Strategies and backtests

AI day trading: what AI actually does, and what the research says

AI can scan tickers, read a chart and send an alert faster than you can, on markets that trade through the night and the weekend. Here is what that actually buys a day trader, what decades of research on human day traders found, and the FINRA margin rule that changed in 2026.

[

November 16, 2026

]
•
[

14

 

min to read

]

The short answer

AI day trading means using AI to scan markets, read charts, send alerts and log trades within a single day, on crypto and stock perpetual futures that trade around the clock. Research on human day traders finds most finish behind once trading costs are counted. AI speeds up the watching, not the odds, so test every rule first.

What is AI day trading?

Day trading means opening and closing a position within the same day, sometimes within minutes of each other. AI day trading means handing part of that work, inside a single session, to software: a scan that flags a setup before the open, a chart read mid-session, an alert the moment a level breaks, or a review of the day’s fills once it is over. It is a faster-moving case of AI trading in general, built from the same kinds of tools: a model that scores a setup, a chatbot you ask mid-session, or an agent that watches continuously and can act within limits you set.

What makes a day trader’s day different from a position trader’s month is the clock. A scan, a chart read or an alert only matters if it arrives before the setup is gone, and on crypto and the stock-linked perpetual futures that now trade alongside it, there is often no closing bell to wait out until tomorrow. That is where this guide starts, and it ends with the single oldest, best-documented finding about day trading: most people who do it for long enough lose money, AI or no AI.

What does AI actually do during a trading day?

Four jobs repeat through a session, and AI now touches all of them, for better and for worse.

What AI does across a day trader’s session
WhenThe jobWhat AI doesThe catch
Before the openScanningScreens thousands of tickers for a setup: a gap, a volume spike, a pattern a model was trained to spotA scan is a filter, not a forecast. It narrows the list; it does not say the trade will work.
During the sessionReading the chartReads several timeframes at once, flags a level, and can answer “why is this moving” with a web search and sourcesA chart read can be as confidently wrong as it is right. Check every price it quotes against the chart itself.
The moment a rule is metAlerting and executingSends an alert, drafts an order for your approval, or places it inside limits set in advanceWho presses the button, and what happens while you are not watching, is a setting you choose, not something to assume.
After the closeThe journalReads the day’s fills and looks for patterns: which setups worked, which did not, what changed under pressureA handful of trades is not a sample. Treat an early read-out as a question, not a verdict.

Our guide to AI stock trading covers the tools for each of these jobs, screeners, chart readers and journaling prompts included, in more depth. Day trading just compresses the same four jobs into hours instead of weeks, which is exactly why a scan or an alert that is a few minutes late is as good as no scan at all.

Why does AI day trading never really stop?

A trader who only trades listed stocks still gets a break: the exchange closes, and the next setup has to wait for the opening bell. That break is smaller than it used to be. Crypto trades every hour of every day, and trade.xyz, an exchange built on Hyperliquid, now lists perpetual futures, contracts that track a price with no expiry date, on stocks such as NVDA and TSLA that keep a price around the clock too. In trade.xyz’s own words, “the key unlock for equity perpetuals is extending the oracle to operate 24/7, beyond external trading sessions of the underlying asset.” When no outside market is pricing the stock, as from Friday evening to Sunday evening New York time, the perp’s reference price moves by a continuous, capped adjustment that follows trading on trade.xyz itself, so it can still change on a Saturday night.

That is not a rounding error. We checked a year of daily prices for two of trade.xyz’s single-stock perps, from Hyperliquid’s public API.

Weekend price moves on two trade.xyz stock perps, year to November 14, 2026, from Hyperliquid’s public API
xyz:TSLAxyz:NVDA
Saturdays and Sundays with a daily candle, year to Nov 14, 2026105105
Of those, moved more than 1% from the day before1314
Biggest weekend move+4.1% (October 18, 2026, a Sunday)−4.2% (November 8, 2026, a Sunday)

On October 18, 2026, a Sunday when the Nasdaq was fully shut, xyz:TSLA moved +4.1%. An AI tool that only checks in on weekday mornings would have slept through it; one that watches continuously, the way an automation does, would not have. It is also a reminder that the old pattern day trader rule, covered below, counted trades in business days, a unit that means little on a market that never closes.

What does the research say about day traders’ results?

