About
I’m currently employed at Citigroup as a quantitative analyst within the credit algorithmic trading business. Previously, as a graduate student at Columbia University, I researched applications of Bayesian statistics in population dynamics under the supervision of Upmanu Lall and Michael J. Puma. As an undergraduate at UCLA, I studied advances in machine learning for market microstrure under the mentorship of Moritz Voss.
Coding and Writing
- Reinforcement Learning for Continuous-Time Optimal Execution
Slides from a lecture I gave for a reading group at Columbia on stochastic control and reinforcement learning - Hawkes Processes and Time Clustering in Finance
Brief discussion on Hawkes processes, a class of self-exciting point processes - Separation Capacity of Randomly Initialized Deep Neural Networks.
Technical report written for EECS 6699: Mathematics of Deep Learning, with Harold Haodong Miao, Weitao Zhu - Trading in the Limit Order Book with CNN-LSTM
Using deep learning to extract order flow information from the limit order book and forecast directional moves
Working Papers
A Bayesian Hierarchical Framework for Capturing Preference Heterogeneity in Migration Flows
(with Upmanu Lall, Michael J. Puma, Emile Esmaili, Rachata Muneepeerakul)
Version: May 2024.
Abstract
The prediction of migration flows, or the number of individuals who will migrate from one location to another, is a fundamental goal of research on population dynamics. Traditional approaches for modeling migration flow include physics-inspired models such as the gravity model and the radiation model. In their simplest form, these models represent migration flow as functions of the relative attractiveness of a locale using a few socio-economic features as proxies. Furthermore, they assume that the parameters governing the relationship between features and migration flow are spatially invariant, regardless of the origin and destination locations of migrants. This assumption of spatial invariance is a key limitation of the classical models. We argue that migrant preferences are likely to vary based on the specific origin-destination contexts. To overcome this limitation, we formulate Bayesian hierarchical models to capture variation in regression coefficients according to origin-destination pair. Applying our hierarchical Bayesian models to interstate migration data from the United States, we demonstrate that accounting for heterogeneity in just one latent migration parameter can explain a large amount of variation in migrant flows. This heterogeneity in migrant preferences likely arises from factors beyond the socio-economic features included in the model. Accounting for such heterogeneity through our hierarchical approach enables it to outperform classical methods as well as some recent machine learning approaches. Our detailed clustering analysis of spatially varying parameters within the hierarchical Bayesian model unveils significant patterns differentiating migration decision-making between low-flow and high-flow paths.
Modeling Migration Flows with Non-Homogeneous Hidden Markov Models
(with Emile Esmaili, Upmanu Lall, Michael J. Puma, Rachata Muneepeerakul)
Version: December 2023.
Abstract
Current models of human mobility rely on static models inferred using multivariate regression that do not explicitly model the temporal structure of the data. We propose a new approach using non-homogeneous hidden Markov models (NHMMs) to reveal underlying space-time patterns in human migration that are not directly observable, but whose persistence and likelihood of occurrence may be identified by exogenous drivers. These drivers may include the migration predictors used in traditional models. We develop NHMMs for state-to-state migrations in the United States using data from 2005 to 2019. We test the performance of these models using out-of-sample forecasts and compare those to selected traditional and newer machine learning models.We find that climate disasters emerge as important drivers of migration in the United-States. The NHMM model outperforms traditional human migration models, as well as some recent deep learning approaches for multivariate time series forecasting. The NHMMs provide insights into the hidden patterns driving complex migrations, while delivering superior forecasting performance compared to both linear and nonlinear approaches.
Education
Columbia University | M.S. in Financial Engineering
Sep 2022 - Dec 2023
University of California, Los Angeles | B.S. in Computational Math and Economics
Sep 2019 - Jun 2022
Teaching
Freelance | Tutor, Monte Carlo simulations and stochastic processes
Feb 2024 - May 2024
Columbia University | Course Assistant, Algorithmic Trading
Jan 2023 - May 2023
Movie Talk
If I’m not at the desk, I’m probably at the Lincoln Square movie theater. I’m a loyal AMC Stubs A-Lister and a self-proclaimed critic on Letterboxd.