Bernhard and Leblang carry out an interesting investigation

Bernhard and Leblang (2006) carry out an interesting investigation of opinion polls and economic news for a parliamentary system. The authors study the relationship between unexpected exchange rate changes and opinion polls in the United Kingdom from the June 11, 1987 election until June 21st, after the 2001 election, using a weekly time series. Their ddr1 is that variables are endogenous in the sense that currency market activity affect opinion polls, as exchange rates can be related to welfare (and vice-versa), as traders anticipate a change in cabinet and thus economic policy. They estimate a voting intention model for the government, including the first difference of the voting intention on the left hand side and controls, such as lagged vote intention, changes in inflation and unemployment and a great number of dummies for economic and political events – for example, dummies for a “honeymoon” period, election dates, the stock market crash of 1987, the foot-and-mouth disease and so on. Results show that changes in unemployment are significantly related to changes in voting intention whereas inflation and other control variables are not significant (with the clear exception of the dummy variables). They use the residuals of this estimation to create a variable that would supposedly reflect unexpected positive and negative news regarding political support. Finally they test the correlation between political news and exchange rate surprises. Their finding is that unexpected depreciations lead to a decrease in the voting intention for the government and that negative public opinion shocks cause a currency depreciation. Also, exchange rate volatility increases when opinion polls do not clearly indicate the outcome of an electoral outcome. The number of studies concerned with the Brazilian Political Economy has been increasing over the recent years. As a first example, Sakurai and Menezes-Filho (2008) find that, in a panel of Brazilian municipalities from 1988 to 2000, mayors who spend more during their terms in office increase the probability of their own reelection or of a successor of the same political party. In particular, higher capital spending over the years preceding elections and higher current expenditures in election years are beneficial to their reelection. Motivated by a different reason, but still regarding political aspects in the Brazilian society, Ferraz and Finan (2011) conclude that first-term mayors are associated with significantly less corruption than those in their second and final term. The level of corruption is especially lower in those cities with a higher level of political competition. According to the authors, these results suggest that the possibility of re-election creates a “discipline effect”, inducing first-term mayors with re-election incentives to extract fewer rents from power. However, in a related empirical assessment regarding Brazil, Pereira et al. (2009) argue that when corruption is not likely to be detected and the pay-offs involved are very large, Brazilian politicians do not align their interests with that of the voters and are likely to engage in corrupt practices. Their results are quite opposite to those of Ferraz and Finan (2011) and suggest that enhancing the quality and quantity of information available to citizens is not enough to foreclose the incentive to commit crimes. However, irrespective of their conclusions and purposes, the latter studies and our own have the Brazilian Political Economy as a common motivation.
Methodology where the subscript t refers to time, a is the approval rating of the president, is the jth economic or political indicator, γ0 is the idiosyncratic characteristic of the president that is important for his/hers popularity (we assume that this is some constant level of approval given by people\'s preferences), γ1 controls for a deterministic time trend, ϵ aggregates all random unobserved variables that affect the president\'s approval and finally, α are parameters. Given that the approval rating might exhibit some dynamics, we add memory to the process of the dependent variable and we also add lags to the independent variables, as can be seen below: