Effects of richer feedback and endogenous stopping/resubmission on optimal search
Investigate how allowing richer feedback and state-transition processes (beyond the simple probability q_i that a low-quality paper becomes high quality upon rejection) and relaxing the exogenous stopping and no-resubmission assumptions affect the optimal submission strategy in the sequential search without recall model. Determine the resulting optimal policies under these modifications and characterize how they alter the search order and outcomes relative to the baseline model analyzed in the paper.
References
In addition, the type of feedback, modelled simply as a potential state transition, is also restrictive: future work should allow for richer information and state transition probabilities. Finally, relaxing the exogenous stopping and no-resubmission policies, while potentially implausible when studying journal-submission decisions, may be useful in understanding other sequential search settings such as the job market, and help better compare the current framework with multi-armed bandit models. The effects these modifications have on optimal search remain open and a fruitful area for future research.
Several theoretical and applied extensions remain open. First, extending the framework to continuous payoff distributions or richer multidimensional state spaces requires moving beyond scalar single-crossing conditions. In such environments, the learning effect becomes multidimensional, and preserving monotonicity will likely demand strict shape restrictions on the posterior predictive distributions, such as log-concavity or the Monotone Likelihood Ratio Property (MLRP).
Several theoretical and applied extensions remain open. First, extending the framework to continuous payoff distributions or richer multidimensional state spaces requires moving beyond scalar single-crossing conditions. In such environments, the learning effect becomes multidimensional, and preserving monotonicity will likely demand strict shape restrictions on the posterior predictive distributions, such as log-concavity or the Monotone Likelihood Ratio Property (MLRP). Second, the information structure could be endogenized within a formal mechanism design or Bayesian persuasion framework. Rather than assuming an exogenous censoring rule, future research could analyze how a principal, such as a hiring platform or a market intermediary, optimally designs the censoring threshold to maximize the searcher's engagement or the aggregate match quality. Finally, the theoretical separation between pathwise and expected monotonicity provides a structural foundation for empirical work. The model generates precise, testable implications regarding how DMs ought to adjust their reservation cutoffs following coarse failures, such as uninformative job-market rejections or opaque price screening. Testing these theoretical bounds against empirical belief-updating behavior will deepen our understanding of when informational coarsening serves strictly to bound learning, and when it acts as a necessary condition for the tractability of sequential choice.