The empirical literature on common international business cycles has largely ignored model misspecification in estimated factor models as the various cycles are typically imposed but not tested for. This paper proposes a Bayesian stochastic factor selection approach for multilevel factor models. The procedure is applied to a three-level dynamic factor model with a global factor, six regional factors and three development level factors. We estimate the factor model using real GDP growth data for a panel of 60 countries over the period $1961-2017$. We find robust evidence for the presence of a global business cycle, four regional cycles (Europe, North America, Latin America and Asia) and two development level cycles (industrial countries and emerging market economies). This suggests that both geographical proximity and the development level of countries are important dimensions of international business cycle synchronization that should be considered simultaneously, a point not previously made in the existing synchronization literature.

关于常见国际商业周期的实证文献在很大程度上忽略了估计因子模型中的模型错误指定，因为通常会施加各种周期但未进行测试。本文提出了一种多级因子模型的贝叶斯随机因子选择方法。该程序应用于具有全球因素、六个区域因素和三个发展水平因素的三级动态因素模型。我们使用同期 60 个国家/地区的实际 GDP 增长数据估计因子模型$\mathrm{1961年}-\mathrm{2017年}$. 我们发现了全球商业周期、四个区域周期（欧洲、北美、拉丁美洲和亚洲）和两个发展水平周期（工业国家和新兴市场经济体）存在的有力证据。这表明，地理邻近性和国家发展水平都是国际商业周期同步的重要维度，应同时考虑，这一点以前在现有同步文献中没有提到。