Chi Wang

Chi Wang, CFA

Ph.D. Candidate in Finance

G. Brint Ryan College of Business  ·  University of North Texas  ·  Expected 2027

I study empirical asset pricing, monetary policy uncertainty, and ETF market structure, with additional interests in corporate finance and ESG investing. My job market paper introduces the Virtual Federal Reserve — a multi-agent LLM simulation calibrated to 1,806 Fed speeches — which recovers the disagreement that institutional consensus suppresses before each FOMC vote. Before my Ph.D., I spent eight years in asset management in China as a portfolio manager specializing in quantitative strategies and structured products. I have held the CFA charter since 2023.

Chi.Wang@unt.edu · +1 (940) 783-6045 · Denton, TX
Download CV Contact

Research

Arbitrage and Monetary Policymaker Disagreement

Working Paper  ·  Presented at SWFA 2026  ·  Submitted to JFQA

Studies how disagreement among Federal Reserve policymakers affects arbitrage in bond ETFs, using FOMC "dot plot" dispersion (2012–2023) as an observable measure of disagreement. Higher near-term dispersion is associated with fewer authorized-participant share creations and wider ETF price–NAV gaps, implying binding limits to arbitrage.

Schematic — qualitative direction only, no fabricated magnitudes
AP Share
Creations
Price–NAV
Deviation
Low Dispersion
AP Share
Creations
Price–NAV
Deviation
High Dispersion
AP Share Creations
Price–NAV Deviation
PDF
This paper examines how disagreement among Federal Reserve policymakers constrains arbitrage in fixed-income ETFs. Using the dispersion of individual FOMC participants' federal funds rate projections from the "dot plot" (2012–2023) as an observable proxy for policymaker disagreement, I find that higher near-term rate-path dispersion is associated with fewer authorized-participant creation transactions and significantly wider ETF price-to-NAV gaps. The results imply that monetary policy disagreement generates information frictions that tighten limits to arbitrage in the bond ETF market.

Measuring Monetary Policy Uncertainty Through an Agentic AI Framework: The Virtual Federal Reserve

Job Market Paper

Working Paper, 2026

A multi-agent LLM simulation of 240 FOMC meetings (1996–2025), one agent per voting member calibrated from 1,806 speeches, predicting individual votes with 91.7% accuracy. Introduces the Agentic AI MPU — a measure of committee disagreement hidden by institutional consensus — that is distinct from existing uncertainty indices.

Agentic AI MPU Index — 240 FOMC meetings, 1996–2025 (real data)
— per-voter prediction
accuracy
— avg suppressed dissent
per meeting
— speech-implied dissents
suppressed to YES
PDF Data
I develop a Virtual Federal Reserve: a multi-agent LLM system that simulates 240 FOMC meetings from January 1996 to December 2025, assigning one calibrated AI agent per voting member trained on 1,806 Fed speeches. The simulation predicts individual member votes with 91.7% accuracy and recovers the latent pre-consensus disagreement — the "true dissent" probability that institutional pressures transform into unanimous votes. The resulting Agentic AI Monetary Policy Uncertainty (MPU) index captures committee disagreement suppressed by herding and is orthogonal to existing text-based uncertainty measures. On average, 39 percentage points of simulated dissent per meeting are absorbed before the official vote, and 97.3% of speech-implied dissents are transformed into YES votes by the consensus process.

ETF Ownership and Idiosyncratic Risk

Working Paper

A one-standard-deviation increase in ETF ownership is associated with lower idiosyncratic volatility and a lower implied cost of equity capital, and is positively associated with R&D spending. Identification uses Russell 2000 index reconstitutions as instruments (2SLS).

Directional findings — schematic, no fabricated magnitudes
↑
ETF Ownership → ↓ Idiosyncratic Volatility
Higher ETF ownership is associated with lower firm-level return volatility (2SLS)
↓
ETF Ownership → ↓ Implied Cost of Equity
Firms with greater ETF ownership face a lower equity risk premium
↑
ETF Ownership → ↑ R&D Spending
ETF-owned firms invest more in research and development
PDF
This paper examines the real effects of ETF ownership on firm risk and investment. Using Russell 2000 index reconstitutions as a quasi-exogenous instrument for ETF ownership (2SLS), I find that a one-standard-deviation increase in ETF ownership is associated with meaningfully lower idiosyncratic stock return volatility and a reduced implied cost of equity capital, alongside higher R&D spending. The results suggest ETF ownership may discipline managers toward long-term value creation by reducing information asymmetry and broadening the shareholder base.

Anomaly Returns at Night and Day

Working Paper

Anomaly premiums shift from daytime to nighttime after academic publication and post-2003. Daytime returns reflect investor mispricing; nighttime returns reflect macroeconomic risk. Increased arbitrage contributes to attenuation in close-to-close anomaly returns.

Premium decomposition — schematic shift from pre- to post-publication
☀️
Daytime
mispricing
🌙
Nighttime
macro risk
After publication & post-2003, arbitrage erodes the daytime mispricing premium; the nighttime macro-risk component becomes dominant.
PDF
This paper decomposes anomaly returns into daytime and overnight components and tracks how the balance shifts after academic publication and post-2003. Before publication, anomaly premiums are concentrated during trading hours, consistent with investor mispricing. After publication, daytime premiums attenuate as sophisticated arbitrageurs trade against the predictability, while overnight premiums — driven by macroeconomic risk exposures — persist or grow. Increased arbitrage activity thus contributes to the observed decay in close-to-close anomaly returns documented in the factor zoo literature.

Data & Resources

Agentic AI MPU Index

240 FOMC meetings  ·  January 1996 – December 2025  ·  Free for research use

Meeting-level Monetary Policy Uncertainty index covering 240 FOMC meetings from January 1996 to December 2025. Each row reports the mean simulated true-dissent probability before institutional herding under four model specifications: a locally-run Qwen 2.5-7B model (agentic_mpu_7b), DeepSeek V4 Flash (agentic_mpu_v4), Claude Sonnet (agentic_mpu_sonnet), and Claude Fable (agentic_mpu_fable). The index captures committee disagreement suppressed by consensus-voting norms and is orthogonal to standard text-based uncertainty measures. Free for research use; please cite Wang (2026).

Download Dataset

Citation: Wang, C. (2026). Measuring Monetary Policy Uncertainty Through an Agentic AI Framework: The Virtual Federal Reserve. Working Paper, University of North Texas.

Teaching

Honors & Awards

Contact Me

I welcome conversations about monetary policy, agentic AI in finance, and ETF market microstructure — as well as collaboration, data, and seminar invitations. Send me a message on the right, or reach me directly using the details below.

Email Chi.Wang@unt.edu
Phone +1 (940) 783-6045
Office G. Brint Ryan College of Business
1307 W. Highland St., Denton, TX 76201
SSRN