Autocorrelation
The correlation of a series with its own earlier values. Price levels are strongly autocorrelated (today is close to yesterday); daily price changes usually aren't.
In plain English
Autocorrelation breaks the independence that most tests assume, so a long series of levels contains far less information than its length suggests. Two trending or wandering series will often look strongly correlated with each other by chance. The usual fix is to work with changes rather than levels.
Uncle Gary explains
In a simulated price series, the correlation of each day's price with the day before was about 0.98; for daily changes it was about 0.00. 41% of pairs of unrelated random walks showed a correlation above 0.5.
Why it matters for your money
Two price charts that 'move together' prove little. Compare their returns, not their levels, before believing in a link or using one to hedge the other.
Where we tested it
- Lesson 31: Time Series: Trends, Seasons and Fake Correlations
- Does the COT Report Predict the Stock Market? We Tested 717 Weeks of S&P 500 Futures Positions
Related terms
All glossary terms. Educational content, not financial advice.