Machine-picked.Market-judged.
Before every U.S. opening bell, GreatEye ranks the 300 most liquid U.S. stocks and ETFs and sells short the five it rates lowest, then closes every position at the final bell. Its research spans 30 years and 7,636 walk-forward sessions, and every trading day the market judges its calls live.
- Walk-forward since 1995
- Survivorship-bias free
- Pre-registered test
- Reproduced bit for bit
Hypothetical walk-forward backtests on vendor reference prices, after modeled costs where stated. Definitions in the evidence and the working paper.
One decision a day. Made before the bell.
GreatEye is a supervised machine-learning system built for a single, repeatable decision: which five of the 300 most liquid U.S. securities are most likely to fall between today’s open and today’s close. The rules are fixed; the model decides.
- 4:00 PMPrevious close
Data locked
Daily records for the most liquid U.S. stocks and ETFs close at 4:00 PM. Nothing after the close enters the next morning’s decision.
- 9:15 AMBefore the open
The call
The model scores the day’s universe and selects the five names it rates lowest, with fixed rank weights. The call is broadcast live.
- 9:30 AMOpening auction
Five shorts
The five positions are sold short in the opening auction, at the official opening price of each name.
- 4:00 PMClosing auction
The verdict
Every position is bought back in the closing auction. The book is flat every night: no exceptions.
Flat every night
Positions live only from the opening to the closing auction. The book never holds a position through an overnight gap, an after-hours earnings release or a weekend.
Liquid by design
The universe is rebuilt every session from the 300 most traded U.S. securities by dollar volume, common stocks and exchange-traded products alike. A name qualifies only if it averages at least US$50 million a day.
Fully systematic
People design, test and monitor the system; the model makes the call. No discretion, no overrides, no emotion at the open.
Thirty years of evidence. A positive signal in every single year.
We evaluated GreatEye walk-forward, with every model trained only on data available at the time, across 7,636 trading sessions from December 1995 to July 2026, on a panel that keeps every delisted security.
Thirty years, compounded within capacityUS$1 million reinvested up to a US$10 million short book, after modeled costs, 1995–2026
A positive signal in all 31 yearsMean daily rank information coefficient by calendar year, 1996–2026
Losers sorted from winnersMean open-to-close return by predicted decile, full period
No random portfolio came closeSharpe ratios before costs: 6,000 placebo books against GreatEye’s, 1995–2026
Key statistics
- Sharpe ratio after modeled costsConstant US$1 million, 1995–2026 · 95% interval 1.36–2.05 1.71
- Financial-crisis era2008–2013, after modeled costs 2.14
- In bear marketsS&P 500 20% or more below its peak (1,834 sessions), after modeled costs; 1.59 in normal markets 1.98
- Profitable calendar yearsComplete years 1996–2025, after modeled costs 29of 30
- Positive monthsAfter modeled costs, 1996–2026 68.1%
- Balanced portfolio½ GreatEye, ¼ S&P 500, ¼ gold, compounded from US$25,000, after modeled costs · deepest drawdown about half that of GreatEye alone 1.90Sharpe
All figures are hypothetical walk-forward backtests on vendor reference prices, not the results of an investor account. After-cost figures include 4 bp of slippage per side and a 2026 broker fee schedule; borrowing, financing and market impact are not modeled. Past performance does not guarantee future results. Full disclosures.
Built to survive scrutiny.
Most trading backtests are flattered by three things: data that quietly encode the future, design choices tuned after seeing the results, and luck in model fitting. GreatEye’s research program was built to separate and measure each one.
Point-in-time, survivorship-free data
A panel rebuilt from unadjusted vendor records since 1992. Every delisted security stays in, keyed to a permanent identifier, and every input ends at the previous close.
Strictly walk-forward
Each model is trained only on the past and scores the sessions that follow it: 7,636 walk-forward sessions in total, each window’s predictions sealed before any of its outcomes was read.
Pre-registered, then tested
The baseline recipe was tested on 1995–2003, eight years withheld while it was designed, under rules registered before a single outcome was read. It cleared its bar by a factor of nine.
Reproduced bit for bit
Separately written code re-hashed every raw file, rebuilt all 8,644 universe sessions and refit 61 of 61 walk-forward models bit for bit.
