading days on a rolling basis. No hallucination, deterministic inference from identical inputs.

04

Signal Delivery

Pre-market via AWS S3. One row per ticker per day. Quintile rank, signal score, recommended side, prediction horizon.

Pre-ChatGPT Architecture

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.

LLM Ceiling Demonstrated

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.

Four Products

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.

Live Track Record

5 Years. 13 Windows. 100% Positive.

Cumulative PnL across 13 overlapping 18-month out-of-sample rolling windows. Every window profitable.

Long Book
Short Book
Combined
+24.9%
Mean Window PnL
1.36
Combined Sharpe
−15.3%
Max Drawdown
4.01
Sortino Ratio
Product Suite

Four Products. One Signal.

Each product packages the same underlying intelligence for a different institutional buyer.

Live • 63 Trials

SIGNAL

The Quantitative Signal Feed

Daily prediction file delivered pre-market. Continuous signal scores, quintile ranks, and 5 years of historical archive for independent backtesting.

IC +0.029
Info Coefficient
200K+
Timestamped Trades
βMom=0.00
Momentum Neutral
+9–14%
Ann. OOS Alpha
Team and Enterprise Pricing
For: Multi-strat platforms, systematic managers, SWFs
Contact Sales →
Product Built

SELECT

Filing-Validated Stock Selection

Quarterly Q5 Long and Short baskets with one-page filing attribution summaries. What changed in the filing and why the model believes it matters.

80.2%
18M Long Win Rate
83.0%
18M Short Accuracy
+44.5%
Mean Long Return
ρ=1.00
Quintile Monotonicity
Per Ticker, Team, and Enterprise Pricing
For: Fundamental L/S managers, equity research teams
Contact Sales →
Product Built

SHIELD

Defensive Exclusion Screen

Quarterly list of 40–80 stocks predicted to underperform. No shorting required. Simply exclude from your long portfolio. Based entirely on public SEC filings.

83%
Exclusion Accuracy
−25.3%
Avg Decline 18M
13/13
Windows Positive
t=8.34
Statistical Sig.
Team and Enterprise Pricing
For: Pensions, endowments, insurance, RIAs
Contact Sales →
Planned

SHIELD+

Dual-Layer Defense

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.

Q5 vs Q1 Return
80–160
Names Excluded
2
Defense Layers
Q5+Q1
Dual Screening
Team and Enterprise Pricing
For: Long-only managers seeking premium screening
Contact Sales →
Case Study

Six Quarters of Early Warning.

How the model identified deteriorating disclosure quality long before the market priced it in.

Q3 2022
First detection. Model flags deteriorating language in risk factor disclosures and management outlook. Stock enters Q5 Short list.
Q4 2022 – Q1 2023
Signal persists. Consecutive quarterly filings show worsening disclosure quality. Model maintains Q5 Short conviction across two more filing cycles.
Q2 2023
Fundamental cracks emerge. Revenue deceleration begins appearing in financial statements. Analyst community starts noticing. Stock still on SHIELD exclusion list.
Q3 2023 – Q1 2024
Price decline accelerates. Stock falls through support levels as fundamental deterioration becomes consensus. Total decline from initial detection: −37.6%.

Werner Enterprises

NASDAQ: WERN • Transportation & Logistics

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.

−37.6%
Stock Decline
6
Quarters Early Warning
Q5 Short
Consistent Signal
By the Numbers

Evidence-First.

Every statistic is sourced from live, out-of-sample data or independently validated analysis.

70%
Combined Win Rate
72.2% long, 67.8% short. Across 200,000+ timestamped predictions over 5 years.
1.36
Sharpe Ratio
Long/Short combined. Long: 1.42. Short: 1.30. Sortino: 4.01 combined.
13/13
Positive Windows
Every rolling 18-month OOS window profitable. 100% consistency across market regimes.
+0.029
Information Coefficient
t = +3.01*** across 676 unfiltered cross-sections. Survives Bonferroni at 11/12 horizons.
0.00
βMomentum
Zero momentum exposure at ALL horizons from 3M to 36M. Genuinely orthogonal to your existing book.
83%
Short Selection Accuracy
Q5 Short names decline over 18 months. t = 8.34, p<0.000002. Extremely robust.
+24.2
FF5 Alpha (bps/trade)
Factor-adjusted, Fama-MacBeth. t = 8.13***. Positive in all 13 windows.
ρ=1.00
Quintile Monotonicity
Perfect Spearman rank from Q1 to Q5. The model correctly orders the entire universe.
−0.14
Market Beta
Near-zero S&P 500 correlation. Market-neutral by construction, not by hedge ratio.
Due Diligence

Fama-French 5-Factor Scorecard

Complete factor attribution using Fama-MacBeth cross-sectional regression, 13 rolling windows.

Fama-French 5-Factor scorecard: all 9 criteria pass for both N10 and N10SF strategies.
CriterionN10N10SFDetail
Q5 Long Alpha SignificantPASSPASS24.2 bps t=8.13 • 38.0 bps t=7.59
Q5 Short Alpha SignificantPASSPASS22.0 bps t=4.08 • 29.1 bps t=4.71
Alpha Positive All 13 WindowsPASSPASSBoth sides, both strategies: 13/13
Quintile Monotonicity Post-AdjustmentPASSPASSQ5–Q1: 41.9 bps (L) • 39.8 bps (S)
Survives Bear Market RegimePASSPASS17.0 / 30.8 bps, t>12
Survives Bull Market (Long)PASSPASS20.5 / 29.9 bps
Market NeutralityPASSPASSNear-zero MKT beta for Long
Size NeutralityPASSPASSSMB loading insignificant in all Q5 specs
Idiosyncratic ComponentPASSPASSR² = 0.22–0.45 — substantial unexplained variation
Market Traction

Institutional Demand at Scale.

63 trials in pipeline across the world's most sophisticated institutional investors, built organically in 5 months.

11
Top-10 Multi-Strategy Platforms
Global multi-strat hedge funds with 50–300+ independent trading teams and $10B–$65B AUM. Centralized data teams evaluating 40+ alt data sets annually.
7
Sovereign Wealth Funds
The world's largest institutional capital allocators with $50B–1.5T under management. Internal quantitative equity teams seeking filing-based intelligence.
45+
Systematic Managers & Partners
Fully systematic investment firms, exchange partners, and data infrastructure platforms. Includes firms trialing the signal and distribution partners exploring licensing.
63
Institutional Trials
$10T+
Combined AUM
The Team

Built by Practitioners.

Sid Ghatak
Founder & CEO

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.

Our Philosophy

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.

Get Started

See What 6,000 Filings Tell You
About Your Portfolio.

Request a trial of any product. We provide the data, the historical archive, and the methodology documentation. You run your own evaluation.

Email Us →

sales@increasealpha.com — We respond within 24 hours.