This paper compares backcasting performance of models based on variable selection to dynamic factor model for backcasting the world-trade growth rate with two months ahead. The variable selection models are specified by applying penalized regressions and an automatic general-to-specific procedure, using a large dataset. A recursive forecast study is carried out to assess the backcasting performance by distinguishing crisis and non-crisis periods. The results show that, some selection-based models exhibit a good backcasting performance during both periods. The more accurate backcasts seem to be SCAD, adaptive Elastic-Net and adaptive SCAD during the GFC and Covid-19 crisis whereas it seems rather Lasso, Elastic-Net, adaptive Lasso and DFM during the non-crisis period. Among the predictors for backcasting world trade growth it appears that the index of global economic conditions proposed by Baumeister et al. (2020), the PMI indicator on new export orders in manufacturing sector ans the MSCI world index are relevant.