Breaking

20-Year Study: LLM Trading Strategies Fail to Beat Buy-and-Hold

June 25, 2026 at 16:51 EDT

Research & Papers
Finance
AI Agents

A research team including the University of Edinburgh found that stock-trading strategies powered by large language models (LLMs) failed to outperform simple buy-and-hold investing when tested over more than 20 years and over 100 stocks. The paper, "Can LLM-based Financial Investing Strategies Outperform the Market in Long Run?" (arXiv:2505.07078), was submitted in May 2025 and revised in February 2026, with authors Weixian Waylon Li, Hyeonjun Kim, Mihai Cucuringu and Tiejun Ma.

June 25, 2026 · Market Research

Over 20 Years, AI Trading Failed to Beat Buy-and-Hold

A reported two-decade study finds that AI-driven trading strategies did not outperform simple passive long-term holding — echoing a long line of evidence that active management rarely beats the market.

20 yrs

Study period examined (exact start/end years unconfirmed)

Net edge of AI strategies over passive holding

Key weaknesses flagged: cost, overfitting, regime shifts

Backtest vs. reality

Where the AI edge disappears

Backtest

Strong on historical data

→

Live trading

Eroded by fees & slippage

Promising test performance shrinks once real-world costs and changing market conditions are applied.

How AI strategies are judged

Benchmark

Buy-and-hold passive long-term investing

Metrics

Returns plus risk-adjusted measures (Sharpe ratio)

Weaknesses

Transaction costs, overfitting, market-regime fragility

Aligns with

SPIVA-style evidence: active rarely beats passive

The optimistic case

Newer-generation models could change the outcome and adapt better to shifting market conditions.

The practitioner caution

Strong backtests routinely lose to fees and slippage; models tuned too tightly to the past break in unfamiliar conditions.

Unconfirmed: The study's title, authors, exact years, models, assets and return gap remain unverified pending further corroboration.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.