Research On Testing Methods In Panel Data Models

Testing in panel data models becomes one of the most important contents of modern econometric theory and method, which has remarkable significance in academic and application. This paper is devoted to making a systematic research on theory and method of panel heteroskedasticity and serial dependence tests, panel unit root tests and panel cointegration tests, and to perfect and expand the tests in panel data models.In summary, the innovation research conclusion is as follows:(1) This paper proposes the certainty tests of heteroskedasticity for one-way fix effect and random effect panel model. The difficulties of empirical research come from the diversity of heteroscedasticity in panel regression model. This paper investigates the seven types of heteroskedasticity for one-way panel regression model in a test framework. The error term is decomposed into the individual effects and time effects, and then the methods and procedures of the certainty tests of heteroskedasticity are given, simultaneously improve the testing efficiency. An empirical analysis is conducted to confirm the usefulness of the heteroskedasticity certainty test, with total consumption, housing consumption and income data of urban residents in China.(2) Spurious unit root tests of panel data with structural change is investigated in this paper through Monte Carlo simulation. The result shows that unit root tests in panel data with expectation shift are effective only when the sample before and after the turning point varies greatly, or expectation shift is not obvious, or the sample N orT is relatively large; and unit root tests in panel data with trend shift are uneffective and will lead to spurious result in most cases. They show that neglect of structural breaks will lead to test failure.(3) This paper proposes a LSTR-IPS panel data unit root test with structural change, by modifying the traditional IPS unit test with nonlinear smooth transition function. The simulation results show that LSTR-IPS test has much more test power than the traditional unit root tests, and furthermore, the bigger sample size contributes a large increase of the test power for the LSTR-IPS test, especially with the increase of time horizon T, the test power of LSTR-IPS test increase obviously. Finally, as an example, we conduct unit root test to the regional GDP growth using LSTR-IPS test, with structural change rather than a unit root process which is previously being concluded, and thus, it is indicated that the LSTR-IPS test has more accuracy and power for unit root test in panel data.(4) Taking into account the progressive nature and smooth asymptotic characteristics of structural change, this paper proposes a cointegration test using LSTR model. The tests are enough to allow for cross-dependence and structural break in both intercept and slope of the cointegrated regression, which may be located at different dates different units. Two statistics by using the LM based tests are developed, and their limiting distributions are derived, and are found to be normal and free of structural break point and cross-dependence. A simulation study shows that these tests have the better small-sample properties.