Resumen de Essays on panel data econometrics: local goverments, politicians
The rst two chapters of this thesis use panel data techniques to study two important questions in political economy: what is the e ect of having more informed voters on the spending decisions of local governments and why members of the Parliament choose to favour their birthplace when allocating government transfers.
Chapter 1 uses a new dataset on disaggregated expenditures and revenues of all Italian municipalities to analyse how the political budget cycle – that is, the strategic increase in public good provisions in proximity of elections – is a ected by giving voters better information on the incumbent government’s spending decisions. The rst contribution of this chapter is to show that the political budget cycle in Italian municipalities is sizeable: investment expenditures rise during the term, peak just before elections and drop in the election year. Specically, compared to the election year, in the year before elections expenditures are, on average, 29% larger than the sample mean. Using variation in voters’ information generated by a reform that, in 2008, required all Italian municipalities to disclose their balance sheet before elections to voters, I compare the budget cycle uctuations when voters have access to this piece of information with when they do not. Results from a di erence-in-di erences specication show that the pre-electoral increase in investment expenditures is reduced after the reform, when voters have access to the information of the balance sheet, suggesting that local governments react to more informed voters by refraining from strategically manipulating spending. To conrm this hypothesis, I then collect data on local newspaper sales and observe that in provinces with relatively more newspaper readers the e ect of the reform is much stronger, whereas in provinces with fewer readers the reform has almost no e ect. The main conclusion of chapter 1 is that, in presence of voters that are well informed about how the provision of public goods is nanced, incumbent local governments alter their spending and nancing behaviour by reducing the budget cycle.
Strategic manipulation of public spending is, however, not limited to local politicians. As the empirical analysis in chapter 2 shows, also member of the Parliament may be in uenced by their career concerns when allocating public resources. The kind of pork-barrel policies studied in this chapter are however, not strictly connected to re-election, but instead they relate to job prospects after the mandate in Parliament is over. Using a particular feature of the Italian legislative election system during the 1994-2006 period, chapter 2 shows that birth towns of a particular kind of parliamentarians – those who were elected in a district that does not include their birth town – are favoured with additional central government transfers while these politicians are in o ce. Part of this bias can be explained by the fact that politicians, foreseeing the moment in which they will have to leave Parliament, use transfers to their birth town to signal their interest in running for o ce in the future either as mayor or as members of the local government.
The rst two empirical chapters are complemented by a more methodological contribution to the analysis of panel data contained in chapter 3. This chapter deals with the problem of testing linear hypothesis in panels with a factor models structure. Factor models are a common choice in several branches of economics. However, distributional results for the estimated factors and loadings only cover the asymptotic distribution of the factors for a given time period and, similarly, the distribution of the vector of loadings for a given individual, but do not show how to perform tests of equality of the coe cients of the model. To ll this gap in the literature, chapter 3 proposes a simple testing procedure that can be used to test hypotheses on the equality of loadings, factors, or both. With simple modications this procedure can be used to test several di erent hypotheses previously considered in the literature.
The asymptotic results allow for both time-series and cross-sectional correlation in the disturbances, often present in the data even after controlling for common factors, and are therefore widely applicable. The chapter ends with an application of the test to the dataset used by Reis and Watson (2010), in order to reconsider their hypothesis about the existence of a pure in ation factor. Even after correcting for serial and cross-sectional correlation one strongly rejects the null of the existence of a pure in ation factor in US sectoral prices data, suggesting that the more robust version of the test should be used in most applications.
This thesis uses modern econometric techniques to provide credible empirical evidence and new methods for the analysis of panel data. Even the use of rigorous econometrics, however, cannot overcome all the limitations of the approach taken here. In particular, although the results in chapter 1 and 2 are econometrically sound, their interpretation as being a consequence of explicit decisions of a politician may be subject to critique, as there is no credible way to completely rule out other possible explanations. Also, the study of political incentives and career concerns is a complex matter and the results of the rst two chapters may be supplemented by further work. Like any other reform evaluation, chapter 1 su ers from a methodological limitation because the di erence-in-di erences approach assumes that no other confounding factor or policy intervention that might have an e ect similar to the reform happened at the same time as the reform. This assumption is relatively strong and further evidence that this concern is not invalidating the results would be useful. Regarding chapter 2, mechanisms behind the birth town bias may not be limited to the post-congressional career concerns but may include, for example, gratitude towards the local party structure or simply the desire to help relatives and acquaintances with additional transfers. Although testing empirically these possibilities is challenging, it would certainly help to understand better and take into account other possible drivers of politicians’ behaviour. Also, the definition of connection used in the empirical analysis is potentially too narrow and may fail to capture other relevant ways in which politicians may be related to local administrations, such as, for instance, political alignment.
The theoretical framework described in chapter 3 also su ers from at least two shortcomings.
First, the assumption that number of common factors is known is convenient for the asymptotic analysis but empirically implausible. Second, the procedure only allows to test for a particular type of linear restrictions on the parameters. Solving both these limitations is, in my opinion, a promising avenue for future research. Also, the testing methodology can be extended in several interesting directions. First, in the context of macroeconomic forecasting, Breitung and Eickmeier (2011), Chen, Dolado and Gonzalo (2014) and Han and Inoue (2013) have recently proposed testing procedures for structural breaks in approximate factor models. In this respect, the methods of chapter 3 could be adapted to models with structural breaks in the loadings, in which case one has to endogenize the timing of the break. Also, recent papers have proposed factor models that use common and block-specic factors to capture the between and within-block variations in the data, hence incorporating structure, taking advantage of it in terms of e ciency, and facilitating the interpretation of the results (Kose, Otrok and Whiteman 2003, Stock and Watson 2009, Moench, Ng and Potter 2013).
Our approach could also be extended to develop specication tests for the aforementioned models. All these topics constitute interesting avenues for further research.
