Bring theory
to the market.

Policies, participants, and economic conditions evolve. A relationship that appears useful in one period may weaken or reverse in another. Adaptation and stability can pull in different directions.

Many stocks respond to the same event. Neighbouring observations may also share information. A large number of records does not automatically provide an equally large number of independent observations.

Average return leaves out part of the story. A few extreme days can substantially change a result, and recovering from a loss can take time. Two portfolios with similar final wealth may have very different journeys.

The size of price moves, position sizes, and holding periods all affect returns. A model can predict many small moves correctly and still lose money on a few large mistakes.

Changing a portfolio involves trading. Fees, bid–ask spreads, and slippage can reduce a paper advantage. A simulation captures only the execution effects included in its rules.

A strategy selected from many alternatives may fit one historical period unusually well. Future observations, revised data, and a language model's memory of historical events can also make a result look more predictive than it is.