| Many empirical literature in economics, social sciences, and medical treatment studies the causal effects of programs, polices or drug effects. In the economic context, the major focus on program evaluation literature is to measure the impact of a particular treatment on a set of individuals, regions, or countries that exposed to such a treatment. It is of particular importance to policy makers, medical practitioners and others. In order to evaluate the effect of the treatment, Rubin (1974) proposed the interpretation of causal statements as comparisons of the so-called potential outcomes, which is defined as a pair of outcomes associated to a particular individual given different levels of exposure to the treatment with only one of the outcomes observed by researchers. Models are developed for this pair of potential outcomes---one for the treated state, another for the control state. In this thesis, a panel data methodology is proposed under the potential outcomes framework to measure the average treatment effect. This methodology is applied to measure the impact of two major Hong Kong polices on the Hong Kong's economy. Due to the model uncertainty involved in forming the counterfactual, forecast combination methods are proposed as a way to reduce this uncertainty. Various existing methods together with some new proposed methods are evaluated in a small scale Monte Carlo studies and two applications. |
