DAVID
WALKER-JONES
Welcome! I am a lecturer with the School of Economics at the University of Surrey and I specialize in micro theory and experimental economics.
Published Papers
This paper studies a new measure for the cost of learning that allows the different attributes of the options faced by an agent to differ in their associated learning costs. The new measure maintains the tractability of Shannon’s classic measure but produces richer choice predictions and identifies a new form of informational bias significant for welfare and counterfactual analysis that is conducted with the multinomial logit model. Necessary and sufficient conditions are provided for optimal agent behavior under the new measure for the cost of learning.
Working Papers
A novel data enrichment demonstrates that experiment subjects are more likely to invest effort into learning about the value of options if simple choice parameters, like price, differ from previous choice problems. This increase in effort in ‘unfamiliar’ choice problems means that the behavior of many subjects violate even the most flexible model of costly learning if the cost for information is assumed to be constant across choice problems with the same prior beliefs. This observation motivates the introduction of heterogeneous decision makers into a standard and more restrictive (posterior separable) model of costly learning to better fit the data.
By weakening Shannon’s original axioms to allow for attributes of the choice environment to differ in their associated learning costs, this paper provides an axiomatic foundation for Multi-Attribute Shannon Entropy, a natural multi-parameter generalization of Shannon Entropy. Sufficient conditions are also provided for a simple dataset that identifies the Multi-Attribute Shannon Entropy cost function for information by analysing stochastic choice data produced by a rationally inattentive agent that is picking between pairs of options when relatively few states of the world have a positive probability of being realized.
(with Yoram Halevy and Lanny Zrill)
We investigate the problem of identifying incomplete preferences in the domain of uncertainty by proposing an incentive-compatible mechanism that bounds the behavior that can be rationalized by very general classes of complete preferences. Hence, choices that do not abide by the bounds indicate that the decision maker cannot rank the alternatives. Data collected from an experiment that implements the proposed mechanism indicates that when choices cannot be rationalized by Subjective Expected Utility they are usually incompatible with general models of complete preferences. Moreover, behavior that is indicative of incomplete preferences is empirically associated with deliberate randomization.
(with Umberto Garfagnini)
This paper introduces the Naive Frequentist model of belief updating, in which a decision maker updates their beliefs by taking simple averages of the numeric values in their most recent pieces of information, and tests the predictions of the model with an experiment. The model provides a parameter free prediction of posterior beliefs in a wide range of updating problems, does not require knowledge of the decision maker’s prior or their perception of the accuracy of information, and in many situations predicts large updates that are in the opposite direction of the change predicted by Bayes’ rule. Our experiment features pieces of repeated, redundant, and inaccurate information, that should all be ignored by a Bayesian, and yet our model predicts should change beliefs in a specific way. Even though our setting is a simple one where Bayesian updating is normally as easy as averaging two integers, we observe large and systematic deviations from the predictions of the Bayesian model that are roughly in line with the predictions of the Naive Frequentist model, and predict simple contexts in another dataset on more standard “ball and urn” updating where a large majority of subjects update in the wrong direction.
Work in Progress
Mechanism Design with Endogenous Information Dissemination
(with Henrique Castro-Pires and Krittanai Laohakunakorn)
We investigate the tradeoffs associated with giving additional commitment power to a biased principal.
Ambiguous Political Platforms and Behavioural Voters
This paper explores a behavioral model of how voters pick a favorite political candidate, and the implications of the model for equilibria in a two candidate election that sees candidates choosing their public platform by selecting the dimensions of their planned policy they wish to disclose. Even if the behavioral voters are ambiguity averse, not declaring a position in certain dimensions can benefit a candidate by focusing the attention of their target voters on dimensions in which the target voters have more homogenous preferences.
Competing Firms and Obfuscation
This paper explores a Bertrand style duopoly environment in which a priori identical firms can choose both the price for their product and how costly it is for prospective buys to observe their realized heterogenous value for the firm's product. In such a setting, competition is not sufficient for achieving prices that are lower than the monopoly price.
Identifying Preferences Using Reaction Times
A theory paper that holds learning costs, as identified by reaction times, constant by using the non-monotonic relationship between the price of a good with uncertain value and the incentive to learn. Holding learning costs constant allows for the identification of preferences.
Contact
