Portfolio Selection Under Rational Inattention with Varying Interest Rates

JOB MARKET PAPER --- UNDER REVIEW AT THE JOURNAL OF ECONOMIC DYNAMICS & CONTROL

WITH DANIEL CONUS (DEPARTMENT OF MATHEMATICS, LEHIGH UNIVERSITY)

This paper extends Huang and Liu (2007) by introducing a stochastic interest-rate process to examine whether a rationally inattentive investor can achieve higher expected utility compared to a constant-rate benchmark. To ensure a valid counterfactual, we focus on Vasicek model that shares the same initial level as the constant-rate benchmark and display no deterministic trend, while the role of trending rates is treated separately. In this framework, a Bayesian investor chooses information precision once at the initial decision time—a static specification relevant for settings such as IPO participation, first-time entry, or infrequent investments—and subsequently makes dynamic portfolio allocations. Using linear filtering methods, the investor infers in real time both the latent appreciation rate and the mean-reverting interest rate from returns on one stock, one bond, and news signals.

The paper shows that stock demand depends on appreciation and price dynamics—defensive, growth, or high-volatility—while bond allocations tilt toward short maturities, though investors may optimally short bonds when risky assets offer strong appreciation potential. Analogous to Markowitz’s mean–variance setting where risk can enhance utility depending on portfolio weights, here interest-rate volatility can raise utility depending on attention allocation and portfolio structure. Rising volatility shifts investors from economizing on information toward sharpening precision, and the model produces corner solutions whenever the marginal utility of a hidden state exceeds its cost. Finally, replacing the constant-rate path with a trend-aware process yields convergence in strategies and wealth trajectories, with the revised value function aligning more closely with the stochastic model and highlighting the importance of trend dynamics.

INTEREST RATE TREND VERSUS CONSTANT RATE: DYNAMIC SIGNAL PRECISION AND PORTFOLIO CHOICE UNDER RATIONAL INATTENTION

WITH DANIEL CONUS (DEPARTMENT OF MATHEMATICS, LEHIGH UNIVERSITY)

Previous dynamic RI research has focused on dynamic signal accumulation (Zhong 2022; Hébert and Woodford 2017; Fudenberg et al. 2018), fully dynamic signal flows (Steiner et al., 2017; Afrouzi and Yang 2019; Miao and Xing 2023), and endogenous precision in static settings (Huang and Liu 2007). This ongoing paper advances this frontier by integrating dynamic signal precision with dynamic portfolio control, bridging information design and stochastic decision-making. We develop a generalized model in which agents dynamically adjust precision in response to evolving information costs and time preferences. Discounted terminal utility generates intertemporal trade-offs, yielding adaptive attention choices. This framework uncovers a recursive Learning–Decision–Value loop, in which precision shapes belief updating, portfolio allocation, and subsequent signal accuracy. The loop introduces a novel dynamic effect—accuracy gains net of cognitive adaptability costs—absent from static benchmarks (JMP). Our key contribution is to move beyond Shannon’s cost view of RI by incorporating feedback-driven attention costs, thereby unifying cognitive effort, signal learning, and stochastic control in a single framework. Applied to monetary environments with uncertain interest-rate trends, embedding dynamic precision into both the constant-rate benchmark (Huang and Liu) and our varying-rate model (JMP) refines portfolio strategies and allows investors to flexibly adapt as beliefs evolve.

HETEROGENEOUS SIGNAL PREFERENCES IN DYNAMIC PORTFOLIO CHOICE UNDER RATIONAL INATTENTION

WITH DANIEL CONUS (DEPARTMENT OF MATHEMATICS, LEHIGH UNIVERSITY)

This project develops a theoretical framework modeling cognitive preferences over information signals—confirmatory versus contradictory—as endogenous reflections of prior confidence. Studies of rational inattention show that confirmation bias can arise from information frictions under homogeneous cognition (Nimark and Sundaresan 2019; Catonini and Mayskaya 2024; Miao and Xing 2024), and that ambiguity aversion reshapes attention under belief uncertainty (Hansen, Miao, and Xing 2025). We extend these insights by linking prior confidence to heterogeneous signal selection: confident agents attend to confirmatory signals, while ambiguity-averse agents seek contradictory ones. We formalize this in a two-agent stochastic control framework, building on the dynamic signal-precision model of WIP1, where belief-persevering and counterfactual-reasoning investors co-evolve. Confirmatory types reinforce priors until major shocks trigger discrete corrections, whereas contradictory types revise beliefs episodically in response to skeptical signals. Despite these contrasting dynamics, we conjecture that both types may ultimately converge toward similar long-run portfolio strategies, reflecting bounded rationality and adaptive survival. We also conjecture that this convergence stems from a symmetry in cumulative information costs: skeptical agents front-load epistemic effort, while confirmatory types defer it, yet total cognitive investment equilibrates across the decision horizon.