AI Trading Book Reference
Updated 27 August 2026 Primary sources only Not investment advice

AI trading, documented — not marketed

An independent reference for Australia, New Zealand and the United Kingdom. Every figure on this site carries a named primary source and a retrieval date: regulator reports, exchange rules, peer-reviewed papers and vendor pricing pages. Where sources disagree, we publish the disagreement instead of averaging it away.

AI trading is legal and regulated in all three markets — under technology-neutral rules (ASIC, FMA, FCA), not AI-specific statutes.

The dominant retail risk is leverage and fraud, not model error. 68% of Australian retail CFD investors lost money in FY2024 (ASIC REP 828, 20 Jan 2026).

Independent evidence does not show AI beating the market after costs. The Eurekahedge AI Hedge Fund Index returned 9.8% annualised (Dec 2009–Jul 2024) against 13.7% for the S&P 500.

Realistic directional accuracy is roughly 55–65% in careful peer-reviewed work. Claims of 90%+ are typically attributable to look-ahead bias.

Read the main guide Open the fact table
68%

of Australian retail CFD investors lost money in FY2024
ASIC REP 828, 20 Jan 2026

11,964

scam and phishing sites removed in 2025, up from 6,270
ASIC 26-063MR, 8 Apr 2026

2%

of UK financial-services AI use cases are fully autonomous
BoE/FCA survey, n=118, 21 Nov 2024

9.8%

AI hedge-fund index annualised vs 13.7% for the S&P 500
Eurekahedge, Dec 2009–Jul 2024

Start with the pillar

Three pages carry most of the ground: what AI trading is, whether the evidence supports it, and what it costs.

Basics · pillar

What AI trading is and how it works

Definition, the nine-step decision loop, eight functional categories with their failure modes, the semantic difference between AI, algorithmic, systematic and high-frequency trading, and a documented timeline from the 2010 Flash Crash to the IOSCO supervisory toolkit of May 2026.

Glossary · 15 terms · 14 FAQs
Evidence · pillar

Does AI trading actually work?

The published record: AI hedge-fund index underperformance, a peer-reviewed review of 27 machine-learning equity experiments finding no conclusive evidence of returns at scale, and why no audited like-for-like live-return comparison between retail vendors exists.

4 charts · 5 tables
Tools · pillar

Platform prices, quotas and limits

Nine platforms with prices taken from vendor pages: TradingView from USD 0 to 199.95, TrendSpider 52.38 to 128.40, Trade Ideas 89 and 178, Cryptohopper 0 to 129, 3Commas, Pionex fees, QuantConnect tiers and the IBKR API. Plus what the subscription price leaves out.

Comparison matrix · 5 head-to-heads

Every section, in order of depth

Evidence

How accurately can AI predict prices?55–65% directional in careful work; why 90%+ signals data leakage Backtest overfitting and the Deflated Sharpe RatioBailey & López de Prado; Harvey, Liu & Zhu on the t-statistic above 3.0 Market size: why estimates run from USD 2.5bn to 57bnSeven publishers, incompatible perimeters, and why not to average them

Regulation

United Kingdom: FCA, MAR 7A and RTS 6"We do not plan to introduce extra regulations for AI" — FCA, updated 13 Feb 2026 Australia: ASIC, REP 798, REP 828 and CP 386624 AI use cases across 23 licensees; leverage capped 30:1 to 2:1 New Zealand: FAP licensing and NZX participant rulesWhen a tool crosses from information into regulated advice

Risks and scams

AI-branded investment scams: how they workFake brands, deepfake endorsements, and the licence check that precedes everything CFD leverage and published loss rates68% Australia, approximately 80% United Kingdom — and what the numbers do not say API-key security for trading botsTrade-only keys, IP allowlists, subaccounts, and a documented breach

Practice

Build and validate a strategy: twelve stepsChronological splits, walk-forward, cost modelling, kill switch, staged capital Data hub: the full fact tableEvery figure on this site with its source URL, publication date and freshness flag

How we source and flag facts

Four independent deep-research passes were compiled for this site. Two of them contained regulatory documents that could not be verified against any regulator's own register — a fabricated FCA policy statement, a consultation paper with the wrong number, a supervisory guidance note that does not appear in the issuing authority's library. None of those claims appear on this site. The rule we apply is simple.

Published

The figure appears in a named primary source — a regulator media release with its reference number, an exchange rulebook edition, a peer-reviewed paper, or the vendor's own pricing page — and we record the retrieval date.

Flagged

The source exists but the page carries no publication date, or the underlying survey is more than twelve months old. It stays, marked stale, with the date it was captured.

Withheld

No primary source, or the claim conflicts with one. It is not published — not softened, not attributed to "industry sources". Where credible sources genuinely disagree, both figures appear side by side.

Editorial policy About this reference Corrections log

Key facts at a glance

A short extract from the full fact table. Values are reproduced as the source states them; qualifiers such as "more than" are not converted into exact numbers.

Table 1. Selected verified figures on AI and algorithmic trading, retrieved August 2026.
FactValueSourceDate
Retail CFD investors losing money, Australia FY202468%ASIC REP 82820 Jan 2026
CFD customers losing money, United Kingdomapproximately 80% (CP16/40 sample: 82%)FCA2019 onward
UK financial firms already using AI75% of 118 respondentsBank of England / FCA survey21 Nov 2024
Fully autonomous AI use cases2%Bank of England / FCA survey21 Nov 2024
Top three cloud providers, share of named providers73%Bank of England / FCA survey21 Nov 2024
Scam and phishing sites removed, calendar 202511,964 (prior 12 months: 6,270)ASIC 26-063MR8 Apr 2026
AI use cases reviewed across 23 licensees624ASIC REP 79829 Oct 2024
Retail CFD leverage caps, AU and UK30:1 major FX down to 2:1 cryptoASIC 20-254MR; FCA PS19/182019–2021
Algorithmic share of trading volume, US/EU/major Asiaapproximately 60–75%Select USA2018, restated to 2025
AI hedge-fund index vs S&P 500, annualised9.8% vs 13.7%EurekahedgeDec 2009–Jul 2024
Realistic directional prediction accuracyapproximately 55–65%Peer-reviewed studies (Yu & Yan; Ding et al.; Sirignano & Cont)2019–2025
FCA review of algorithmic trading controls10 principal trading firmsFCA multi-firm review21 Aug 2025

Short answers

Is AI trading legal in Australia, New Zealand and the UK?

Yes. None of the three regulators bans the technology or has a bespoke AI trading law; existing financial-services obligations apply to AI and non-AI systems alike. What triggers a licence is the activity — dealing, advice, portfolio management, market access — not the label on the software. Full detail by jurisdiction.

Do AI trading bots guarantee profit?

No, and no audited benchmark of live retail bot returns exists. Backtest-overfitting research shows that selecting the best result out of many trials manufactures apparent skill; the published AI hedge-fund record is below the S&P 500 over fifteen years. The evidence in full.

What does an AI trading platform cost?

Retail tools run from USD 0 to about USD 200 per month, with Trade Ideas Premium at USD 178 per month or USD 2,136 per year at the top of the compared set. Subscriptions exclude broker commissions, exchange data, spreads, slippage and tax. Price and quota tables.

What is the single biggest risk?

Fraud and leverage, in that order — not model error. Between 68% and roughly 82% of retail CFD accounts lose money, and AI-branded scam sites are being removed by the thousand each year. Verify the promoter's licence before anything else. Scam patterns and checks.