Research interests
Research profile
During the first part of my PhD course, I kept working on some issues raised during the preparation of my B.A. thesis. To partially relief the computation burden associated with fractionally integrated models, I have derived the analytic versions of the gradient and the Hessian matrix of FIGARCH models and I have analyzed, in a Monte Carlo study, the benefits or their employment, both for what concerns the accuracy of the estimates and the quickness of execution. The work was firstly presented at the conference S.Co. 2001 and then published on Statistical Methods and Applications.
The main body of the thesis was extended and integrated to form a monograph, on time series models with long memory, with a particular focus on the estimation and computational issues.
The last by-product of my B.A. thesis was a paper in which I have analyzed the memory features of daily price ranges of stocks. The paper was presented at the SIS2002 conference. This work triggered my interest in the UHF dynamics of stock markets. Together with Massimilano Cecconi, I earned a grant for young researchers issued by the University of Florence. Thanks to this grant, we could buy the trades and quotes data from the NYSE and set up, in collaboration with Christian T. Brownlees, a database for their storage, extraction and filtering. A first survey work concerning data issues and models of the ACD class was published on Scienza & Business. A second and more substantial paper was presented at the conferences MAF2004 in Salerno and at the International Workshop on Statistical Modelling and deals with the exploitation of infra-daily information for the description of the daily volatility dynamics.
Concurrently, I have completed and discussed my PhD thesis, which deals with simulation-based estimation procedures for a-stable distributions and processes. Those distributions originate from a generalized version of the central limit theorem and allow for skewness and heavy tailedness. However, due to estimation difficulties, their diffusion among practitioners has been limited. My project was to employ simulation-based estimation methods for the estimations of statistical models based on alpha-stable distributions. The first part of the thesis concerns the indirect inference approach. It is shown that indirect inference estimators for the parameters of a-stable models can be constructed straightforwardly by using skew-t distributions as auxiliary models. The relative paper is now being considered for publication on the Econometrics Journal.
The second part of the thesis deals with Bayesian MCMC methods. It is shown that, by employing a FFT-based approximation of the density function, a random walk Metropolis algorithm can be straightforwardly implemented and is computationally less expensive with respect to other MCMC samplers that have been proposed in the literature. The paper has been published on Computational Statistics and Data Analysis.
The last part of my thesis was completed during my visit to the department of engineering at the University of Cambridge. There, I have worked on simulation-based filtering methods for time series (particle filters). A particle filter that allows the noise term in the observations equations to be a--stable, or more in general a scale mixture of normals, was introduced. This filter was shown to outperform standard Gaussian filters in audio processing problems when the signal is highly corrupted. The resulting paper was presented at the conference EUSIPCO2004 and has been published on IEEE Transactions on Signal Processing.
After completing my Ph.D. thesis I have worked as research fellow in the project “Ultra-high frequency dynamics of financial markets” (national coordinator Prof. Rosario N. Mantegna, local coordinator Prof. Giampiero M. Gallo), concentrating on simulation based estimation methods for infra-daily stochastic volatility models. I have proposed an indirect estimation approach to stochastic volatility models with heavy-tailed innovations in either the returns or the volatility equation, and I point out how such a kind of models can be suitable for the analysis of currency crises; I have been invited to present a first contribution at the 2005 Bundesbank Fall conference in honour of Benoît Mandelbrot; the relative paper was later submitted to the Journal of Financial Econometrics.
After that, I joined the Money and Banking Statistics Division of the European Central Bank. Besides being involved in the data production and seasonal adjustment of monetary aggregates, I have also participated in two distinct research projects, under the supervision of Antonio Matas-Mir. The first one concerns detection and dating of breakpoints in the seasonal components of macroeconomic time series, with a particular focus on the forecasting performance of models allowing for trends and breaks in the seasonal factors. The second deals with the effects of the application of moving average-based filters (e.g. X11, X12 or TRAMO-SEATS) on the dating of business cycles in a Markov switching modelling framework; the relative paper has been accepted for publication on Journal of Applied Econometrics.
Furthermore, I have recently started to work with Silvia Sgherri on a number of projects related to the application of sequential Monte Carlo methods for the estimation of DSGE models; in our first contributions, we employ a particle filter to estimate the natural rate of interest and compare the US and the European monetary policy stance. In addition to extend this analysis to other central banks, with a special focus on those adopting an explicit inflation target, we also plan to apply sequential Monte Carlo methods to investigate the role of learning as a source of inflation inertia. As a more ambitious project, I would like to examine the relationship between particles filters and economic theory, based on the intuition that the death-birth mechanism governing the evolution of the filter can provide an useful framework in which to inscribe some economic concepts such as learning and inertia.
Affiliations