Abstract:
This article sets a framework for high-frequency nowcasts of private consumption, in terms of percentage changes, which are originally observed quarterly.
The problem has been considered through two mixed-frequency models, both projecting quarterly observed percentage changes into monthly covariates: imports of consumer durables, retail trade, revenue of services, credit card and VAT payments, all adjusted for seasonal variation and available from official sources. The first model uses the MIDAS specification and produces quarterly nowcasts, updated monthly. The second model builds on state-space representation and enables monthly nowcasts.
Timeliness of monthly official information is an important parameter of mixed-frequency models, taken into account either by appropriated parameterization or by endpoint estimates which fulfill incomplete monthly data. I simulate end-of-sample estimates using consumption-related Web-query indices, available weekly from Google Insights for Search. Proceeding from a predetermined pool of 41 Google predictors, a two-step procedure for filtering series has been applied, which selects the best explanatory subset, in accordance with the AIC information criteria. As the predictive content of query indices is time-varying, the best subset is reassigned each month, based on the information known up to the previous month.
The MIDAS specification enables quarterly nowcasts for two months earlier than the bridge department equation, with almost the same RMSFE/AMFE. The monthly nowcasts of private consumption simulated from the state-space model are found (ex-post) to be consistent and highly positively correlated with quarterly realizations.
Preliminary empirical evidence was found that monthly estimated "local-demand" series are more appropriate in inflation equations than the overall index of real economic activity.