Theory and Estimation of Dynamic Factor Models for Macroeconomic Forecasting in Commodity-Dependent Developing Economies: Evidence from Nigeria

Macroeconomic forecasting in commodity-dependent developing economies such as Nigeria requires factor models capable of extracting latent business cycle signals from large panels of economic indicators while accommodating structural instability induced by commodity price cycles, exchange rate volatility, and fiscal policy reversals. This dissertation develops the theory and estimation of dynamic factor models (DFMs) for macroeconomic forecasting in commodity-dependent developing economies, with Nigeria as the primary empirical application. New identification theory for DFMs under time-varying factor loadings is developed, providing conditions under which factor extraction remains consistent during structural breaks. A novel commodity-augmented DFM (CA-DFM) incorporating crude oil price factors with time-varying factor loadings is introduced and its theoretical properties derived. Estimation employs a sequential Kalman filter-EM algorithm with structural break detection. The CA-DFM is estimated on a panel of 82 Nigerian macroeconomic and financial monthly series from 2005 to 2023. Out-of-sample forecasting experiments show CA-DFM achieves 28 percent lower RMSE for GDP growth forecasting than standard DFM without commodity augmentation. The CA-DFM correctly predicted the 2015 to 2016 recession one quarter earlier than competing models. Oil price shock transmission to GDP growth exhibits time-varying factor loadings with significantly larger transmission in post-2020 periods. The dissertation provides methodological contributions relevant to central bank and ministry of finance forecasting units in commodity-exporting developing economies. Keywords: dynamic factor model, macroeconomic forecasting, commodity dependence, Nigeria, time-varying loading