David Xiaoyu Xu

Despite the secular shift to nonbank lending, banks retain a distinctive advantage in providing credit lines (Kashyap, Rajan, and Stein, 2002). We document a new pattern in nonbank lending: business development companies (BDCs) extend substantial credit commitments to borrowers, with commitment-to-asset ratios comparable to those of banks. These off-balance-sheet commitments are concentrated, exposing BDCs to undiversified borrower liquidity shocks—a risk they manage with credit lines provided by banks. We develop a model of layered liquidity insurance along the credit chain: the BDC pools its portfolio firms to partially diversify idiosyncratic shocks and uses a bank credit line to backstop residual liquidity risk. In equilibrium, the nonbank optimally provides only partial liquidity insurance, which generates a novel externality whereby firms sharing the constrained liquidity pool inefficiently underinvest in liquidity management.

Loan sales weaken banks' incentives to produce information about borrowers, but may create incentives for outside investors to produce such information. In the leveraged loan market, bookbuilding lets banks learn from investors, so loan sales need not reduce information production overall. Lead banks observe nonbank investor demand during syndication and can revise loan spreads before issuance. We show that upward adjustments in spread predict higher subsequent borrower default and upward revisions in banks' internal estimates of default risk. These patterns indicate that investor demand can reveal adverse borrower information not fully reflected in banks' initial terms. Overall, the evidence suggests that loan sales can induce information production by investors, while bookbuilding transmits that information back to lead banks and incorporates it into loan prices and bank beliefs.

This paper examines the growth and performance of AI-driven investing in asset management. Using investment advisers' regulatory disclosures, labor market data, and fund strategy descriptions, we document that AI-driven investing has grown steadily since the early 2010s and is concentrated among hedge funds, particularly those employing systematic diversified macro strategies. AI funds outperformed non-AI funds in the early years, but this outperformance declined over time along with the growth of AI-driven investing, even among early adopters. Contrary to concerns about AI-driven strategy homogeneity, AI funds exhibit lower return comovement than non-AI peers. Our findings highlight both the alpha-generating potential and the limitations of AI as a source of investment performance.

This paper investigates how the spatial distribution of financial intermediaries' human capital affects credit supply. Focusing on residential mortgage markets, we find that shocks to mortgage demand are not met with changes in the number of local loan officers, but instead affect workloads and remote lending. Importantly, evidence from shocks to the supply of loan officers shows these responses are not perfect substitutes for local intermediation. Local loan officers improve the availability of mortgage credit, the efficiency of the application process, and households' refinancing outcomes. Our findings suggest that labor market frictions can distort local credit supply.

We study the contribution of AI-generated reference prices to intraday price discovery in markets with infrequent trading. Using corporate bond transactions and MarketAxess CP+ quotes, we find that CP+ is more informative about future trade prices than the last trade. Regression analysis shows that CP+ quote updates are systematically related to market-wide movements in bond, equity, and options markets, as well as bond-specific non-public information from the RFQ process. CP+ provides broad coverage across bonds and trading days. Its contribution to price discovery exhibits a bell-shaped relationship with liquidity and increases under market uncertainty. Following a trade report, CP+ updates quickly in the direction of the trade. We show that this can limit its contribution during periods driven by large transitory price shocks.

Mutual funds create liquidity by issuing demandable equity shares while holding illiquid securities. We study this liquidity creation using frequent trading suspensions in China, which temporarily eliminate market liquidity in affected stocks. These suspensions cause significant mutual fund mispricing due to inaccurate valuations of illiquid holdings. We find that investors actively acquire information about suspended stocks held by mutual funds, driving flows into underpriced funds. This information is subsequently incorporated into stock prices when trading resumes. Our findings suggest that liquidity creation by nonbank intermediaries, such as mutual funds and semi-liquid private funds, stimulates information acquisition about illiquid assets.

I study dynamic portfolio rebalancing in Collateralized Loan Obligations (CLOs) by developing an industry equilibrium model of nonbank lending, in which CLOs and loan funds arise endogenously in response to a premium for safe securities. When loans deteriorate after issuance, CLOs rebalance their portfolios to maintain collateral quality, which protects senior tranches at the expense of equity investors. This "self-healing" mechanism lowers CLOs' ex-ante funding costs by enabling the issuance of larger safe tranches. As more lenders operate CLOs, their portfolio rebalancing generates greater non-fundamental price pressures, incentivizing other lenders to operate loan funds. Overall, portfolio dynamics facilitate risk sharing across nonbank lenders and increase both total lending and the supply of safe securities relative to static portfolios.

Using detailed data on company visits by Chinese mutual funds, we provide direct evidence of mutual fund information acquisition activities and the consequent informational advantages mutual funds establish in local firms. Mutual funds are more likely to visit local and nearby firms both in and outside of their portfolios, but the ease of travel between fund and firm locations can substantially alleviate geographic distance constraints. Company visits by mutual funds are strongly associated with both fund trading activities and fund trading performance. Our results show that geographic constraints and costly information acquisition amplify information asymmetry in financial markets.