The evidence on day trading predates AI by decades, and it is unusually direct, because in a few countries researchers have studied every trade in every account, not just the ones winners choose to post.

Taiwan: a 15-year, whole-market study

Brad Barber, Yi-Tsung Lee, Yu-Jane Liu and Terrance Odean followed every day trader on the Taiwan Stock Exchange from 1992 to 2006 (Journal of Financial Markets, 2014). In a typical year, roughly 450,000 people day traded. About 20% of them made money once trading costs were counted. Fewer than 1% of them, something like 4,000 traders, beat the market by a margin wide and steady enough to call it skill rather than luck, and kept doing it year after year.

Brazil: “virtually impossible … for a living”

Fernando Chague, Rodrigo De-Losso and Bruno Giovannetti tracked everyone who began day trading Brazil’s equity futures market, the world’s third largest by volume, between 2013 and 2015, and who stuck with it for more than 300 days (working paper, 2019). Their own summary is blunt: “it is virtually impossible for individuals to day trade for a living, contrary to what course providers claim.” 97% of the traders they followed lost money. Only 1.1% earned more than Brazil’s minimum wage from it, and only 0.5% earned more than a bank teller’s starting salary, “all with great risk.”

What about ChatGPT, Claude or Grok?

Plugging a language model into the same job does not change the odds by itself. In Season 1 of nof1’s Alpha Arena, from October 17 to November 3, 2025, six AI models got $10,000 each to trade crypto perpetual futures on Hyperliquid, with the same prompt and no human stepping in. Only Qwen3-Max (+22.0%) and DeepSeek V3.1 (+4.5%) finished up; the other four lost between 31.1% and 62.7% (nof1’s own account totals at the close). The public tests of single models, from academic benchmarks to live contests, are covered in can ChatGPT trade for you? and can Grok trade for you?; neither found a model that reliably beats the market. Our wider comparison of trading LLMs lines up Claude, ChatGPT, Grok and Gemini on the same tasks.

What do costs and speed do to an AI day trader’s edge?

Day trading multiplies two things that eat into any edge: how often you trade, and how fast you have to be.

Costs compound with frequency

Every fill pays a spread or a fee, and a market order can fill worse than the price you last saw, which is slippage. A position trader who makes ten trades a year barely notices this. A day trader who makes ten trades a day pays it about 2,500 times over a year of trading days, 250 times as often, so a small edge per trade has to clear the same cost that many more times. Robert Novy-Marx and Mihail Velikov measured this across a large number of stock-market anomalies, return patterns found in published research: most of those that turned over less than half the portfolio a month kept a statistically significant edge once trading costs were counted, and few of the faster ones did. Our guide to quantitative trading goes through the statistics on turnover and costs, with a worked example on crypto perps.

Speed is not what people think it is

“Fast” for an AI agent does not mean fast the way a high-frequency trading firm is fast. The SEC’s own description of high-frequency trading includes servers placed next to the exchange’s matching engine and positions held for very short periods; a language-model agent, by contrast, reads a request, calls a few tools and reasons through several steps before it answers, which takes seconds, not microseconds. That is fast enough to watch a market tirelessly and far too slow to compete with firms whose whole business is speed. What an AI agent can realistically offer a day trader is coverage and discipline: watching more tickers and more hours than a person can, and running the same rule the same way at 3am as at 3pm, not winning a race to the exchange.

What is the pattern day trader rule, and has it changed?

For nearly 25 years, US brokerage firms ran day trading in margin accounts under one well-known rule, part of FINRA Rule 4210. A “pattern day trader” was any customer who made four or more day trades within five business days, and that customer had to keep at least $25,000 of equity in the account; below that, no more day trading until the equity was back (SEC order, April 14, 2026).

That rule has now been replaced. FINRA’s own page puts it plainly: “FINRA is replacing current day trading margin provisions, including the requirements for ‘pattern day traders’ (the PDT rules), with new intraday margin requirements.” The SEC approved the change on April 14, 2026, and FINRA’s Regulatory Notice 26-10 made it effective on June 4, 2026, with a transition period through October 20, 2027, for brokerage firms that need more time. FINRA’s investor page sums up the difference in two lines: “There’s no $25,000 minimum equity requirement for day trading.” “There’s no ‘pattern day trader’ designation based on counting trades.” Margin itself stays: “$2,000 is the minimum equity required to engage in leveraged trading (trading on margin).” If your positions during the day need more equity than the account holds, the shortfall is an “intraday margin deficit,” which you are expected to satisfy “as promptly as possible”; fail to do that repeatedly, and the account may be restricted for up to 90 days. Because firms can phase the change in until October 2027, check which rules your own broker applies today.