Placebo-controlled
Random portfolios drawn from the same sessions, instrument classes, liquidity and volatility. None came within reach of GreatEye’s 2.17 before costs.
Robust to the random seed
Ten refits with different random seeds. The signal stayed positive in every calendar year, in every refit; a second data vendor covered 98.97% of the traded positions since 2004.
Read the full study.
Every number on this page is computed from the frozen results of a 34-page working paper written to the standards of a scientific journal: the data, the method, every test, the complete cost model and the statistical inference.
- The pre-registered test on eight withheld years, and the rules it had to beat
- Performance by era, in bear markets and under costs adjusted to each era
- How the data layer was verified, record by record, and reproduced bit for bit
- Placebo controls, seed robustness and the full statistical inference
Don’t trust us. Watch us.
Every U.S. trading day, GreatEye’s five calls go out live before the opening bell, and the market delivers its verdict at the close. A public record, built one session at a time.
- 9:00 AM ETPre-market
- 9:15 AM ETThe call
- 9:30 AM ETOpening auction
- 4:00 PM ETThe verdict
Questions investors ask first.
What is GreatEye?
GreatEye is BlackShift’s machine-learning system for U.S. equities. Every trading day it ranks the most liquid U.S. stocks and exchange-traded products by how likely each is to fall between the open and the close, and sells short the five it rates lowest. Every position is bought back at the close.
How do you guard against overfitting?
Each known source of backtest optimism is tested separately. The evaluation is strictly walk-forward over 7,636 sessions; the baseline recipe passed a pre-registered test on eight years withheld while it was designed; 6,000 random placebo portfolios never came close; ten refits with different random seeds kept the signal positive in every year; and separately written code reproduced the data layer and the models bit for bit.
Are these results live or backtested?
The 30-year record is a hypothetical walk-forward backtest, built to the standards above and documented in the working paper. The live record is kept separately and built in public: the calls are broadcast before the open and judged by the market at the close, every trading day.
Today’s broadcast calls come from GreatEye’s earlier production model. Its successor, fitted on the point-in-time panel evaluated in the paper under a procedure registered in advance, is being built; its calls will be recorded before every open.
Why intraday short selling?
Short sellers are among the best-informed participants in equity markets, and a book that is flat every night carries no overnight position. GreatEye’s after-cost Sharpe ratio was higher in bear markets, 1.98, than in normal markets, 1.59: it tends to work hardest when long portfolios suffer.
How are trading costs handled?
In the main results, every trade is charged 4 basis points of slippage per side plus a full 2026 broker fee schedule, including commissions, exchange auction fees and regulatory fees. Where an exchange ran no opening or closing auction, as on Nasdaq before 2004, the paper’s central era-aware variant charges at least one tick on each such trade.
How does it combine with a traditional portfolio?
In the study, a balanced portfolio holding half its capital in GreatEye and a quarter each in an S&P 500 fund and gold, compounded from US$25,000, reached an after-cost Sharpe ratio of 1.90 from 1995 to 2026, higher than GreatEye alone, with a deepest drawdown roughly half that of GreatEye alone.
How much capital can GreatEye run?
GreatEye trades only the most liquid U.S. securities: a name enters the universe only if it averages at least US$50 million of daily dollar volume. In the study, a US$1 million book never took more than 0.70% of a security’s average daily dollar volume, and at US$10 million the median position was 0.56% of it.
Scaling further, with more names per session, liquidity-aware sizing and execution beyond the opening and closing auctions, is the next stage of the research program.
Is the model disclosed?
The method, the tests and the results are documented in full in the working paper. The proprietary inputs, formulas and settings of the model are not.
Can I invest?
BlackShift does not offer securities or managed accounts through this website. We are opening conversations with qualified investors, family offices, allocators and strategic partners. Request the paper and investor materials below and we will reply personally.
Request the research paper.
Qualified investors, allocators and research partners can request the full working paper and our investor materials. Every request is read by the team.
- The working paper34 pages in English, 38 in the Portuguese edition: data, method, tests, costs and inference.
- Answers to your diligence questionsOn the research, the system and the live record.
- A conversation with the founderFor qualified investors and partners, on request.
Prefer email? investing@blackshiftcap.com
Request received.
Thank you. We read every request and will reply personally.