Risk Disclosures

Backtests, models, futuretests, grades and example accounts are hypothetical. They show what a rule or a portfolio would have done on past data, under stated assumptions. No one traded them with real money, and they were made knowing how the past turned out.

Hypothetical results have limits that real results do not. They leave out the pressure of real decisions, they can be fitted too closely to the past, and markets change. Past results, real or hypothetical, do not promise future returns. Every result shows its losing years. Read those first.

Usually not. Unless a page says otherwise, results are before costs and taxes: before trading costs (commissions, spreads, slippage and market impact), financing and fees. Real results after those would be lower, sometimes much lower for rules that trade often or hold small companies.

• Entry at the next open: a backtest or model checks its rule at a day's close and buys or sells at the next session's open, never at a price it could not have had.
• Point in time: every portfolio forms from data known on its date. We never run today's holdings back through history.
• How far back: backtests reach stocks since 1990, the S&P 500 since 1928. Some early index history is price only, without dividends; each chart names what it shows. Before 1998 our stock data holds mostly companies that survived, so the early years look better than they were. A run that starts earlier says so.
• Futuretests start at the first close after an event was known and run your portfolio through each past event of that kind. The cone shows the spread of those outcomes. It is not a forecast.
• Benchmarks are named on each chart (for example, the S&P 500 with dividends, through an index or SPY, an S&P 500 fund, where the index is not available).
• Grades sum up evidence about a company from its filings and prices. A grade is not a prediction, and not a rating to buy or sell.
• Numbers come from our engine and warehouse. Each chart says when it is from (Live, or Updated with a time).

Check it. AI can be wrong: it can misread a number or make one up. Ask AI writes words around our numbers, so check any figure against the page it came from. Your AI trader acts only inside the limits you set, in the sandbox, and you stay responsible for what you approve and turn on.

Orders in the preview go only to Alpaca's broker sandbox, in accounts QuantGPT opened with play money. Sandbox fills, prices and balances can differ from real markets, so a sandbox result does not mean you would get the same result with real money.

No. Posts, comments, models and takes from members are their own views. QuantGPT does not check or endorse them.

The Data sources page, linked from the footer of every page, lists each data source with its license and credit. Here is the same list:

• Prices: EODHD end of day, via the QuantGPT engine; coins: Binance daily closes.
• Rates and macro: FRED, Federal Reserve Bank of St. Louis, through the QuantGPT engine (this product uses the FRED API but is not endorsed or certified by the Federal Reserve Bank of St. Louis); each series is its owner's, among them VIX and GVZ: Cboe; consumer sentiment: University of Michigan; Baa yields: Moody's; ICE BofA spreads: ICE Data Indices, LLC. Currencies also from the ECB (Frankfurter).
• Weekend tape: Hyperliquid perpetual futures, through the QuantGPT engine: perp prices, not exchange prices.
• Filings and companies: SEC EDGAR, Wikidata.
• Short interest: FINRA.
• Futures positioning: CFTC Commitments of Traders, legacy futures-only report (US government work).
• Bank failures: FDIC Failed Bank List and BankFind Suite API (US government work).
• Index, country and home exchange members: Wikipedia (CC BY-SA) through the MediaWiki API, Wikidata (CC0) listings.
• Search by what companies do: each company's own 10-K or 20-F on SEC EDGAR, Wikipedia short descriptions and articles (CC BY-SA) through the MediaWiki API, Wikidata descriptions (CC0); industries: GICS, via the QuantGPT engine.
• Events and news: USGS, GDACS, GDELT, WSJ, MarketWatch, Financial Times, Fox Business, Investor's Business Daily, the Federal Reserve, the SEC; per-company news, earnings dates, analyst ratings and corporate actions: EODHD, via the QuantGPT engine.
• Maps: OpenFreeMap and OpenMapTiles, OpenStreetMap contributors, NASA GIBS (Blue Marble), AWS Terrain Tiles.
• Hazards (God's Eye): Earthquakes: USGS Earthquake Hazards Program; active fires: NASA FIRMS (Fire Information for Resource Management System, LANCE MODIS and VIIRS); tropical cyclones and their tracks: NOAA National Hurricane Center; space weather: NOAA Space Weather Prediction Center (all US government work in the public domain); the operators of places matched to SEC EDGAR's list of company tickers.
• Conflicts: UCDP Candidate Events Dataset, Uppsala Conflict Data Program (CC BY 4.0; Hegre, Croicu, Eck and Hogbladh 2020), and GDELT.
• Places: Ports: NGA World Port Index; airports: OurAirports; oil chokepoints and US refineries: EIA; power plants: Global Power Plant Database, World Resources Institute (CC BY 4.0, modified); mines, refineries and oil and gas fields outside the US: USGS; chip plants and stock exchanges: Wikipedia lists (CC BY-SA), placed with Wikidata (CC0).
• Logos: Logo.dev, Wikimedia Commons, company websites, Google's favicon service; marks belong to their owners.
• Flags: circle-flags by HatScripts (MIT license).
• Advanced chart: zoom, pan and relative strength math after VeganSurge (MIT license).
• Research numbers: the QuantGPT engine and warehouse.
• Landing illustration: AI-generated.
• Open-source software and its licenses: see Open-source licenses.

Data can be late, wrong or missing. It is provided as is.

S&P 500 and Dow Jones are trademarks of S&P Dow Jones Indices LLC or its affiliates; Nasdaq-100 of Nasdaq, Inc.; Russell 2000 of FTSE Russell. GICS is a trademark of MSCI and S&P Dow Jones Indices. QuantGPT is not affiliated with or endorsed by them.

These disclosures are made by QuantGPT LLC, a Florida limited liability company, 15350 N Florida Ave, Tampa, FL 33613. Write to desk@quantgpt.co.