Skip to main content

Quantitative research · Structured credit

Howard
Zeng.

Models built to survive contact with real data.

I build explainable, production-grade models for structured credit—from loan-level data and factor design to daily risk and portfolio decisions.

View experience Start a conversation by email
Portrait of Howard Zeng

Current role

Quantitative Researcher

LibreMax Capital
New York, NY

RMBS / CRT

Research focus

100M+

Loan-level records processed

End to end

Research → models → risk

01 / Experience

Research in practice

From mortgage modeling to production structured-credit research.

Feb 2025 - Present

Current

LibreMax Capital (opens in a new tab)

New York, NY

Quantitative Researcher

  • Build loan-level prepayment, delinquency, and loss models across CRT, MIR, Non-QM, jumbo, and HELOC portfolios, with factors designed for stability under stress.
  • Productionize the research workflow from SQL extraction and feature engineering through GAM/XGBoost fitting, automated QC, and daily risk refreshes.
  • Translate model behavior into tracking and risk explanations used by portfolio managers.

RMBS · CRT · Non-QM · Jumbo · Python · SQL

May 2024 - Aug 2024

Navy Federal Credit Union (opens in a new tab)

Vienna, VA

Data Scientist Intern | Lending Analytics, Pricing Team

  • Built an executive dashboard for a $6B home-equity loan portfolio; surfaced pricing anomalies that informed portfolio strategy.
  • SQL/Python pipelines on Azure/Teradata processing 100M+ records with automated anomaly detection.
  • End-to-end delivery: KPI design → Plotly visualizations → presentation to senior leadership.

Python · SQL · Azure · Pricing · Dashboards

Jan 2024 - May 2024

Gravity Investments (opens in a new tab)

Ithaca, NY

Quantitative Research Engineer & Team Leader | Asset Pricing

  • Led a 6-person team building and evaluating predictive models against an S&P 500 benchmark across rolling backtests.
  • Migrated Temporal Fusion Transformers to AWS SageMaker, optimizing 20M+ record workloads.
  • Drove full pipeline architecture — feature engineering through deployment — as team lead.

Python · PyTorch · AWS SageMaker · Time Series

Jan 2023 - Aug 2023

WeiCepts LLC (opens in a new tab)

Conroe, TX

Financial Data & Model Analyst Intern | Mortgage Modeling

  • Reduced a ~9M-record mortgage dataset by 91% through aggregation, paired with validation checks to preserve the modeling workflow.
  • SMOTE resampling + Optuna-tuned XGBoost/RF/Logistic classifiers with cross-validation.
  • End-to-end mortgage default modeling — data prep through model selection and evaluation.

Mortgage · Python · PostgreSQL · XGBoost

Earlier experience 6 roles

Nov 2022 - Mar 2023

UW Foster School of Business

Seattle, WA

Research Assistant

Sep 2022 - Dec 2022

Huatai Securities

Shanghai, China

Equity Research Intern | Research Institute (Hardware & Software)

Feb 2022 - Apr 2022

China Securities

Beijing, China

Quantitative Research Intern | Derivatives Trading

Sep 2020 - Feb 2021

UW Human Centered Design & Engineering

Seattle, WA

Research Assistant

May 2019 - Jan 2021

iRent

Dublin, Ireland

Mobile Full Stack Developer | Founder & Team Leader

Sep 2019 - Jan 2020

UW Information School

Seattle, WA

Research Assistant

02 / Research

Selected work

Research systems for noisy data, unstable regimes, and decisions that need more than a headline metric.

Flagship research / 01

Python · XGBoost · Pipeline · Walk-forward

AlphaCycle — Stock Prediction Framework

Modular data/model/report pipeline for multi-horizon equity prediction.

Built a config-driven research framework spanning data ingestion, feature storage, model training, and reporting, with walk-forward validation and regime labels.

Read the case study: AlphaCycle — Stock Prediction Framework

Research system

2015–24

Walk-forward window

Workflow / not performance

01

Data

02

Features

03

Models

04

Reports

03 / Profile

How I work

Research is only valuable when it is explainable, reproducible, and connected to a decision.

Current focus

I build loan-level prepayment, delinquency, and loss models across residential credit. My work spans factor design, model fitting, automated QC, daily risk refreshes, and decision-ready explanations for portfolio managers.

Operating principle

Own the full research loop, make the assumptions legible, and fix the root cause—not the symptom. The standard is not a good backtest; it is a system people can trust repeatedly.

Selected prior contexts

Navy Federal Credit Union · Gravity Investments · Huatai Securities · China Securities

04 / Capabilities

Research stack

01

Structured Credit / Modeling

RMBS · CRT (STACR/CAS) · Non-QM · Jumbo · HELOC · Prepayment · Default/Loss · Scenario/Stress Testing

02

ML / Statistics

GAM (mgcv/bam) · XGBoost · LightGBM · Logistic Regression · Time Series · Imbalanced Classification · Walk-forward Validation

03

Data / Engineering

Python (Polars/Pandas) · C++ · R (Tidyverse) · SQL (Redshift/SQL Server) · Parquet · Jenkins · PowerShell · Automation/QC

04

Systems / Tooling

Git · Linux · Ray (distributed compute) · Azure · AWS · Teradata · Reproducible Configs · Model Versioning

05 / Background

Education

Cornell University

2023 - 2024

MPS Applied Statistics — Data Science

GPA 4.08 / 4.3

Large-Scale Machine Learning · Deep Learning · Natural Language Processing · Reinforcement Learning · Stochastic Processes · +4 more

University of Washington

2019 - 2022

BS Economics — Econometrics

GPA 3.6 / 4.0 · Minor: Applied Math & Data Science

Econometric Theory & Applications · Causal Inference · Data Science for Pricing · Financial Economics · Database Systems (SQL) · +4 more

06 / Contact

Let’s talk
research.

Open to quantitative research opportunities and rigorous problems at the intersection of markets, statistics, and production systems.

howieecon@gmail.com
LinkedIn (opens in a new tab) GitHub (opens in a new tab)