关键词: 中间投入, 投入产出表, 即时预测, 高维动态因子模型, 自适应稀疏主成分分析

Abstract: As a balance table reflecting the relationship between various departments, the input-output table is an important data basis for input-output analysis. The intermediate input provides strong support to observe economic cycles and formulate macroeconomic and industrial policies. However, the compilation of input-output tables requires high standards of basic data, and it is difficult to increase the update frequency. Inspired by GDP Nowcasting (Giannone et al., 2008), we use a large number of high-frequency macro and industrial economic indicators to estimate and forecast the intermediate input immediately and effectively. Firstly, we construct a high-dimensional macroeconomic indicator set composed of more than 2000 sequences. Then we propose a high-dimensional dynamic factor model based on auto-adaptive sparse principal component analysis. Next, we estimate the quarterly intermediate inputs of various departments and forecast the relevant data for 2018 and 2019. The research results show that the high-dimensional model constructed in this paper can estimate the quarterly intermediate input data, and its fitting and forecasting performance are better than the traditional time series model and traditional dynamic factor model (Giannone et al., 2008). In addition, we also analyze the estimated intermediate input, intermediate input rate and intermediate input contribution coefficient after frequency increase. Finally, we attempt to construct the intermediate flow matrix in the input-output table based on the method of Nowcasting, whichverifies the feasibility and stability of the high-dimensional dynamic factor model in forecasting the input-output table.

Key words: