For institutions
Welcome to ATLAS for Institutions.
Our performances, model calibrations, metrics and services.
All figures as of 13 September 2026. Methodology and definitions are one click away, never in the way.
Coverage
Every market, one engine.
From New York to London, Frankfurt, Paris, Zurich, Hong Kong, Shanghai, Seoul, Mumbai, Riyadh, São Paulo and Mexico City: prices, full history, company data and simulations on every exchange.
Calibration
For every stock, ATLAS gives a low and a high price for 30 days ahead. 8 times out of 10, the real price ends up between them.
Checked on 903,345 forecasts across 5,043 stocks, 2 Jan → 14 Oct 2026: 83.2%.
| How sure | Should land inside | Did land inside | Gap |
|---|---|---|---|
| 50% | 5 in 10 | 58.0% | +8.0 |
| 80% | 8 in 10 | 83.2% | +3.2 |
| 90% | 9 in 10 | 90.4% | +0.4 |
On a 30-day horizon. Recomputed nightly; figures as of 13 Sep 2026.
How coverage is measured
Each range comes from a return distribution: 10,000+ Monte Carlo paths per stock and horizon, with volatility recalibrated nightly to the current regime. The 50%, 80% and 90% ranges are the corresponding central intervals of that distribution; Value at Risk and Expected Shortfall at 95% are read from the same paths, so a calibrated range means a calibrated risk figure.
On each trading day ATLAS publishes, for every tracked stock, the 30-day range at all three levels. Thirty days later the realized close is compared with the range ATLAS gave. A prediction counts as covered if the close lands inside the range.
Actual coverage is the share of covered predictions at each level over the whole window. The gap is actual minus stated: a positive gap means the range was wider than it needed to be, a negative gap means it was too narrow. Every prediction is scored, none are removed, and the full log is public on the track-record page.
Backtesting · Egyptian Exchange
A market backtest, verifying our model.
Several portfolios across different years, consistently beating the country's index, surviving market crashes coupled to high volatility. Results are after simulated trading costs (spreads and an assumed broker fee of 2.5% on each buy and each sell) and include dividends and the exchange rate.
All three windows side by side
| Window | Bought → sold | Sessions | Eligible | Return EGP | Return USD | EGX30 | Ahead by | Worst fall | Winners | Forecast → actual |
|---|
The ten shares in the selected window
| # | Ticker | Company | Sector | Size | Weight | Bought EGP | Sold EGP | Dividends EGP | Price change |
|---|
Weights as allocated on the decision date. Dividends were reinvested in the same share. Price change is sale price over purchase price, before dividends and costs.
The other 74 portfolios ATLAS generated the same day
Every size from one to twenty-five shares, each under three ways of splitting the capital: equally, by conviction, by steadiness. All bought and sold on the same dates. The highlighted bar is the ten-share portfolio above.
Methodology, costs and what was left out
- Decision, then execution. The ranking is run after the close on the decision date and every share is bought at the next session's price. Nothing is traded until the sale date.
- Universe. Every share on the Egyptian Exchange above a daily traded-value floor on the decision date (USD 25,000 in 2020, USD 250,000 in 2023, USD 500,000 in 2025). Delistings and suspensions are kept until they happen.
- Costs. An assumed broker fee of 2.5% on each buy and each sell, the highest assumption we apply. It is the cost an investor would pay a broker; ATLAS charges no commission and does not execute trades. Spreads are the Egyptian Exchange's monthly figures published through the World Federation of Exchanges, by size segment, charged half on purchase and half on sale.
- Dividends and currency. Cash dividends are reinvested in the paying share. Dollar returns use the official USDEGP closing rate on the buy and sell dates.
- Not included. Taxes, intraday execution, market impact beyond the published spread.
- Reproducibility. Every calculated cell in the source workbooks is a live formula. The workbooks and the EGX disclosure documents they draw on are available on request.
Public fund performance
A live fund running on the US market.
On the ATLAS platform, run on virtual capital with every trade timestamped and visible to anyone.
- Fund
- ATLAS Strategies
- Started
- 9 Apr 2026
- Capital
- $10,000 virtual
- Benchmark
- S&P 500 (SPY)
- Status
- Active · indefinite
Figures captured from the fund page on 13 Sep 2026, market closed. Curve traced from the fund page chart and pinned to the published start, peak and end values; the exact daily series is available as CSV on the fund page. Virtual capital: the fund tracks real prices and real execution rules, no client money is invested. Past performance is not indicative of future results.
Coverage & data
What goes in, and when.
Every stock on the NYSE, Nasdaq, AMEX and the Egyptian Exchange, rebuilt from public data after each close.
Prices
Full daily history for every stock since its IPO. Drift and volatility are re-estimated per stock from roughly 252 trading days of recent history.
- Exchanges
- NYSE · Nasdaq · AMEX · EGX
- US stocks tracked
- 5,043
- Refresh
- Nightly
Earnings & filings
Earnings calendar refreshed from regulated filings. Each stock's historical post-earnings behaviour is blended into any horizon that contains a report date.
- Source
- SEC filings
- Overlay weight
- Per stock, 0.30–0.79
- Refresh
- Nightly
Macro regime
15+ FRED indicators (VIX, NFCI, term spreads, unemployment trends, credit spreads) classify the market as Normal, Neutral, Risk-on, Bearish or Stress; each regime applies its own drift and volatility multiplier to every stock.
- Source
- FRED
- Regimes
- 5 states
- Refresh
- Nightly
News & events
Every news article and SEC 8-K for every covered stock is sentiment-scored nightly and aggregated per stock, per sector and globally. Regulatory actions, fraud allegations, going-concern warnings and delisting notices are flagged on the output.
- Sectors
- 11 GICS
- Hard-risk flags
- 4 types
- Refresh
- Nightly
Data quality, limits and what the model cannot know
- Flagged names. Stocks with short histories (recent IPOs), very low volume or pending corporate actions carry wider error bands and are flagged on the output. A newly public stock with four quarters of data gets an earnings-overlay weight near 0.30; a long, consistent record like AAPL lands near 0.79.
- Horizons. From 7 to 365 days.
- Cadence. One intraday session contributes about 0.4% to the underlying distribution, so probabilities change once a night, all at once, and are identical at 11 AM and 4 PM.
- Public information only. Events enter the model once they are public. Undisclosed fraud, tomorrow's analyst rerating and single-name binary events (FDA decisions, M&A) are not pre-warned.
- Regime breaks. Unprecedented macro shocks the engine has not seen are the stated failure mode; the FRED-driven regime reacts in the next nightly batch.
Services & delivery
Two ways to take the numbers.
Plug the engine into your own systems, or give your analysts the platform.
Services
What ATLAS does, and where each service is available.
| Service | API | Platform |
|---|---|---|
| Analyze stocksOdds up and down, price range, VaR, on any listed US and EGX stock, from 7 to 365 days. | ✓ | ✓ |
| Build portfoliosBudget, target and horizon in; weights and share counts out. Joint risk, 50% sector cap. | ✓ | ✓ |
| Track market eventsEvery position re-simulated nightly; SELL / TRIM / HOLD / ADD in the morning brief by 9 AM ET. | — | ✓ |
| Recover from lossesOdds of getting back to break-even in the time left, the expected days to do it, and the stocks with better odds of recovering the loss. | — | ✓ |
| BYOP (Bring your own portfolio) Coming soonLoad an existing portfolio into ATLAS and get the read on it: how to raise the expected return, cut the risk and lift the probability of reaching the target, with what to buy, hold or sell and how to reweight. | ◌ | ◌ |
REST API
One endpoint returns every probability, threshold and transparency field as structured JSON. All derivations are precomputed server-side; client code does no financial math.
- Endpoint
- POST /api/simulate-full
- Per request
- Up to 50 tickers
- Horizons
- From 7 to 365 days
- Cached mode
- < 100 ms
- Enriched mode
- 8–12 s cold · < 100 ms cached
- Rate limit
- 2,000/min cached · 500/min enriched
- Auth
- Bearer API key
Request, response and limits
- Two tiers. Cached (
skipNews: true) returns probability, risk profile, verdict and the full technical block in under 100 ms. Enriched adds analyst narrative, catalysts, risks and news sentiment: 8–12 s on a cold cache, sub-100 ms within the 4–5 hour cache window. - Keys. Sandbox key at 100/min, issued within 24 hours. Production keys after a use-case review: 2,000/min cached, 500/min enriched, burst allowance of 3× the steady rate over 30 seconds,
X-RateLimit-*headers on every response. Rotation keeps the old key valid for 24 hours. - Partial success. If one ticker in a batch fails, the others still return their full payload; the failed symbol appears in an
errorfield. - Transparency block. Every response exposes raw σ, the macro multiplier, adjusted σ, the drift chain, simulation parameters, the news multiplier and the earnings blend formula, so a model-risk team can reproduce the number.
Model governance
How every forecast is recorded.
Live since 1 October 2026
Since 1 October 2026 every prediction is saved before the open and never edited. January to September was rebuilt day by day with only the data available the evening before, after the previous database was lost, and is labelled as rebuilt.
Scored after, not fitted before
Accuracy is recomputed nightly on every logged prediction: what the model said before, against what the market did after.
Changes are dated and measured
Every methodology change is published with its effect on calibration; the April 2026 earnings refinement moved it +0.3 pts. Past results are never restated.
Reproducible outputs
Every API response carries its model version and every intermediate value behind the number, so a risk team can rebuild it. 10,000 paths, about 1% standard error.
Stated limits
- Ledger imperfections, as documented: survivorship bias, earnings-timing edge cases and macro regime breaks.
- Delistings. A ticker delisted mid-horizon has its result invalidated and is reported as such; in a portfolio it is removed and reallocated on the next nightly run.
- Scope. Listed equities on the NYSE, Nasdaq, AMEX and the Egyptian Exchange. No crypto. Raw simulation paths are not in the public API; per-path output is available for partner research arrangements.
- Status. Outputs are probabilistic estimates, not personalised advice and not a recommendation to buy or sell.
Contact
Get the tear sheet.
One PDF with the method, the coverage and the calibration. Your model-risk team can start the same day.