Policy Effectiveness
Insufficient statistics?

The concept of Granger-causality is commonly used in studies of causal relationships between time-series variables such as money and national income. If a forecasting equation for a variable, y, which already makes optimal use of past values of y itself, can be improved by including past values of another variable, x, then x is said to 'Granger-cause' y. Does this mean that x can be used to control y? In Discussion Paper No. 126, Research Fellow Willem Buiter argues that Granger-causality tests, which in principle are a useful 'model-free' statistical techniques, have been abused to make invalid inferences about policy effectiveness.

In previous work, Buiter has demonstrated that for x to Granger- cause y is neither necessary nor sufficient for inferring that x could be used to control y. In a rational expectations model, for example, the endogenous variables may depend on current, past or expected future values of policy instruments. If the policy instruments in turn depend on current or lagged values of the endogenous variables (through the policy rule), then the policy instruments do not Granger-cause the endogenous variables, even though changing the policy rule may alter the relationship between the endogenous variables and their own past values and the exogenous variables in the model. Buiter established that a policy instrument x might fail to Granger-cause a policy target y (while y did, in some cases, Granger-cause x) if x were set by a variety of policy feedback rules. Yet in all the examples considered by Buiter, x was an effective policy instrument.

In the present paper Buiter argues that this result is still valid in more general models. He considers cases in which governments may use a different information set to determine the values of instruments from that used by the public to form expectations. He also relaxes the assumption that policy-makers have a perfectly specified or estimated model of the economy with which to design policies.

Buiter concludes that Granger-causality tests are not useful or relevant for establishing policy effectiveness. Statistical techniques alone cannot provide insights about the ability of one set of variables to influence or control the behaviour of another set of variables.


Granger Causality and Policy Ineffectiveness: A Rejoinder
Willem H Buiter

Discussion Paper No. 126, September 1986 (IM)