Can GDP Measurement Be Further Improved? Data Revision and Reconciliation

Recent years have seen many attempts to combine expenditure-side estimates of U.S. real output (GDE) growth with income-side estimates (GDI) to improve estimates of real GDP growth. We show how to incorporate information from multiple releases of noisy data to provide more precise estimates while avoiding some of the identifying assumptions required in earlier work. This relies on a new insight: using multiple data releases allows us to distinguish news and noise measurement errors in situations where a single vintage does not. We find that (a) the data prefer averaging across multiple releases instead of discarding early releases in favor of later ones, and (b) that initial estimates of GDI are quite informative. Our new measure, GDP⁺⁺, undergoes smaller revisions and tracks expenditure measures of GDP growth more closely than either the simple average of the expenditure and income measures published by the BEA or the GDP growth measure of Aruoba et al. published by the Federal Reserve Bank of Philadelphia.

近年来，学界已开展诸多尝试，将美国实际产出增长率的支出端估计值（GDE）与收入端估计值（GDI）相结合，以优化实际GDP增长率的测算结果。本文阐述了如何纳入多批次含噪数据的信息以生成更精准的估计结果，同时规避了既往研究中所需的部分识别假设。这一方法依托一项新的理论洞见：在单一批次数据无法实现的场景中，利用多批次数据可甄别新闻型与噪声型测量误差。本文的研究结果显示：其一，数据拟合更支持对多批次数据进行平均，而非舍弃早期发布数据转而采用后期数据；其二，收入端的初始估计值具备较强的信息价值。本文提出的全新测算指标GDP++，其修订幅度更小，且与GDP增长率的支出端测算值贴合度更高——相较于美国经济分析局（BEA）发布的支出端与收入端指标的简单平均值，或是费城联邦储备银行发布的Aruoba等学者提出的GDP增长率测算指标，该指标均表现更优。