Publications

AI availability and U.S. corporate bond markets (with Feng Gao and Chuleng Qiu)
Journal of Financial Markets, forthcoming

We investigate the impact of AI adoption by dealers on U.S. corporate bond liquidity. We use LinkedIn profile data to estimate dealer-specific AI adoption and subsequently construct AI availability at the bond level. Analyzing transaction-level data, we show that AI adoption by bond dealers significantly enhances bond liquidity across various metrics. Specifically, AI adoption improves dealers’ liquidity provision to customer investors and reduces frictions in
the over-the-counter dealer-customer market. The reduction in market frictions further enhances the performance of various multi-factor bond pricing models, and mitigates associated anomalies by eliminating arbitrage opportunities.

Social Optimal Search Intensity in Over-the-Counter Markets
Review of Economic Dynamics, 53 (2024): 224-282.

This paper analyzes OTC market participants' endogeneous search intensity in competitive equilibrium and social optimal cases. We develop a random search-and-match model where agents (market participants) are allowed to choose and adjust their search intensities based on two idiosyncratic states: asset position and liquidity need. We find that: [1] in competitive equilibria with different market parameters, agents can switch between the core and periphery on the trading network. [2] it is the social optimal case that there is no intermediation, in the sense that no agent searches at positive speeds on both the buy and sell sides of the market. In competitive equilibrium, there always exist some agents over-searching and some other agents under-searching. We also discuss related policy implications.

Search Friction, Liquidity Risk and Bond Misallocation
Journal of Financial Markets, Volume 70, September 2024, 100912

Systemic search friction is an important liquidity factor which drives all corporate bonds' yield spread changes. In cross section, bonds have different levels of this yield spread loading. To explain this cross-sectional heterogeneity, we propose a measure of bond-level misallocation among traders, which is defined as the covariance of traders' private valuation and inventory position for each bond. Using transaction-level data, we find that: bonds with a higher level of misallocation have a lower absolute value of yield spread loading on systemic search friction. This relationship is specific to the decentralized market structure, where transactions rely on traders’ searching activity.

Corporate Bond Liquidity During the COVID-19 Crisis (with Mahyar Kargar, Benjamin Lester, David Lindsay, Pierre-Olivier Weill, and Diego Zuniga), Review of Financial Studies (2021), 34(11), 5352–5401

We study liquidity conditions in the corporate bond market since the onset of the COVID-19 pandemic. We find that in mid-March 2020, as selling pressure surged, dealers were wary of accumulating inventory on their balance sheets, perhaps out of concern for violating regulatory requirements. As a result, the cost to investors of trading immediately with a dealer surged. A portion of transactions migrated to a slower, less costly process wherein dealers arranged for trades directly between customers without using their own balance sheet space. Interventions by the Federal Reserve appear to have relaxed balance sheet constraints: soon after they were announced, dealers began absorbing inventory, bid-ask spreads declined, and market liquidity started to improve. Interestingly, liquidity conditions improved for bonds that were eligible for the Fed’s lending/purchase programs and for bonds that were ineligible. Hence, by allowing dealers to unload certain assets from their balance sheet, the Fed’s interventions may have helped dealers to better intermediate a wide variety of assets, including those not directly targeted.

Working papers

Why Voice Trading Persists: The Value of Interdealer Relationships (with Ping He and Shengxing Zhang)

Why does relationship-based voice trade remain dominant in the U.S. corporate bond interdealer market despite lower-friction electronic alternatives, and is this market structure efficient? Using transaction-level data, we show that dealers increasingly choose voice venue when inventory imbalances make immediacy valuable, because voice venue provides access to established trading relationships. Voice execution is slower than electronic execution for micro trades but faster for orders of at least $100,000, which account for nearly all interdealer par volume. We develop and calibrate a searchand-matching model where dealers choose trading venues contingent on their inventory balances. Voice venue offers two advantages: on average dealers meet counterparties faster, and because dealers with large imbalances sort into the voice venue, they meet better counterparties, whose positions are more likely to offset their own. The endogenous second channel accounts for roughly three quarters of the voice venue's advantage in the calibrated model. In equilibrium, voice venues are valuable but inefficiently allocated and overused by dealers with limited need for immediacy. A social planner retains the most imbalanced dealers in voice but reallocates near-target dealers to electronic trading, reducing the share of dealers searching in voice from 72% to 25%.

Interdealer Price Dispersion and Intermediation Capacity (with Andrea L. Eisfeldt and Bernard Herskovic)

Intermediation capacity varies across dealers and, as a result, misallocation of credit risk reduces the risk-bearing capacity of the dealer sector and increases effective market-level risk aversion. When the efficient reallocation of credit risk within the dealer sector is impaired, interdealer price dispersion increases. Empirically, interdealer price dispersion is a strong determinant of yield spread changes. When interdealer price dispersion is high, bond prices are low. Interdealer price dispersion explains a substantial portion of bond yield spread changes, the cross-section of bond returns, and the basis between yield spread changes and changes in fair-value spreads. We conclude that frictions within the dealer sector reduce the risk-bearing capacity of intermediaries and are thus crucial for intermediary bond pricing.

Dealers' Search Intensity in U.S. Corporate Bond Markets

Are dealers' search efforts endogeneous in decentralized markets? How do dealers' search efforts affect market efficiency? We propose a model with dealers choosing idiosyncratic search intensities, and estimate the model using transaction data on U.S. corporate bonds. We find that: [1] with dealers ranked by their private valuations for a bond, the middle-type dealer chooses the highest search intensity, and she reallocates bond positions from lower-type dealers to higher-type dealers; [2] the estimated model predicts that the search costs and bond misallocation in current OTC markets generate 13.7% welfare loss relative to a counterfactual frictionless market.

Venue Participation and Transaction Cost: Evidence from China Government Bonds Market (with Jiaer Liu)

This paper examines trading activity and transaction costs differences across the Over-the-Counter (OTC) and Central Limit Order Book (CLOB) venues in the China government bonds market. Using novel trade-level data, we estimate that CLOB lowers transaction costs by 0.66 basis points relative to OTC, highlighting efficiency gains from the centralized trading mechanism. We also document selection in venue participation: CLOB disproportionately facilitates core traders, standardized orders, and newly issued bonds, indicating potential redistributional effects. Despite CLOB's cost advantage, OTC remains active due to complementary benefits, including mitigating information leakage, designating counterparties, and facilitating holding position rebalancing.

Underwriting as Network Formation: Evidence from Municipal Bond Auctions (with Jiawei Dai, Manyi Wang and Xiang Yin)

This paper studies how municipal bond dealers use primary market participation to expand their over-the-counter (OTC) trading networks. Linking auction data to secondary market transactions involving insurance company clients, we find that dealers bid more aggressively in primary auctions when the issuer’s client base includes prospective secondary market counterparties. Winning dealers are significantly more likely to initiate trades with these clients, particularly those with no prior trading relationship. These relationships persist over time and spill over into other securities. The findings highlight a novel channel through which primary market activity facilitates endogenous network formation in OTC markets, with implications for underwriting spreads and liquidity provision.