Maximum Likelihood Estimation of Factor Models on Datasets With Arbitrary Pattern of Missing Data

Abstract

4Non-technical summary 5

Now-casting inflation using high frequency data

“…In order to deal with these two issues we adopt the estimation methodology proposed in Banbura and Modugno (2010), that generalizes, for the case of missing data, the methodology proposed by Watson and Engle (1983). The latter methodology is based on the ExpectationMaximization (EM) algorithm under the assumption of an exact factor model, i.e.…”

“…This allows to impose a factor structure on the i th variable only when y it is available. Moreover, Banbura and Modugno (2010) show how to impose restrictions on the parameters, in order to impose a block structure to the factor model.…”

“…The estimation adopts the methodology exposed in Banbura and Modugno (2010), which is in turn based on the methodology proposed by Watson and Engle (1983). Doz, Giannone and Reichlin (2006) show that the latter allows to estimate factor models by maximum likelihood to deal with panels with arbitrary patterns of missing data, e.g.…”

“…MIDAS, the methodology proposed in Banbura and Modugno (2010) models data within a unified single framework that allows to forecast all the involved variables. This is achieved by using a factor model, which, by definition, does not suffer from overparametrisation.…”

“…Banbura and Modugno (2010) generalize this methodology in order proposed originally by Ghysels, Santa-Clara, and Valkanov (2002) has been used by Monteforte…”