Collusion Along the Learning Curve: Theory and Evidence from the Semiconductor Industry
This paper formulates a theory of collusion with learning-by-doing and multiproduct competition and tests it with data from an explicit cartel. The model shows that collusion is harder to sustain on a new product generation, where learning is high, than an old generation, where learning is low. Collusion on the old generation shifts demand toward the new generation, raising its output. Empirical analysis exploits variation between cartelization and competition in the DRAM market to identify counterfactual quantities and prices. Consistent with the model, cartel firms cut output of older generations by up to 50% and increased output of newer generations manifold.
JEL Classification: D43, L13, L41, L63