中国季度GDP的即时预测与混频分析

Nowcasting China’s Quarterly GDP Using Mixed-Frequency Data

及时、准确地获得GDP短期预测值对于宏观调控和企业决策至关重要。本文在收集我国实时碎尾数据集的基础上,采用混频动态因子模型,将我国季度GDP的预测频率由"季度"提高到"日度"。研究结果表明,相对于混频抽样模型以及MFVAR等现有模型,混频动态因子模型能够有效解决实时预测中需要面临的数据问题,包括混频指标、碎尾特征、数据的周期性缺失等。本文模型在每个数据发布日,均可更新GDP的预测结果,这不仅将最新的经济活动信息迅速地体现到GDP预测中,而且显著提高了GDP即时预测的准确性,且预测结果随着月度数据信息的增加趋近于GDP真实值。此外,本文还估算了拟GDP季度同比增长率和GDP月度同比增长率两个月度数据序列,为我国宏观经济监测与政策分析提供一定的数据支撑。

As GDP can comprehensively reflect the economic condition of a country or a region,GDP predictions are carefully scrutinized by many institutions.However,because GDP is usually only calculated on a quarterly frequency and released after a delay of 3 weeks,classical forecasting models cannot provide accurate and timely GDP predictions.However,some macroeconomic variables that are highly correlated with GDP,such as industrial added value,import and export volumes,and the total retail sales of consumer goods,are released monthly with a much smaller delay.The incorporation of this monthly information into GDP prediction could therefore improve the timeliness of GDP forecasting,enable the correct evaluation of economic conditions,and facilitate the formulation of appropriate macroeconomic regulations.However,the incorporation of these monthly indicators into economic forecasting models will require the solution of key problems associated with these indicators’underlying data,namely its mixture of data frequencies,the ragged-edge behavior of real-time data,the presence of data revision,and the periodic absence of data.To deal with the problems of absent data and ragged-edge data,we nowcast China’s GDP based on Zheng and Wang’s(2012,2013)mixed-frequency dynamic factor model for year-on-year growth rates.Compared with mixed data sampling(MIDAS)and mixed-frequency vector autoregression(MFVAR)models,the mixed-frequency dynamic factor model accounts for missing data in addition to dealing with ragged-edge data,and thus makes full,accurate,and timely use of the data.In addition,a year-on-year growth rate model is more useful in China,as the National Bureau of Statistics announces only year-on-year growth rates for most macroeconomic indicators,and policymakers focus on year-on-year GDP growth rates.Moreover,as year-on-year growth rates are based on data for the same month or quarter each year,they can mitigate effects due to seasonality,which is not generally accounted for in official year-on-year economic growth data released in China(and is another reason why a year-on-year growth rate model is appropriate for China).Thus,once new data are released,we can immediately update our nowcasting result.This means that we can nowcast the quarterly GDP growth rate in real time using the most up-to-date data and thus provide a reliable and timely economic prediction for decision-makers.Our results also show that the mixed-frequency dynamic factor model provides more accurate predictions than the MIDAS and the MFVAR models.In addition to developing GDP nowcasting,we derive some other important results.First,in contrast to GDP data,which are announced quarterly,in April,July,October,and January,with an approximately 3-week delay,we can estimate the quarterly GDP growth rate on a monthly basis.For example,we estimate the year-on-year growth rate of GDP from February to April,which is crucial information for decision-makers and for economic modeling.Second,in addition to determining the smoothed estimator of a common factor,we also obtain the smoothed estimator of the idiosyncratic factor.The sum of these factors is then used to derive a smoothed estimate of the monthly year-on-year growth rate of real GDP.Third,we circumvent the missing data problem,which is due to the restriction of statistical rules,the effect of the Spring Festival,and other factors,by estimating the missing data from the observations of other indicators,which affords a complete time series of data.In summary,our GDP nowcasting method enables daily(rather than quarterly)forecasting of China’s quarterly GDP growth rate,which means that we can incorporate the latest information about economic conditions into our forecasts.Thus,we can provide more timely and reliable economic predictions to policymakers.From the perspective of macroeconomic regulation,these economic predictions may allow policymakers to generate real-time updates of its judgments on current economic conditions and thereby formulate more timely and suitable macroeconomic policies.From the perspective of microeconomic decision-making,these economic predictions may enable enterprise managers to understand the current economic situation more clearly and in a timely fashion,and thereby to efficiently adjust investment plans and development strategies.Thus,we believe that our nowcasting-based economic predictions will be invaluable for developing more effective national-level macroeconomic control and enabling better market-level microeconomic decisions.