使用贝叶斯结构时间序列模型预测增长

Nowcasting Growth Usingthe Bayesian StructuralTime Series Model: Prepared by Sunwoo Lee WP/26/49 2026MAR IMFWorking Paper African Department Nowcasting Growth UsingtheBayesian Structural Time Series Model: Application to Tanzania Authorized for distribution by Justin Tyson IMF Working Papersdescribe research in progress by the author(s) and are published to elicitcomments and to encourage debate.The views expressed in IMF Working Papers are those of the ABSTRACT:In light of recent global shocks and rising external volatility, there is a growing need to effectivelymonitor short-term economic fluctuations, especially in countries with limited access to high-frequency growthdata. This paper examines the application of the Bayesian Structural Time Series (BSTS) model to the case ofnowcasting quarterly economic growth in Tanzania, leveraging a range of high-frequency economic indicators. RECOMMENDED CITATION:Lee, Sunwoo, 2026,“Nowcasting Growth Using the Bayesian Structural TimeSeries Model: Application to Tanzania”. IMF Working Paper WP/26/49. WORKING PAPERS Nowcasting Growth UsingtheBayesian Structural Time Series PreparedbySunwoo Lee Contents I.Introduction..................................................................................................................................................3II.Model Components and Features...........................................................................................................4III.Model Inputs and Methodology...............................................................................................................6IV.Application to nowcasting Tanzania’s growth.......................................................................................8V.Extensions...............................................................................................................................................16VI.Conclusion..............................................................................................................................................20Annex.................................................................................................................................................................22References.........................................................................................................................................................23 FIGURESFigure 1.Contribution to predictions by components, different priors of expected model sizek..........................7Figure 2. Tanzania: Real GDP growth (year-over-year, percentage)...................................................................9 TABLESTable 1.Prior and Posterior Distribution of Model Parameters and In-sample Prediction Errors.......................12Table 2. Prior and Posterior Distribution of Model Parameters and In-sample Prediction Errors, Three-period-ahead Forecasts.................................................................................................................................................17 I.Introduction Amid recent global shocks and increased external volatility affecting economies, there is a growing need formonitoring short-term economic fluctuations. However, high-frequency growth data seriesare not readilyavailablein many countries. Official quarterly Gross Domestic Product (GDP) figures are absent in over 60 At the same time, an increasing number of high-frequency economic indicators have become available.New technologies have introduced novel forms of data, expanding the pool of relevant information sources.Furthermore,more countries have begunto collect and publish various statistics relevant to economic Nowcasting economic growth using high-frequency indicators has thus become a popular option tocomplement official statisticsthat are typically published at lower frequencies or with delays. Among thesetools, univariate models, such as the Bridge and Mixed Data Sampling (MIDAS) models, are favored for theirsimplicity and effectiveness in capturing short-term fluctuations witha limited set of indicators(Clements andGalvao (2007); Armesto et al. (2010)). Multivariate approaches, including Vector Auto-regressive Models (VAR)and Dynamic Factor Models (DFM), offer greater complexity by accounting for interdependencies among Introduced by Scott and Varian (2014), theBayesian Structural Time Series (BSTS)modelprovidesanadditional tool for time series nowcasting and forecasting. Itsmodularstate-space framework allows for theflexible incorporation of trend, seasonality, and regressioncomponents. The use of spike-and-slab priors for In developing economies–where data availability may be limited, growth dynamics are evolving, and clearpolicy communication is essential–the model can offer several advantages. It can accommodate sparse andirregular growth data, and its modularity provides a transparent narrative of what drives the nowcast.As new In contextswhere formal nowcasting frameworks are s