The objective of the current research project is to establish a new econometric estimation and inference methodology that extends cointegration analysis to structural systems with time series exhibiting general persistence characteristics.
Existing econometric methodology of inference in cointegrating regression is largely confined to models with exact unit roots, since even a small deviation from a unit root data generating mechanism invalidates inference and may result to severe size distortions in econometric testing.
The project develops a new general approach to inference that resolves this difficulty and is robust to the persistence characteristics of the regressors. A feasible instrumental variables procedure is developed in which instrument persistence is explicitly controlled, independently of the unknown degree of persistence in the original regression. A unified inference framework for cointegrating regression is established that cab to accommodate various diverse classes of persistent processes.
Financial and macroeconomic applications of the new methodology are considered in the form of robust hypothesis tests for stock return predictability and residual based tests for cointegration. The new methodology should become a useful tool for applied research, as it helps alleviate practical concerns about the use of cointegration techniques when there is uncertainty about the type of regressor persistence.