Panel, Cross-Section or Event Stream?

Before running any regression or backtest, it helps to name the shape of the data you actually have, because the shape determines which statistical tools are even valid. A cross-section is one snapshot — every stock's valuation ratio on a single day — with no time dimension at all. A panel tracks the same set of entities repeatedly over time — every stock's valuation ratio every month for ten years — giving you both a cross-sectional dimension and a time dimension, with the same tickers reappearing each period. An event stream is neither: it's an irregular sequence of things happening whenever they happen — trades, order cancellations, earnings announcements — with no fixed grid of entities or dates to line up against.

The distinction matters because standard errors and identification strategies assume a particular shape. A panel lets you use fixed effects to strip out anything constant about a company (like its industry) and isolate what changes over time; applying that same trick to a single cross-section is meaningless because there's no "over time" to speak of. An event stream can't be cleanly averaged into a panel without first deciding how to bucket events into time windows, and that bucketing choice — hourly, daily, per-trade — can change the answer a study gets, so it deserves being stated explicitly rather than assumed.

Misreading the shape is a common source of a subtly wrong analysis: treating panel data as if it were a plain cross-section (ignoring that the same company appears many times) understates how correlated your errors really are, because one bad quarter for a company shows up in every one of its rows.