The Basics
- Simple definition: Correlation between a variable and its own past values.
- Core idea: Today’s value is related to yesterday’s value.
- Think of it as: Economic memory – what happened before influences what happens now.
What It Actually Means
In time series data, autocorrelation (serial correlation) means errors are not independent across time. If inflation is high today, it tends to be high tomorrow – observations aren’t independent. This violates regression assumptions, making standard errors wrong and tests unreliable. Positive autocorrelation (common in economics) means high values follow high values; negative autocorrelation means low values follow high values.
Example
Pakistan’s monthly inflation shows autocorrelation – high inflation this month often means high inflation next month, because momentum and expectations persist. Forecasting models must account for this.
Why It Matters
Ignoring autocorrelation leads to overconfident, wrong conclusions. Economists test for it (Durbin-Watson, Breusch-Godfrey) and correct using methods like Newey-West standard errors or including lagged variables.
See also
Time Series Analysis • Heteroscedasticity • Stationarity • Durbin-Watson Statistic • ARIMA Models
Read more about this with MASEconomics:
Understanding and Dealing with Autocorrelation in Time Series Econometrics