Al fine di migliorare la tua esperienza di navigazione, questo sito utilizza i cookie di profilazione di terze parti. Chiudendo questo banner o accedendo ad un qualunque elemento sottostante acconsenti all’uso dei cookie.

Econometric inference of stochastic volatility processes: The Pseudo-Adaptive Block-ABC

6 aprile 2016 ore 12:00 - 13:30

Aula 104, Sede di Viale Romania, 32

Speaker: Giuseppe Brandi , LUISS

Abstract: Approximate Bayesian Computation (ABC) is a
simulation-based method used when the likelihood function of the model under
examination is not available or difficult to compute. This is particularly the
case of high dimensional continuous time models or in presence of latent
variables, as in the stochastic volatility model (in which the volatility
process is treated as latent). The ABC method relies on the fact that it is
easy to simulate from the structural model given some parameters. Using the
simulation series and the actual data, summary statistics on the two datasets
are computed and a distance between them evaluated. If the distance is small
enough, the proposed parameters are accepted. We are going to introduce a new technique to
estimate parameters of a stochastic process via ABC. We want to show that, when
the financial time series is long, the standard approach relying on summary
statistics on the full dataset fails to correctly estimate the
parameters. To overcome this issue, we introduce a blocking approach, enlarging
the set of summary statistics to subsamples summary statistics, which leads to
a more precise estimation. We will also cover the issue of chain mixing with a
pseudo-adaptive algorithm extension and a simulated-data dependent prior