UP | HOME

granger causality

1. linear regression

  • \(y\) and \(x\) are two time sries.
  • We want to test whether \(x\) causes \(y\)
  • consider the time series \(y_t\) and the time series \(x_t\)
  • choose a lag value \(m\), and make an auto-regressive linear regression
    • \(y_t = a_0 + a_1 y_{t-1} \cdots a_{m+1} y_{t-m}\)
  • Now add lagged values of \(x\) to this regression
  • \(y_t = a_0 + a_1 y_{t-1} \cdots a_{m+1} y_{t-m} + b_0 + b_1 x_{t-1} + b_{m+1} x_{t-m}\)
  • first check that the regression is a good fit (ex. using an F-test)
  • then, check whether the coefficients \(b_i\) are significant according to their t-statistic
    • if any of them are, then reject the null hypothesis that \(x\) does not Granger-cause \(y\)

2. see also

Created: 2024-07-15 Mon 01:28