Testing for cointegration with structural changes in very small sample

This article proposes an adaptation of existing tests of cointegration with endogenous structural changes to very small sample sizes. Size-corrected critical values for both testing cointegration with endogenous structural breaks and testing structural breaks in the parameters in a cointegration model are computed in this context. We show that the power of such a testing procedure is satisfying in sample sizes smaller than fifty observations. This is of interest for macroeconometric studies of emerging economies for which the data history is usually not long enough to apply conventional methods. When the serial correlation is low, we find the tests to be powerful for even less than thirty observations. A combined procedure of testing for cointegration and structural change allows us to improve the power of testing cointegration in very small sample sizes while staying agnostic about the underlying data generating processes. An example using the Chinese data finds a cointegration relationship with two structural breaks between the national household consumption expenditures, the retail sales of consumer goods, and the investment in fixed assets during the last four decades.