FINRA says the new requirements reflect technology that lets firms monitor risk in real time, and that they will typically cover all the margin you use during the day, including for zero-day-to-expiration (0DTE) options, contracts traded on the day they expire.

How do you start AI day trading without losing money fast?

  1. Pick one setup, not a feed of ideas.

    A single, narrow pattern you can describe in a sentence: a gap and go, a level reclaim, an RSI extreme. A feed of AI-flagged ideas is not a plan if you cannot say in advance what makes you take or skip one.

  2. Write the exit before the entry.

    The stop, the target, and the point in the day where you walk away regardless of what the screen says. A day trader’s worst days usually come from one bad trade followed by three more trying to fix it.

  3. Backtest and paper trade the exact rule.

    Count every trade, win and lose, over real history, with trading costs included, before it runs with money. Algorithmic trading for beginners and paper trading crypto cover the two steps in order.

  4. Size for the number of trades you will make, not the one you want to win.

    A trader who is wrong 60% of the time can still do fine with a good stop and a bigger average win; one who is right 60% of the time can still lose with a bad one. Know which trader you are from the backtest, not the hope.

  5. Decide who is accountable for the bad days.

    An alert you act on is yours. An order an agent places on its own is still yours too, which is worth remembering before you turn approvals off.

How does Ethy help with AI day trading?

We build Ethy, so weigh this section knowing that. Ethy trades Hyperliquid perpetual futures, crypto plus stocks, indices, commodities and currencies through trade.xyz, and spot swaps on Base. FINRA’s margin rules cover margin accounts at brokerage firms; Ethy’s trades run on Hyperliquid and on Base, not through a brokerage margin account. If you also day trade in a margin account at a broker, that account follows your broker’s rules.

  • It reads across the session. A chart read on four timeframes together (1D, 4H, 1H and 15m), support and resistance, RSI and MACD divergences, and a web search with sources when you want to know why a market just moved.
  • It watches around the clock. Price alerts (a move of a set size over 1, 6 or 24 hours, or a level) and indicator alerts (an RSI level, a 9/21 EMA cross, a MACD cross), delivered in the app or on Telegram, day or night, weekday or weekend.
  • It turns a setup into a tested rule. Describe it in plain words, and Ethy backtests the exact version on up to 365 days of history before anything runs, then can trade it live or on paper.
  • Its limits sit in the platform, not in a promise. Stop loss and take profit are native Hyperliquid orders that fire even when Ethy is closed. Automations add a maximum daily loss and a maximum drawdown in dollars, trading-hours windows for when you only want it running, and a pause after 10 losing trades in a row, on by default.

Your trading plan, the rules you set for risk, markets, process and goals, is read at the start of every chat, and Ethy says so when a fast-moving idea goes against it. Ethy is software for research and analysis; it does not give investment advice.

Summary

  • AI day trading means using AI for the four jobs inside a session: scanning before the open, reading the chart and alerting during it, executing at the trigger, and reviewing the fills after the close.
  • Crypto never closes, and trade.xyz’s stock perpetual futures now price around the clock too; in the year to November 14, 2026, xyz:TSLA moved +4.1% on a single Sunday while the Nasdaq was shut.
  • The best evidence on human day traders is blunt: in Taiwan (1992 to 2006) fewer than 1% of day traders beat the market consistently, and in Brazil (2013 to 2015) 97% of those who kept at it for 300-plus days lost money.
  • AI models fare no better unsupervised: most of the six models in nof1’s 2025 Alpha Arena lost money trading crypto live with the same instructions.
  • Trading costs multiply with how often you trade, and an AI agent is fast enough to watch continuously but far slower than the high-frequency firms that actually compete on speed.
  • FINRA replaced its $25,000 pattern day trader rule with intraday margin requirements, effective June 4, 2026, with a transition through October 20, 2027, for brokerage firms that need more time; it changes which account gets restricted, not the odds of making money.
  • Start with one written setup, test it with costs included before any money follows it, and decide your exits before you look for an entry.

