VAR
Vector Autoregression is a statistical model used to capture the linear interdependencies among multiple time series.
What is VAR?
Unlike ARIMA which models a single variable, a VAR model treats all variables as endogenous (interacting dynamically). Each variable is modeled as a linear function of its own past values and the past values of all other variables in the system, allowing researchers to see how a shock to one variable ripples through the others over time.
Why VAR Matters
It is essential in macroeconomics and finance for forecasting interconnected systems and conducting impulse response analysis without imposing strict theoretical constraints on which variables cause which.
Example
A central bank economist uses a VAR model to analyze how an unexpected increase in the interest rate affects inflation and unemployment over the next twelve quarters.
Common Mistakes
- Including too many lags or variables, which rapidly depletes degrees of freedom and causes overfitting.
- Estimating a VAR with non-stationary variables that are cointegrated, which requires a VECM instead.