Job Market Paper

Common Investment Advisors for Insurance Companies: Implications for Liquidity and Systemic Risk
Abstract: Life insurers —the largest investors in corporate bonds— outsource nearly $1 trillion in bond investments to a few common external advisers, raising concern for systemic risk through similar portfolio strategies. Investors with the same advisor benefit from cost savings while exhibiting higher portfolio similarity. Portfolio similarity is three times higher between insurer pairs with the same adviser, and 1.5 times higher between insurer-mutual fund pairs sharing an adviser. This interconnectedness has two major implications for systemic risk. On one hand, portfolio similarity amplifies fire sale risks when these institutions face common shocks, such as during monetary tightening cycles. On the other hand, when institutions face divergent shocks, portfolio similarity enhances financial stability. During mutual fund outflows, insurers purchasing bonds sold by associated mutual funds provide a stabilizing effect on market prices.

Working Papers

• Behavioral Economics of AI: LLM Biases and Corrections - with Lin William Cong, Xing Huang, and Lawrence Jin
  Abstract: Do generative AI models, as epitomized and popularized by large language models (LLMs), exhibit systematic behavioral biases in economic and financial decisions? If so, how can we mitigate these biases? Following the cognitive psychology literature and the experimental economics studies, we conduct the most comprehensive set of experiments to date—originally designed to document human biases—on prominent LLM families with variations in model version and scale. We document systematic patterns in the behavioral biases that LLMs exhibit. For experiments concerning the psycholoogy of preferences, LLM responses become increasingly irrational and human-like as the models become more advanced or larger; however, for experiments concerning the psychology of beliefs, the most advanced large-scale models frequently generate rational responses. Further exploring various methods for correcting these behavioral biases re- veals that prompting LLMs to make rational decisions according to the Expected Utility framework seems the most effective.

• Prediction Risks of Asset Pricing via Machine Learning - with Lin William Cong
  (draft coming soon)
  Abstract: We quantify uncertainty in machine learning predictions of stock returns. While machine learning methods consistently outperform traditional regression-based approaches in asset return predictions, the point-estimate nature and the lack of closed-form solutions for prediction intervals in most models presents a significant challenge for practical applications. We address this limitation by implementing conformal prediction techniques, enabling uncertainty quantification for predictions generated by arbitrary algorithms. Through extensive empirical analysis of various machine learning models, we document substantial heterogeneity in prediction uncertainty across both models and time. Notably, we find that out-of-sample performance and prediction uncertainty are not necessarily aligned—models with the best apparent predictive and investment performances out-of-sample do not systematically exhibit the lowest prediction uncertainty. These findings contribute to better understanding applications and implications of machine learning methods in problems portfolio choice.

• Alpha in Insurers Corporate Bond Portfolios - with David Ng and Xing (Alex) Zhou
  Abstract: This paper assesses the performance of corporate bond portfolios of life insurers based on detailed daily portfolios constructed from the combination of holdings and transaction data from regulatory filings. On average, life insurers' portfolios do not outperform the broader market, although performance varies significantly across insurers. Investments in illiquid bonds can yield higher returns and alpha; however, regulatory restrictions on managing the duration gap between assets and liabilities offset these gains.