دانلود رایگان مقاله شناسایی نقاط عطف چرخه کسب و کار در زمان واقعی با کوانتیزاسیون برداری

abstract

We propose a simple machine-learning algorithm known as Learning Vector Quantization (LVQ) for the purpose of identifying new U.S. business cycle turning points quickly in real time. LVQ is used widely for real-time statistical classification in many other fields, but has not previously been applied to the classification of economic variables, to the best of our knowledge. The algorithm is intuitive and simple to implement, and easily incorporates salient features of the real-time nowcasting environment, such as differences in data reporting lags across series. We evaluate the algorithm’s real-time ability to establish new business cycle turning points in the United States quickly and accurately over the past five NBER recessions. Despite its relative simplicity, the algorithm’s performance appears to be very competitive with those of commonly used alternatives. © 2016 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.

4. Conclusion

Non-parametric machine learning algorithms are used commonly in many disciplines to classify data as belonging to one of a set of classes. We have proposed a particularly salient algorithm, known as learning vector quantization, for the purpose of classifying economic data into expansion and recession regimes. Of particular interest is the ability of the algorithm to identify US business cycle turning points accurately and quickly in real time.We evaluate the real-time performance of the LVQ algorithm for identifying business cycle turning points in the United States over the past 35 years and five recessions. The LVQ algorithm identified the dates of all five recessions over this period accurately, with no false positives, and at an impressive speed. For example, the LVQ algorithm would have identified the December 2007 peak in economic activity by early June 2008, several months ahead of the statistical tracking procedures reviewed by Hamilton (2011) as being in use at the time. Looking across all recessions, the algorithm’s speed of identifying peaks and troughs over our sample period is very competitive with that of the dynamic factor Markov-switching model, a technique that is used commonly for dating business cycles in real time.