Research

Job Market paper:

Non-parametrically estimating preferences in presence of strategic reports: Application to college admissions

(With Kala Krishna and Sergey Lychagin). We propose a new approach to estimate the distribution of cardinal preferences for university programs from strategic reports. As applicants misrepresent preferences, the primary challenge is obtaining an estimate of the true preferences from the stated ones. To do this, we propose an estimation strategy based on the revealed preference approach.

We assume that each applicant has one of a finite set of preferences and, unlike the usual Logit approach, there is no random element to preferences. Thus, each applicant belongs to a preference group all members of which have exactly the same preferences. We estimate the cardinal utilities associated with each preference group along with its prevalence in the population, as well as the costs of adding to the list and the outside option that vary by applicant.

We are able to predict the equilibrium effects of policy changes to the Turkish university placement mechanism. We predict that an affirmative action policy that gives low-income applicants a one standard deviation benefit in terms of the acceptance cutoffs improves their expected utility by 69%, which is less than a third of what using a standard Logit setup would predict.

Our approach is flexible and can be scaled to very large problems because we re-cast the estimation problem as a sequence of linear programs, as opposed to a large scale non-linear optimization problem.

The latest version is available here: Job Market Paper (please let me know if the link does not work: rji5040@psu.edu)

Numerical algorithm for inverting the Pure Characteristics model of demand

I propose a method for inversion of market shares in the Pure Characteristics Model (PCM) of demand based on Quasi-Newton's method. The method uses the fact that the market shares function in PCM is piece-wise differentiable, and solves the system of non-linear equations using the Knitro solver with supplied Jacobian.

A series of MC-simulations finds that the algorithm inverts market shares for small and large number of products with high accuracy (1e-12). An algorithm to calculate a good initial guess with a large number of products is also provided.

I also show that using the quadrature integration rules is valid in this setting. I believe this set of algorithms may be useful in empirical work and be incorporated into estimation procedures that estimate structural models with random coefficients based on the Pure Characteristics Model.

Temporary draft: Pure characteristics model Inversion

The rest of my papers can be found in my CV.