Questions people ask

Does AI make day trading easier?

No. AI can scan more tickers, read a chart faster and watch continuously, but it cannot change the odds a day trader faces. Research on human day traders consistently finds that most lose money once trading costs are counted, and Alpha Arena, a 2025 contest that let AI models trade crypto live, ended with most of them down too. Treat AI as coverage and discipline, not an edge.

What is the pattern day trader rule, and is it still $25,000?

Not under FINRA’s new rules. The pattern day trader rule required $25,000 of equity from anyone who made four or more day trades in five business days. FINRA replaced it with intraday margin requirements, effective June 4, 2026, with a transition through October 20, 2027, for brokerage firms that need more time, so check which rules your broker applies today. $2,000 is still the minimum equity for trading on margin. The change affects how a brokerage restricts an account, not how likely day trading is to make you money.

Can crypto and stock perpetual futures be day traded 24/7?

Yes. Crypto markets never close, and stock-linked perpetual futures on exchanges such as trade.xyz keep pricing around the clock, weekends included, using a reference price that keeps moving even when the underlying stock market is shut. In the year to November 14, 2026, xyz:TSLA’s biggest weekend day was +4.1% and xyz:NVDA’s −4.2%, so “overnight” is not automatically quiet.

Can ChatGPT, Claude or Grok day trade for you?

Only through a broker or exchange connection, and nothing in the public tests of these models suggests they have a reliable edge within a day. In nof1’s Alpha Arena, six AI models traded crypto live with the same instructions in late 2025, and most lost money. Use a language model to read a chart or draft a rule, and test that rule before any money follows it.

Does Ethy day trade for me?

Ethy can read a chart on four timeframes, send alerts day and night, backtest a setup on real history, and run a tested rule live or on paper on Hyperliquid, including stock, index and commodity perpetual futures through trade.xyz. FINRA’s margin rules cover margin accounts at brokerage firms; Ethy’s trades run on Hyperliquid and on Base, not through a brokerage margin account. It is software for research and analysis and does not give investment advice.

Sources

  1. FINRA: Understanding the New Intraday Margin Requirements (April 20, 2026)finra.org
  2. FINRA Regulatory Notice 26-10: FINRA Adopts New Intraday Margin Standards to Replace the Day Trading Margin Requirements (April 20, 2026)finra.org
  3. SEC order approving FINRA’s Rule 4210 amendment, Release No. 34-105226 (April 14, 2026; Federal Register, via govinfo.gov, April 17, 2026)govinfo.gov
  4. FINRA’s proposed rule change, SR-FINRA-2025-017 (Federal Register, via govinfo.gov, January 14, 2026)govinfo.gov
  5. CFTC: Customer advisory on AI trading botscftc.gov
  6. trade.xyz Docs: Oracle Price (24/7 equity perpetuals)docs.trade.xyz
  7. trade.xyz Docs: Equities, Internal Session Hours (August 13, 2026)docs.trade.xyz
  8. Barber, Lee, Liu and Odean: The Cross-Section of Speculator Skill, Evidence from Day Trading, Journal of Financial Markets (2014)doi.org
  9. Chague, De-Losso and Giovannetti: Day Trading for a Living? (working paper, 2019)doi.org
  10. Novy-Marx and Velikov: A Taxonomy of Anomalies and Their Trading Costs, NBER Working Paper 20721 (2014); Review of Financial Studies (2016)nber.org
  11. Hyperliquid public API, candleSnapshot (daily candles taken November 14, 2026)api.hyperliquid.xyz
  12. Nof1: Alpha Arena Season 1 account totals at the close (Wayback Machine copy of November 4, 2025)web.archive.org
  13. Nof1: Exploring the Limits of Large Language Models as Quant Traders (October 27, 2025; Wayback Machine copy of November 12, 2025)web.archive.org
  14. SEC staff: Staff Report on Algorithmic Trading in U.S. Capital Markets (August 5, 2020)sec.gov
  15. SEC: Day Trading: Your Dollars at Risksec.gov

Trading involves risk, including the loss of your money. Ethy AI is a research and trading tool and does not give investment advice or recommendations. Backtests are simulations on historical data and do not predict future results.

Ethy blog

Read more

The $ETHY ecosystem

Stop watching charts.
Start living.

Deploy your AI trading agent in one click.