In this study, we use a long panel dataset of Ethiopian manufacturing firms for the period 1996-2009 to test certain hypotheses regarding firm performance and exporting, namely export hysteresis, self-selection, and learning-by-exporting. We find evidence for export hysteresis and the presence of high sunk costs associated with exporting. However, the magnitude of this effect has become lower when controlling for unobserved firm heterogeneity using a dynamic random effect probit model suggesting that not addressing the unobserved heterogeneity can lead to an overestimation of the coefficient of the sunk costs. Moreover, we find no evidence of self-selection into exporting when controlling unobserved firm heterogeneity and unlike to most previous studies. On the other hand, we find strong robust evidence in support of the learning-by-exporting. Our results support the view that export promotion efforts might have long-term effects in terms of sustaining exports and industrial competitiveness.