Where Are We Now? Real-Time Estimates of the Macro Economy

This paper describes a method for calculating daily real-time estimates of the current state of the U.S. economy. The estimates are computed from data on scheduled U.S. macroeconomic announcements using an econometric model that allows for variable reporting lags, temporal aggregation, and other complications in the data. The model can be applied to …nd real-time estimates of GDP, in‡ation, unemployment or any other macroeconomic variable of interest. In this paper I focus on the problem of estimating the current level of and growth rate in GDP. I construct daily real-time estimates of GDP that incorporate public information known on the day in question. The real-time estimates produced by the model are uniquely-suited to studying how perceived developments the macro economy are linked to asset prices over a wide range of frequencies. The estimates also provide, for the …rst time, daily time series that can be used in practical policy decisions. Keywords: Real-time data, Kalman Filtering, Forecasting GDP JEL Codes: E37 C32 Introduction Information about the current state of real economic activity is widely dispersed across consumers, …rms and policymakers. While individual consumers and …rms know the recent history of their own decisions, they are unaware of the contemporaneous consumption, saving, investment and employment decisions made by other private sector agents. Similarly, policymakers do not have access to accurate contemporaneous information concerning private sector activity. Although information on real economic activity is collected by a number of government agencies, the collection, aggregation and dissemination process takes time. Thus, while U.S. macroeconomic data are released on an almost daily basis, the data represent o¢ cial aggregations of past rather than current economic activity. The lack of timely information concerning the current state of the economy is well-recognized among policymakers. This is especially true in the case of GDP, the broadest measure of real activity. The Federal Reserve’s ability to make timely changes in monetary policy is made much more complicated by the lack of contemporaneous and accurate information on GDP. The lack of timely information concerning macroeconomic aggregates is also important for understanding private sector behavior, and in particular the behavior of asset prices. When agents make trading decisions based on their own estimate of current macroeconomic conditions, they transmit information to their trading partners. This trading activity leads to the aggregation of dispersed information and in the process a¤ects the behavior of asset prices. Evans and Lyons (2004a) show that the lack of timely information concerning the state of the macroeconomy can signi…cantly alter the dynamics of exchange and interest rates by changing the trading-based process of information aggregation. This paper describes a method for estimating the current state of the economy on a continual basis using the ‡ow of information from a wide range of macroeconomic data releases. These real-time estimates are computed from an econometric model that allows for variable reporting lags, temporal aggregation, and other complications that characterize the daily ‡ow of macroeconomic information. The model can be applied to …nd real-time estimates of GDP, in‡ation, unemployment or any other macroeconomic variable of interest. In this paper, I focus on the problem of estimating GDP in real time. The real-time estimates derived here are conceptually distinct from the real-time data series studied by Croushore and Stark (1999, 2001), Orphanides (2001) and others. A real-time data series comprises a set of historical values for a variable that are known on a particular date. This date identi…es the vintage of the real-time data. For example, the March 31st. vintage of real-time GDP data would include data releases on GDP growth up to the forth quarter of the previous year. This vintage incorporates current revisions to earlier GDP releases but does not include a contemporaneous estimate of GDP growth in the …rst quarter. As such, it represents a subset of public information available on March 31st. By contrast, the March 31st. real-time estimate of GDP growth comprises an estimate of GDP growth in the …rst quarter based on information available on March 31st. The real-time estimates derived in this paper use an information set that spans the history of data releases on GDP and 18 other macroeconomic variables. A number of papers have studied the problem of estimating GDP at a monthly frequency. Chow and Lin (1971) …rst showed how a monthly series could be constructed from regression estimates using monthly data related to GDP and quarterly GDP data. This technique has been subsequently integrated into VAR forecasting procedures (see, for example, Robertson and Tallman 1999). More recently, papers by Lui and Hall (2000) and Mariano and Murasawa (2003) have used state space models to combine quarterly GDP