This dissertation includes three essays on unit root tests and cointegration.
The first essay uses econometric tools to explore the possible causes of global warming. By examining the time series properties of the global mean temperature, solar irradiance and greenhouse gases, it finds static long-run steady-state and dynamic long-run steady-state relations between temperature and radiative forcing of solar irradiance and a set of three greenhouse gas series. Estimates of the adjustment coefficients indicate that temperature series is error correcting around 5% to 65% of the disequilibria each year, depending on the type of the long-run relation. The estimates of the I(1) and I(2) trends indicate that they are driven by linear combinations of three greenhouse gases and their loadings indicate strong impact on the temperature series. The equilibrium temperature change for a doubling of carbon dioxide is between 2.15 and 3.40 Celsius degree, which is in agreement with past literature and the report of the Intergovernmental Panel on Climate Change using fifteen different general circulation models.
The second essay examines the effect of random initial condition on the power of unit root tests. It develops the test statistics in the context of unknown structural change models using generalized least square detrended data when the initial observation is drawn from its unconditional distribution. It derives the limiting distributions of M-tests, ADF test and a family of feasible point optimal tests. Using two methods to estimate the break point, this study calculates the power envelopes and asymptotic power functions, and compares them with the case where the initial condition is fixed. Finite sample Monte Carlo simulations under various forms of error processes are performed using different lag length selection methods. Empirical applications are also provided.
The third essay employs a multivariate framework to improve the power of unit root tests. In the context of unknown structural change models, it shows that when testing for a unit root against stationarity with an unknown structural break, substantial power gains can be achieved by incorporating extra information contained in an arbitrary number of covariates. The power gains are dependent on the long-run correlations between the shocks of covariates and quasi-differences of potentially integrated time series. The higher correlation between covariates and quasi-differences of time series, the higher power gains can be realized. A feasible statistic to estimate the unknown correlation is also derived. In finite sample experiments, Monte Carlo simulation results confirm large power improvements without size deteriorations.
| Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/29358 |
| Date | January 2006 |
| Creators | Liu, Hui |
| Publisher | University of Ottawa (Canada) |
| Source Sets | Université d’Ottawa |
| Language | English |
| Detected Language | English |
| Type | Thesis |
| Format | 102 p. |
Page generated in 0.0306 seconds