A Structural Model of Correlated Learning and Late-Mover Advantages: The Case of Statins
49 Pages Posted: 19 Sep 2015 Last revised: 25 Oct 2016
Date Written: October 18, 2016
When Lipitor entered the statin market in 1997, some incumbent drugs had already obtained strong clinical evidence to show their efficacy in preventing heart diseases. Although it lacked such important evidence, Lipitor quickly became the most popular statin. To explain this puzzle, we propose a theory of correlated learning and indirect inference. We introduce a concept of “efficiency ratio,” which measures how efficiently a drug can convert reduction in the bad cholesterol to reduction in heart disease risks. We assume physicians are uncertain about drugs’ efficiency ratios, and their initial prior belief could be correlated across drugs. Hence, a new clinical trial information signal on a drug’s efficiency ratio can update physicians’ belief on other statins’ efficiency ratios. Physicians then infer each statin’s ability in reducing heart disease risks based on its perceived efficiency ratio and its ability in reducing the bad cholesterol. Consequently, correlated learning may allow late entrants to gain late-mover advantages by free-riding on the clinical evidence and informative detailing of incumbents. To estimate our model, we use data on market shares, switching and discontinuing rates, detailing expenditure, clinical trials and media coverage from 1993 to 2004. Our results show that correlated learning is strong. Moreover, because the two late entrants (Lipitor and Crestor) are more effective in reducing the bad cholesterol, correlated learning and indirect inference allow them to gain late-mover advantages and grow much faster in the actual world compared with a counterfactual world where there is no correlated learning.
Keywords: Correlated Learning, Late Mover Advantage, Swiching Costs, Clinical Trials, Detailing
JEL Classification: D12, I11, L65, M30, M31, M37
Suggested Citation: Suggested Citation