Credit card lenders individualize contracts primarily through risk-based credit limits rather than interest rates. To understand lenders’ credit limit choices, I use novel statement-level data on the near-universe of UK credit cards active between 2010–2015 to estimate a structural model of the credit card market. The model explains differences in the shape of lenders’ credit limit distributions through a screening technology that provides lenders with a noisy signal of customers’ risk. I identify model parameters using a novel cost shock that results from the April 2011 case in the England and Wales High Court concerning the mis-selling of payment protection insurance. I use the estimated model to evaluate a counterfactual scenario in which lenders can freely individualize interest rates and credit limits, which the existing environment precludes. As a result, individualized interest rates and credit limits emerge, profits increase, and borrowing becomes more dispersed as a result.
We develop and estimate a dynamic structural model of the patent screening process. The model incorporates incentives, intrinsic motivation and bargaining structure. We estimate the model using novel negotiation-round-level data on examiner decisions and text data from 24 million patent claims. From the claim text data, we use modern natural language processing methods to develop a new measure of patent distance. Our model estimates imply substantial variation in examiners’ intrinsic motivation relative to examiners’ time costs, with senior examiners less intrinsically motivated than juniors on average. With the estimated model, we calculate changes to timeliness and examination quality resulting from changes to agents’ incentives and the bargaining structure. We find that a reduction in the number of negotiation rounds would improve both timeliness and quality of the patent screening process.
We introduce multivariate ordered discrete response models that exhibit non-lattice structures. From the perspective of behavioral economics, these models correspond to broad bracketing in decision making, whereas lattice models, which researchers typically estimate in practice, correspond to narrow bracketing. There is also a class of hierarchical models, which nests lattice models. A special case of non-lattice models, hierarchical models correspond to sequential decision making and can be represented as binary decision trees. In each of these three cases, we specify latent processes as a sum of an index of covariates and an unobserved error, with unobservables for different latent processes potentially correlated. This additional dependence further complicates the identification of model parameters in non-lattice models. We provide conditions sufficient to guarantee identification under the independence of errors and covariates, compare these to identification conditions in lattice models, and outline an estimation approach. Finally, we provide simulations and empirical examples, with particular focus on probit specifications.
Other Work in Progress
On the Validity of Leniency Instruments, with Mark Schankerman