شناسایی نقاط عطف چرخه کسب و کار در زمان واقعی با کوانتیزاسیون برداری  Identifying business cycle turning points in real time with vector quantization

1. Introduction A traditional view of the US business cycle is that of alternating phases of expansion and recession, where an expansion corresponds to widespread, persistent growth in economic activity, and a recession consists of a widespread, relatively rapid, decline in economic activity. The timely identification of the turning points between these phases, or peaks and troughs, is of considerable importance to policymakers, financial markets, firms, and individuals. A substantial body of literature has focused on the prediction of future turning points using a variety of leading economic and financial time series, with some limited success.1 A smaller body of literature has focusedon the identification of turning points that have already occurred, using economic variables that are coincident with the business cycle. The problem of the ex-post identification of turning points is of particular interest, because there are many examples of turning points that have not been predicted exante. This spotty forecasting record means that economic agents are left trying to determine whether a new business cycle phase has already begun. Even this is a difficult task, with new turning points usually not being identified until many months after they occur. The official chronology of business cycle turning points in the United States is provided by the National Bureau of Economic Research’s (NBER) Business Cycle Dating Committee, which has historically announced new turning points with a lag of between four and 21 months. Statistical models improve on the NBER’s timeliness considerably, with little difference in the timing of the turning point dates established.2 However, in general these models don’t identify turning points until several months after they occurred. As a recent example of this, Hamilton (2011) surveys a wide range of statistical models that were in place to identify business cycle turning points in real time, and finds that such models did not send definitive signals regarding the December 2007 NBER peak until late 2008.3 One important reason for these identification lags is data reporting lags, as many key coincident indicators are released only with a lag of one to two months. Another factor is the need for several months of negative or positive data to accumulate before a definitive turning point signal can be uncovered.