Microsimulation and analysis of income distribution : an application to Italy
The first chapters of the thesis put special emphasis on tax-benefit microsimulation models. The state of the art in the economic literature of tax-benefit microsimulation models is reviewed and discussed. Particular attention is paid to issues such as the reliability of estimation and the grossing-up of the sample. In order to analyze tax-benefit microsimulation, a new model is developed focusing on the case of Italy: it shares many features with other country-specific tax-benefit microsimulation models. The model, appropriately calibrated to population totals, is also used for an estimation of tax evasion via comparison with a number of different data sources. Non-parametric density estimation is used to improve the understanding of policy simulations and to analyze the effect of fiscal reform: an application to the 1998 Italian personal income taxation reform is provided. The first part concludes with an analysis of the reliability of microsimulation models, which has been addressed by few authors before. The analysis is undertaken using the bootstrap, which tends to show a better performance in finite sample than asymptotic approximations. The main result is that static microsimulation does not by itself make confidence intervals larger: on the contrary they can also make it smaller. To improve the reliability of microsimulation models the best way to proceed is to reduce the sampling error of the available data sets. In the remaining chapters the thesis analyzes how microsimulation models can be useful in understanding the causes of inequality trends. As a preliminary step, the review and discussion of the literature about the main methods for inequality decomposition is provided. Based on this, a combination of two recent microsimu-lation methods is proposed to analyze the trend of inequality in Italy in 1977-2000. It is found that analysis using traditional methods of inequality decomposition can be seriously misleading if the sample is not representative of the whole population in some of its dimensions, such as female labor force participation. Microsimulation techniques can overcome this problem and can account for the major factors that driving inequality. Finally, the thesis discusses the issue of inference with thick-tailed distributions, such as the Pareto distribution with infinite second moment, that is of special relevance to empirical analysis of income distribution. It is shown that inference based on the standard t-ratio statistic can induce a non negligible error in rejection probability. Some solutions are suggested with an application to Italian household income data.