Francesco Bloise (Sapienza University of Rome)
Francesco Giuli, Edoardo Santoni and Margherita Scarlato
Whether the informal wage gap reflects labor market segmentation or competitive sorting on productivity remains a central question in development economics, with direct implications for policy design. Existing evidence is largely based on average wage comparisons that may conceal economically meaningful heterogeneity. We revisit this debate by estimating the full distribution of individual-level informal wage penalties in urban South Africa. Using nationally representative data from the National Income Dynamics Study and a causal machine learning frameworkdouble machine learning with doubly robust scoreswe flexibly control for a high-dimensional set of worker characteristics and recover individualized treatment effects without pre-specifying subgroups. We find an average informal wage penalty of approximately 21%, consistent with the existing literature on developing economies. However, this mean masks substantial heterogeneity: individual penalties range from near zero to almost 50% at the 1st percentile, while a small fraction of workers faces no penalty or even a mild premium. Crucially, the dimensions along which heterogeneity is most pronounced are identified endogenously by the machine learning algorithm rather than imposed ex ante by the researcher. A classification analysis reveals that the largest penalties concentrate among workers belonging to the African racial group and speakers of minority languages, while White and Coloured workers, Afrikaans speakers, and Western Cape residents experience comparatively smaller gaps. These patterns are difficult to reconcile with a frictionless competitive model and point toward search frictions, unequal outside options, and employer market power as plausiblethough not uniquely identifiedmechanisms shaping wage inequality across both formal and informal sectors.