where and are paremeters n p and T are self

where α, β, γ2 and γ3 are paremeters; n, p and T are self-explanatory. The implicit test assumption is that the linear combination of fundamentals can be a proxy for the macroeconomic conditions that affect opinion polls. The characteristics that make Lula differ from FHC will be summarized in an intercept dummy, DLula, which assumes 1 during Lula government and 0 during FHC\'s – if this variable is statistically significant, there is evidence that citizen\'s perception about their President is not only related to economic indicators, but also to his or her own idiosyncrasy. In order to control for changes in the political scenario stemming from corruption scandals, we will use an index variable that will be explained in the next section. As mentioned earlier, we use the automated selection procedure embedded in the econometric package Oxmetrics – Autometrics. This algorithm performs a general-to-specific model selection based on the merely of reduction. Designed to simplify dynamic and linear model regressions, they build on the search processes put forward by Hoover and Perez (1999). Autometrics is able to select the relevant variables from those that compose a General Unrestricted Model (GUM), according to specified diagnostic tests and significance levels. If the GUM contains the variables that are important to the Data Generation Process (DGP), it is shown to retrieve a final model that is encompassing (Hendry and Krolzig, 2005). Political Science, New Political Economy and Economic theory help us to specify the variables in the GUM, to ensure that variables are orthogonalized, to perform appropriate data transformations, to calibrate the algorithm and, finally, to interpret the results. The method is appropriate because, for the explanatory variables, lags and deterministic specification in our General Unrestricted Model, we would have to estimate separately 216 sub models and consider 16! possible paths. This computational burden justifies the need for the automated process. We are also able to use a standardized testing procedure for different models and can benefit from the rigour of the “theory of reduction”. Autometrics considers a tree search that corresponds to the whole model space, which are tested until a dominant encompassing reduction is selected (Doornik, 2009). The objective is to reduce a model, possibly finding a specification that is absent of misspecification.
Results We first show the results obtained with the Instrumental Variables Estimation (IVE) using two stage least squares and the GUM as presented in Table 6. Besides the economic and political explanatory variables, the constant, the time trend and Lula\'s dummies, we also included two other dummies in order to control for the variation in approval rates that were not properly captured by these explanatory variables, especially during Cardoso\'s presidency. The first dummy, “negative FHC”, receives the value 1 for June 2000, May and June 2001, −1 for July 2000 and 0 otherwise. We believe that it reflects the volatility in the rate of approval due to the federal political crisis that was sparkled by the violation of the senate\'s electronic voting panel by senator Antônio Carlos Magalhães, a president\'s political ally, as mentioned before. The other dummy is interesting and points out to the existence of a calendar effect during FHC\'s second term. Inspection of the data shows that approval rates increased from 38% in 1999M12 to 56% in 2000M1, from 57.5% in 2000M12 to 65.8% in 2001M1 and, finally, from 60.9% in 2001M12 to 65.6% in 2002M1 without any major change in the economic or political scenario, as measured by our choice of variables. Its inclusion considerably reduced the normality problem of the GUM. Hence, we included a dummy variable, named “January FHC”, which receives the value 1 in all January months during Cardoso\'s presidency and zero otherwise.