Anmerkungen:
Nach Informationen von SSRN wurde die ursprüngliche Fassung des Dokuments September 10, 2022 erstellt
Beschreibung:
We consider competition between firms that design and use algorithms to target consumers. Firms first choose the design of a supervised learning algorithm in terms of the complexity of the model or the number of variables to accommodate. Each firm then appoints a data analyst to estimate demand for multiple consumer segments by running the chosen design of the algorithm. Based on the estimates, each firm devises a targeting policy to maximize estimated profit. The firms face the general trade-off between bias and variance in model selection. We show that competition may induce firms to choose algorithms with more bias leading to simpler (less flexible) algorithmic choice. This implies that complex (more flexible) algorithms such as deep learning that show greater variance in the estimates are more valuable to firms with greater monopoly power