Dynamic Factor Models

Nowcasting aggregate services trade

“…Google Trends search data have been used as a tool to monitor economic crises (Askitas and Zimmermann, 2009[35]) or directly measure investor interest (Varian and Choi, 2009); (Jun, Yoo and Choi, 2018 [34]; Choi and Varian, 2012[36]). For instance, Choi and Varian (2012[36]) applied Google search data to forecast automobile sales, unemployment claims, travel destination planning and consumer confidence and D'Amuri and Marcucci (2017 [37]) to forecast unemployment. Gil et al (2018[38]) found that combining Google Trends search data with traditional indicators helps to predict private consumption.…”

“…Those models have been found to be useful for nowcasting major economic aggregates (Doz and Fuleky, 2019) and have been widely used to predict global trade (Guichard and Rusticelli, 2011[6]; Martínez-Martín and Rusticelli (2020 [7]); Cantú (2018 [8])). More specifically N variables (xit), for i = 1, …, N and t = 1, …, T, where t refers to the time index, are each assumed to be the sum of two unobservable orthogonal components: one component resulting from the factors that are common to the set of variables (or common component), ( ), and an idiosyncratic component ( ), which covers the shocks specific to each of the variables.…”

Permanent-Transitory decomposition of cointegrated time series via dynamic factor models, with an application to commodity prices

“…Essentially, the vector of variables is split into a common component, capturing the joint movement of all the observable series, and an idiosyncratic component, which is variable-specific. Although DFMs are now a standard tool in applied macroeconomics and finance (extensive surveys can be found in Bai and Ng (2008); Watson (2011, 2016); Doz and Fuleky (2020)), it is only recently that the issues of non-stationarity and cointegration have begun to receive systematic attention in the literature.…”

On the truncation criteria in infinite factor models

“…When k < p , the factor analytic decomposition is condensing the available information into independent aggregates that could be potentially meaningful. In view of this, factor models are commonly used in many applications in order to deal with data sets where a large number of observed variables is thought to reflect a smaller number of latent variables, especially if the latter present an easy interpretation, as in the individual skills and behaviours in psychology (see, e.g., the review article of Fabrigar et al 1999) or economics (Heckman et al 2006; Doz & Fuleky, 2020). In these settings, the choice of the number of factors k is a crucial task in determining the model success.…”