I will join the U.S. Securities and Exchange Commission this summer as a financial economist in the Office of Markets. I am currently a postdoctoral research associate in the finance department at the WashU Olin Business School.
I completed my Ph.D. in Finance at the Washington University in St. Louis Olin Business School in 2021. Previously, I worked as a research assistant at the Federal Reserve Board of Governors.
My main research field is empirical asset pricing. My current projects use machine learning, large-scale optimization, high-dimensional statistics, and natural language processing to study questions in asset pricing.
I estimate characteristic-sparse stochastic discount factors (SDFs) for the cross-section of expected returns via cardinality constraints. Each resulting SDF is the best feasible SDF of $K$ assets chosen from a given panel of $N$ assets. Characteristic-sparse SDFs estimated in this manner price the cross-section quite well and better than both sparse SDFs estimated via the lasso or elastic net and factor-based SDFs. The cardinality constrained model’s specification is also much more amenable to economic interpretation than specifications working with factor transformations of asset returns. Overall, the results indicate that small, parsimonious, and characteristic-sparse SDF specifications are a viable approach to explaining expected returns despite the rapid growth in the number of firm characteristics associated with cross-sectional variation in expected returns.
Characteristic interactions play an important role in describing the cross-section of expected returns. I use a Fama-Macbeth regression modified to accommodate more vari- ables than observations to study the cross-sectional relationship between characteristic interactions and expected returns. The modified Fama-Macbeth regression uses a form of dimension reduction called an envelope, which does not require variable selection or slope regularization. I use the method to estimate the information in 3,655 character- istic interactions about the cross-section of expected returns. About 100 interactions have incremental information about expected returns. Standard long-short portfolios constructed from interaction-based estimates of expected returns have significant risk- adjusted returns compared to standard factor models.
Public companies report “the most significant factors that make” their common stock “speculative or risky” in section “Item 1A. Risk Factors” of their annual filings. This paper uses textual analysis to estimate common risks from Item 1A texts and study these risks’ effect on public companies’ stock returns. I find the textual relevance of common Item 1A risks to the cross-section of firms’ Item 1A texts predicts the cross-section of expected stock returns. A factor portfolio aggregating information about returns from fifty individual Item 1A risks has an average monthly return of 0.97% and a risk-adjusted return of 1.06%. Factor portfolios for nineteen individual Item 1A risks have significant average returns. Eighteen individual Item 1A risks provide independent information about stock returns.
We investigate the informational content of options-implied probability density functions (PDFs) for the future price of oil. Using a semiparametric variant of the methodology in Breeden and Litzenberger (1978), we investigate the fit and smoothness of distributions derived from alternative PDF estimation methods, and develop a set of robust summary statistics. Using PDFs estimated around episodes of high geopolitical tensions, oil supply disruptions, macroeconomic data releases, and shifts in OPEC production strategy, we explore the extent to which oil price movements are expected or unexpected, and whether agents believe these movements to be persistent or temporary.
Data-Base Management System (DBMS) is the current standard for storing information. A DBMS organizes and maintains a structure of storage of data. Databases make it possible to store vast amounts of randomly created information and then retrieve items using associative reasoning in search routines. However, design of databases is cumbersome. If one is to use a database primarily to directly input information, each field must be predefined manually, and the fields must be organized to permit coherent data input. This static requirement is problematic and requires that database table(s) be predefined and customized at the outset, a difficult proposition since current DBMS lack a user friendly front end to allow flexible design of the input model. Furthermore, databases are primarily text based, making it difficult to process graphical data. We have developed a general and nonproprietary approach to the problem of input modeling designed to make use of the known informational architecture to map data to a database and then retrieve the original document in freely editable form.