This dissertation comprises three chapters within the fields of Econometrics of Social Networks and Productivity Estimation.

The first chapter studies the estimation of demand with social network effects, where individuals' product choices are affected by the choices of their friends. I construct a sequential game choice model that captures the impact of large-scale network effects on consumers' product choices and propose a tractable Bayesian MCMC estimation procedure. Using granular data from the Steam online video game platform, I estimate the impact of friends' choices on individual gaming decisions. My findings reveal a positive network effect on game choices, with multiplayer games exhibiting a significantly larger effect. The empirical results highlight the importance of considering network effects in demand estimation, as failure to do so can result in an overestimation of the coefficients associated with game characteristics. This suggests the presence of demand amplification effects arising from social networks. Lastly, I conduct counterfactual analyses to examine demand elasticity and explore pricing and marketing strategies leveraging network effects.

The second chapter is a collaborative work that estimates firms’ production functions using a novel correlated random coefficient methodology. To incorporate firms’ endogenous decisions on input choices, we specify firm-specific and time-varying productivity parameters while allowing arbitrary correlations between productivity parameters and input variables. Applying the methodology to panel data from manufacturing industries in Chile and India, we estimate the average and dispersion of productivity across firms and use the estimates to decompose the aggregated productivity growth.

In the third chapter, I explore a specialized case related to the first chapter, where agents possess private information on links and payoffs. To address this scenario, I propose a two-stage model. The first stage involves network formation, where agents follow an incentive-compatible mechanism to form network links. In the second stage, given the endogenously formed links, agents engage in a continuous-action network game with incomplete information. To jointly estimate the two stages and account for the endogeneity of links, I utilize a latent variable and adopt a tractable Bayesian MCMC approach.