Likelihood-Based Panel Cointegration Methodology and Its Applications in Macroeconomics and Financial Market Analysis
Final Report Abstract
Cointegration defines a long-run equilibrium relationship between non-stationary variables. These non-stationary variables are called random walks, since it is not possible to predict in which direction they will evolve in time. Cointegration tests are tools, which are used to determine the existence of long-run equilibrium relationships between variables. However, it is know that the standard cointegration tests have low power if the time span is short. As a remedy to this problem two decades ago, the first panel cointegration tests were proposed. By including observations from different cross-sections (countries, regions, cities, etc.) and by extending the conventional cointegration tests to the panel data framework, it was possible to propose tests with higher power even when the time series is short. Most of the panel cointegration tests in the literature are tests, which are only able to determine the existence of a cointegrating relationship (i.e. residual-based tests). Therefore, this project concentrated on the development of “likelihood-based panel cointegration tests”, which are not only appropriate to find out the existence of a long-run relationship, but also suitable to determine the number of cointegrating relations between variables. Simulation studies have shown that neglecting the cross-sectional correlation between the panel units (i.e cross-sectional dependence) may lead to wrong statistical inference. In this project, we have proposed several likelihood-based panel cointegration tests, which are robust to various types of cross-sectional dependence. We have modelled cross-sectional dependence by; (i) using unobserved factors, which are affecting the crosssections heterogeneously, (ii) using the correlation between the innovations to model the correlation between the statistics, and (iii) combining p-values using the Bonferroni correction. In our simulation experiments, we have compared our tests with the existing residual-based panel cointegration tests in the literature. In most of the simulation settings, our tests have demonstrated good size and power properties in small samples. It is also known that if structural breaks (i.e. German reunification, Asian Crisis, Great Recession) are not taken into account, the panel cointegration tests may lead to wrong test results. Therefore, within the project we have developed the “first likelihood-based panel cointegration test” in the literature, which takes both structural breaks and cross-sectional dependence into account. The test not only allows for structural breaks and cross-sectional dependence, but also has various other advantages: It is even applicable when the number of time observations are not equal for all cross-sectional units of the panel. The break dates and the number of breaks can also vary across panel units. By using simulation techniques, we were able to show the nice finite sample propertied of our test with structural breaks. Moreover, we applied the new tests that we have developed, to various macroeconomic research questions. Here are some examples: (i) Money demand relation: By analysing selected 22 OECD countries we have concluded that to keep the money demand stable, in the long-run the economies should take some unobserved factors such as technological progress, oil prices etc. into account. (ii) Housing prices: In the long-run global stochastic trends are driving the real house prices in 12 selected OECD countries. (iii) Trade with China: Import from China has positive contributions to the gross domestic product in 22 emerging and developing countries. However, the contribution of trade with advanced and other emerging and developing countries is higher than the contribution of trade with China. (iv) Regional housing prices in the U.S.: There is not any long-run equilibrium relationship between the state level housing prices and the personal income in the US, when the recent financial crisis is taken into account.
Publications
- (2020) Intersection tests for the cointegrating rank in dependent panel data. Communications in Statistics - Simulation and Computation 49 (4)918–941
Arsova, Antonia; Örsal, Deniz Dilan Karaman
(See online at https://doi.org/10.1080/03610918.2018.1489552) - (2011) “Corrigendum to likelihood-based cointegration tests in heterogeneous panels”, Econometrics Journal, 14: 21-125
Karaman Örsal, D. D. and Droge, B.
(See online at https://doi.org/10.1111/j.1368-423X.2010.00327.x) - (2014) “Do the global stochastic trends drive the house prices in OECD countries?”, Economics Letters, 123: 9-13
Karaman Örsal, D. D.
(See online at https://doi.org/10.1016/j.econlet.2014.01.007) - (2014) “Panel cointegration testing in the presence of a time trend”, Computational Statistics and Data Analysis, 76: 377-390
Karaman Örsal, D. D. and Droge, B.
(See online at https://doi.org/10.1016/j.csda.2012.05.017) - (2017) “Analyzing money demand relation for OECD countries using common factors”, Applied Economics, 49: 6003-6013
Karaman Örsal, D. D.
(See online at https://doi.org/10.1080/00036846.2017.1371842) - (2017) “Meta-analytic cointegrating rank tests for dependent panels”, Econometrics and Statistics, 2: 62-72
Karaman Örsal, D. D. and Arsova, A.
(See online at https://doi.org/10.1016/j.ecosta.2016.10.001) - (2018) “Likelihood-based panel cointegration test in the presence of a linear time trend and cross-sectional dependence”, Econometric Reviews, 37:10, 1033-1050
Arsova, A. and Karaman Örsal, D. D.
(See online at https://doi.org/10.1080/07474938.2016.1183070)
