Working Papers
Too Costly To Follow Blindly: Endogenous Learning and Herding (2024) [link to paper]
This paper explores the role of social learning on private learning behavior when agents have heterogeneous preferences and both types of learning are costly. I find a non-monotone relationship between social and private learning. They are substitutes when private learning is sufficiently cheap and become complementary for higher private learning costs and eventually becomes uninformative. This happens because increased private learning costs make social learning less informative. Furthermore, herding is an equilibrium only for an intermediate range of values for marginal private learning costs.
Swapping Signals: When and How Learning is Obfuscated (2024) (Joint work with Eric Spurlino) [link to paper]
In this paper, we consider a market where information is readily available but often cognitively costly and sellers can directly affect the learning,i.e., obfuscate information for the buyer in myriad of ways. Using the framework of a one-shot strategic communication game, we model the equilibrium obfuscation behavior of the seller. In our model, the buyer pays a cognitive cost of learning and the seller can garble the posterior belief distribution of the buyer directly. We find that in equilibrium, if the buyer’s
belief is favorable it is optimal to obfuscate fully, however, in case of unfavorable belief an intermediate (or zero) level of obfuscation becomes optimal. We also find that the range of parameters where obfuscation is optimal expands with the cognitive capacity of the buyer. Our framework is agnostic about the form of obfuscation. Furthermore, we use two examples, namely, hiding information and providing misleading information to demonstrate how our model can be used to make testable predictions across different obfuscation practices.
Communication Policy in the Presence of Negative Externality (2022) [link to paper]
This paper studies the optimal disclosure policy of a planner under negative externality. The strategic communication between the planner and the agent is modeled following the information design a la Kamenica and Gentzkow (2011). Additionally, we assume that relevant information is scientific, thus both the planner and the agent have access to the information subject to a cost function following Rational Inattention, a la Caplin, Dean, and Leahy (2018). We show the planner cannot truthfully recommend in the worst state where the cost of making mistakes is the highest. The strategic learning problem can lead to lower welfare generation compared to a non-strategic environment.
Multidimensional and Selective Learning: a case study of Bt cotton farmers in India (2019) [link to paper]
Most production technologies require using an optimal combination of multiple inputs. Farmers need to choose the best combination of seeds, fertilizers, pesticides, etc. to maximize yield. They can learn about the production function by observing the conditional productivity of combinations of input cells or by the marginal productivity of each input across cells average, where both types of learning are costly. I characterize the optimal learning strategy: observing an average is optimal for higher uncertainty and observing a cell is optimal for lower uncertainty. In a sequential learning problem with an optimal stopping time, the optimal learning strategy is to start with observing averages and then switch permanently to observing cells. Depending on the uncertainty of averages, learning about averages only can be optimal, at the cost of a higher probability of error (“selective learning”). Selective learning describes the behavior of Indian cotton farmers when they switched to pest-resistant Bt seeds, as they did not reduce their pesticide use sufficiently. This informs about optimal extension policies (what type of information) for various types of production functions. I also show that the learning mechanism in a laboratory setting predicts the behavior of subjects in the lab.