WORKING PAPERS
WITH DANIEL CONUS (DEPARTMENT OF MATHEMATICS, LEHIGH UNIVERSITY)
This research extends Huang and Liu’s (2007) rational-inattention portfolio framework by introducing a stochastic interest rate alongside the original latent appreciation rate, resulting in two latent states and allowing the interaction between information acquisition and portfolio choice to be studied under interest-rate uncertainty. When signal precision and portfolio allocation are jointly optimized, a rationally inattentive investor can attain higher expected terminal utility under interest-rate fluctuations than in the constant-rate benchmark. A preference for short-maturity bonds does not necessarily imply a positive bond position, since the optimal allocation depends on the composition of the portfolio. As interest-rate volatility rises, the dominant force in information acquisition shifts from economizing on information costs toward improving signal accuracy for timely portfolio adjustment.
The framework combines stochastic filtering with continuous-time stochastic control and Hamilton–Jacobi–Bellman analysis. Following Huang and Liu, information arrives continuously and beliefs and portfolio decisions evolve dynamically, while signal precision is chosen once at the initial date. With multiple latent states, the resulting attention-allocation problem can generate corner solutions in which attention is concentrated on either the firm-appreciation signal or the interest-rate signal. These solutions arise naturally from the trade-off between the marginal value and marginal cost of information; they do not appear in Huang and Liu because their model contains only one unknown state.
This point-in-time concentration of attention motivates my work on dynamic signal precision. When precision can adjust over time, instantaneous corner choices may still arise, but their frequencies over the investment horizon generate an effective time-averaged attention allocation, allowing non-corner allocations to emerge intertemporally. This provides a natural bridge from static information design to dynamic learning and portfolio choice under uncertainty.
DYNAMIC SIGNAL PRECISION IN RATIONALLY INATTENTIVE DECISION-MAKING (Job Market Paper)
WITH DANIEL CONUS (DEPARTMENT OF MATHEMATICS, LEHIGH UNIVERSITY)
This research builds on Huang and Liu’s single-state framework, in which the latent state is the firm-appreciation rate, by allowing signal precision to evolve dynamically together with information inflow and portfolio decisions. This dynamic perspective provides an intertemporal interpretation of an interior attention share: although attention may remain concentrated on one signal at any instant, its time-averaged allocation over the investment horizon can be interior. The resulting feedback loop links signal precision and information inflow to belief updating, portfolio choice, and the value function. Newly observed information changes posterior uncertainty and the continuation value of learning, which in turn determine subsequent optimal precision.
Dynamic precision therefore generates an additional forward-looking ex ante effect in the value function that is absent under a one-shot precision choice. This ex ante value gain is complemented by a terminal-wealth result. Dynamic precision need not exceed fixed precision at every date, nor must its posterior variance remain pointwise below its static counterpart. It is sufficient that the dynamic policy generate a weakly lower integrated posterior variance than the static benchmark over the full investment horizon. Under this condition, expected terminal log wealth is weakly higher and the expected terminal-wealth ratio is at least one. The benefit of dynamic precision therefore comes not from maximizing accuracy at every instant, but from allocating costly information acquisition more effectively over time.
Over a fixed investment horizon, the baseline precision level acts as a terminal attractor. Moving variance thresholds classify the information environment into low-, moderate-, and high-risk regimes and generate distinct transient precision dynamics. Optimal precision generally declines as the remaining horizon shortens, although sufficiently large uncertainty shocks can induce delayed intensification. These results extend the analysis of rational inattention beyond the allocation of information quantity across states to the dynamic allocation of information quality, represented by signal precision, across states and over time. The relevant questions are therefore which states to attend to, how precisely to learn them, and when greater precision is most valuable. The framework also provides a foundation for future analysis of heterogeneous signal preferences and strategic interaction through a mean-field-game extension.
Draft available upon request.
WORK IN PROGRESS
SIGNAL PREFERENCE IN MEAN-FIELD GAMES WITH DYNAMIC SIGNAL PRECISION
WITH DANIEL CONUS (DEPARTMENT OF MATHEMATICS, LEHIGH UNIVERSITY)
The motivation for this project is that heterogeneity based solely on observed risk-taking behavior may be misleading. Agents who appear risk-loving may operate in structurally low-risk environments supported by wealth, institutional protection, or other safety buffers, while resource-constrained agents may be genuinely risk-averse yet still face consequential decisions under substantial downside risk. Observed risk-taking therefore need not reveal an underlying preference for risk, but may instead reflect differences in agents’ environments and constraints.
This motivates an alternative classification based on signal preference. Building on the dynamic signal-precision framework, the project considers heterogeneous agents who differ in their tendency to attend to confirmatory versus contradictory signals. Signal preference is interpreted as reflecting prior confidence: relatively confident agents may tilt toward confirmatory information, while agents more responsive to negative or opposing evidence may place greater weight on contradictory signals.
The proposed framework introduces these heterogeneous information types into a mean-field stochastic game in which agents choose signal precision endogenously and interact through the aggregate evolution of beliefs and portfolio behavior. A central conjecture is that confirmatory and contradictory types may follow sharply different belief-updating paths while converging toward similar long-run portfolio behavior under common cost and feedback constraints. One possible mechanism is that skeptical agents front-load information acquisition whereas confirmatory agents defer it, potentially producing comparable cumulative information costs over the decision horizon.
PRESENTATION