FRM Part I · Quantitative Analysis · Free Lesson

Time Series, Stationary and Non-Stationary Processes

Free GARP FRM Part I lesson in Quantitative Analysis. 19 min read, ~2,834 words.

You backtest a momentum signal on equity prices over five years. The strategy looks profitable. You run the regression on returns instead of prices and the edge vanishes. The difference is stationarity. And the exam tests whether you can spot the trap before your trade does.

Time series modeling assumes the data-generating process is stable across the sample. If the mean or variance shifts, your regression on past data does not predict future data. Worse, two unrelated non-stationary series can look highly correlated by accident, the spurious-regression problem.

A series is covariance stationary when three conditions hold. First, the unconditional mean is constant. Second, the unconditional variance is constant and finite. Third, the autocovariance between Y at time t and Y at time t-k depends only on k, not on t. Trends violate the first; random walks violate the second; structural breaks violate any of the three.

KEY: Stationarity is a property of the data-generating process, not of the data alone. A short sample from a non-stationary process can look stationary; a long sample from a stationary...

Read the full lesson, free →
Worked examples and practice. Free with a free account, no card.

Common mistakes

Bottom line

Exam shortcut

When a question gives you φ in an AR(1) and asks for the long-run mean, divide alpha by one minus phi. When it gives you ACF cuts off at q and PACF decays, the answer is MA(q). When it gives you a series with R-squared of 0.7 between two unrelated random walks, the trap answer is "strongly correlated". The right answer is spurious regression. ADF first, then model.

The full lesson (about 2,834 words, 19 min read) adds 2 worked examples, all 6 common mistakes, a self-check, free in the app.

Learning objectives

Browse all free FRM Part I lessons or jump into free FRM Part I practice questions.