پیش بینی فعالیت های اقتصادی فنلاندی با استفاده از داده ها در سطح بنگاه   Predicting Finnish economic activity using firm-level data

1. Introduction Statistical agencies, central banks, and numerous public and private entities collect hundreds, if not thousands, of economic time series every year. This ever-growing amount of data has helped policymakers and researchers with key activities such as forecasting, evaluating the performances of economic models, and designing fiscal and monetary policies. Unfortunately, this wealth of data is not matched by a high degree of timeliness. Most notably, variables measuring economic activity are generally published with long lags. For example, the first estimates of the US and UK quarterly GDP are published four weeks after the end of each quarter, while the lag is usually six weeks for the Euro area (see Banbura, Giannone, & Reichlin, 2011). In recent years, this problem of the timeliness of data releases has been addressed in the literature on nowcasting models and coincident economic indicators (for the latter, see e.g. Altissimo, Cristadoro, Forni, Lippi, & Veronese, 2010; Stock & Watson, 1989). Nowcasting methods have been applied chiefly in the prediction of low frequencydata, and quarterly data in particular, by exploiting the release of monthly data (see, e.g., Aastveit & Trovik, 2014; Banbura et al., 2011; Evans, 2005; Giannone, Reichlin, & Small, 2008). In these papers, the focus has been on the creation of early estimates of the quarterly GDP growth, which are updated as new information is released. These revisions are analyzed by checking the contributions of the news carried by additional data. Most of the nowcasting papers are interested in quarterly variables, though Modugno (2013) and Proietti (2011) focus on computing monthly nowcasts of GDP. Aruoba, Diebold, and Scotti (2009) propose a real-time economic activity indicator that is built on data observed at mixed frequencies, including daily data. Recent examples of nowcasting applications are those of Camacho and Garcia-Serrador (2014) and Camacho and Perez-Quiros (2010), who use a single-index dynamic factor model based on both real and financial indicators, and Foroni and Marcellino (2014), who apply various different approaches (bridge equations, state space and mixed data sampling models) to the nowcasting of Euro area GDP components. Finally, a recent survey on nowcasting with parsimonious mixed-frequency methods is provided by Camacho, Perez-Quiros, and Poncela (2013). The novel idea introduced in this study is to exploit the information contained in large firm-level datasets in order to compute early estimates of economic activity. In particular, we compute nowcasts of the Finnish monthly economic activity indicator, the Trend Indicator of Output (TIO), using a two-step procedure. In the first step, we extract common factors from a large firm-level dataset of turnovers, then in the second step we use these common factors as predictors for nowcasting regressions. The estimates constructed for TIO are also used subsequently to compute early figures of the Finnish quarterly GDP.