Overview

A profitable equity backtest can disappear when a quarterly earnings number moves from its reporting date to its actual publication date. The investment idea may be sensible and the calculations correct, yet the simulated trader acted on information that did not exist. This topic establishes what the researcher intends to test and what the historical trader was entitled to know.

The organising object is a research specification: an economic explanation, a trading decision, an information set and a possible reason to reject the idea. A dataset then becomes an implementation of that specification. The researcher must retain firms that subsequently fail, use the version of each observation available at the decision time, and distinguish a raw execution price from a return series adjusted for distributions. These are conditions on the experiment, not cosmetic data-cleaning preferences.

By the end of this note you should be able to:

• turn a market observation into a falsifiable prediction and a trade specification;
• select historical records by their availability and revision history;
• distinguish event, observation, publication and ingestion clocks;
• reconcile splits, dividends and delistings with a holdings-and-cash ledger;
• explain which apparent profits would disappear when the data contract is enforced.

1. Contents
2. The big picture