Article dans une revue: We study the instability of hyper-specialization of exports at the 4-digit level in 1998–2010. (1) Specializations are surprisingly un-stable. Export ranks are not persistent, and new top products and destinations replace old ones. Measurement error is unlikely to be the main or only determinant of this pattern. (2) Source country factors are not the main explanation of this instability. Only 16–20% of variation in export growth is accounted for by source country plus source country-product factors that do not vary across destinations. The high share of idiosyncratic variance (source-product-destination residual) of 41–55%, indicates the difficulty to predict export success using source country characteristics. While we are cautious in interpreting factors that are jointly determined in global general equilibrium, our results suggest that destination and product-specific factors importantly matter at least as much as source country factors.