Footnote:
Nach Informationen von SSRN wurde die ursprüngliche Fassung des Dokuments February 28, 2022 erstellt
Description:
Since most official estimates of major macroeconomic variables that market participants and policy makers keenly monitor in real time come out with significant publication lags, nowcasting literature especially targeted on key macro variables has trended over the recent years. Nowcasting refers to forecast present values of monthly or quarterly variables by distilling information from big data in real time. Hence, nowcasting is regarded to be a data-driven approach with fewer subjective choices on model specifications involved along the way.In an attemp to build up nowcasting platforms, major central banks, such as the European Central Bank, Bank of England, and Federal Reserve branches in the United States, have delved into fashioning newer nowcasting methodologies over the course. Most markedly, Federal Reserve Bank of New York had regularly published their GDP nowcasting estimates on their website until September 2021.1 GDP is one of the most representative macroeconomic variables that even laypersons can intuitively apprehend, but its official estimates usually are out with 1 or 2 months lagged.A dynamic factor model that the New York Fed has deployed (see Bok et al., 2018) enables to combine information scattered across from different sectors of economies and nowcast based on the dynamics of intercorrelated economic variables. Its underlying mixed-frequency model allows for the stacking of latest observations of varying frequencies and computerized algorithms extract a handful of factors from a hodgepodge of market observations, which describe the overall underlying movements of a big data set. Lastly, factors recursively updated at each window churn out nowcasting estimates on a recursive basis.Recently, nowcasting literature has taken up more sophisticated deep-learning algorithms and introduced survey data and text data from newspapers and the internet (Aguilar et al., 2021). Hirschb hl et al. (2021) have found that soft data and text data are especially informative for short-term forecasts and can play a central role when constructing market sentiment metrics. In this study, we attempt to nowcast Korean industrial production index adopting the Bok et al. (2018)’s methodology. Industrial production index is also one of real variables to be frequently referred to if one is to get the grasp of current industrial activities. We come up with a systemized framework for nowcasting the industrial production index and found that real variables, such as electricity sales and exports, can be of a great help to track the real-time movements of industrial activities.The next section describes our nowcasting model framework with details on a state-space model framework and its constituent explanatory variables. In section 3, we analyze our empirical results with the primary objective being to confirm the forecast superiority of our dynamic factor model. Finally, we wrap up our analysis with policy implications and further technical suggestions to improve our framework