Kecheng Shi

Portrait of Kecheng Shi in a light blue shirt

Hi, I’m Kecheng. I earned my master’s in Mathematics of Finance at Columbia, and I’m based in New York. I enjoy taking a mathematical idea through to working code, then testing it against real market data. My research has taken me from trading strategies in Shanghai to cross-asset portfolios in New York.

I’m curious by nature and skeptical of a result that looks too good—including my own. I want to understand what drives it, which assumptions matter, and what would change my mind. I also enjoy making complicated ideas easier to discuss, something I practiced as a teaching assistant for Machine Learning for Finance.

Outside work, I’m a 2300-rated blitz chess player. I enjoy making decisions under pressure, then going back to understand what I missed. That habit of thinking, testing, and reviewing carries into my research. The projects below give you a closer look at how I work.

Projects & research.

01 Public project · AlgoTrading

Strategy research & paper trading

Signals connected to paper orders and risk checks

Tracked strategies from backtests through paper trades. I connected Alpaca order execution, data checks, and risk monitoring, with saved results and a dashboard for review.

PythonAlpacaPostgreSQL

GitHub

Details

I built a workflow that connects strategy research with paper execution and operational review. It starts with EMA–ADX trend following and Z-score mean reversion, then follows the resulting orders, positions, and risk measures through separate services. The aim was to make a research run inspectable beyond its final equity curve.

How I built it

The Python backtester writes results and configuration metadata to CSV and JSON. A separate executor submits paper orders through Alpaca, while an order listener observes broker updates. An independent risk worker reads positions and market data, records snapshots, and can enforce configured controls. PostgreSQL stores order and risk telemetry, and a Streamlit dashboard brings that information together with saved research runs. A standalone data guard checks freshness, gaps, and spreads.

What the report covers

The report follows the strategies from signal definitions and trading assumptions to historical comparisons with buy-and-hold. Saved figures show the equity paths; run metadata records the configuration behind them. The architecture below distinguishes the independent services and their shared data, so the implementation can be examined alongside the analysis.

Project architecture

System map drawn from repository source · scroll to explore, use + to zoom

Strategy research & paper trading: source-derived architecture overview

The data guard runs independently: freshness, gaps, spreads, and warning output.

Read the report

Project report · 27 pages · scroll to read, use + to zoom

Strategy research & paper trading, original report page 1
View a saved result
Unchanged saved EMA strategy equity and buy-and-hold comparison
Saved backtest figure · original repository output. Source

Research scope. Historical, in-sample research. The report’s cost assumptions need reconciliation; this is not evidence of a validated trading edge.

Source: kecheng-shi/AlgoTrading · eaa36f1

02 Numeraxial · Professional research

Cross-asset portfolio research

Equity, macroeconomic, and FX data used to build a portfolio allocation

Made portfolio comparisons repeatable with consistent market data. At Numeraxial, I standardized Bloomberg inputs and built reusable workflows to compare allocations, turnover, and downside risk.

PythonSQLBloomberg

Experience

Details

What I built

I used Python and SQL to standardize Bloomberg equity and FX data, identifiers, trading calendars, and corporate actions. Input checks, saved configurations and results, and structured logs made the research runs repeatable.

I also used hidden Markov models (HMMs) and clustering to study market regimes, their stability, and drawdowns within each regime.

Comparing allocations

I checked the matrices and portfolio constraints, then compared mean–variance, Black–Litterman, and hierarchical risk parity allocations against saved benchmarks. I evaluated Sharpe ratios, drawdowns, turnover, and risk contributions.

03 Columbia · Academic project

Factor risk & portfolio construction

59 industry and six style factors connected to portfolio weights

Evaluated portfolio returns alongside total and stock-specific risk. I compared regression and neural-network models using 59 industry and six style factors, then calculated portfolio weights.

PythonNumPySciPy

GitHub

Details

This study connects return modeling to portfolio construction. I compared linear and neural-network models of factor-adjusted equity returns, then combined their outputs with a factor risk model to calculate portfolio weights. The analysis follows the modeling choices through to historical portfolio profit, exposures, and risk.

How I built it

The notebook forms a market-cap-filtered universe, winsorizes raw returns, and removes exposures to 59 industry and six style factors. Dates divide the data into an initial 80% training period and a final 20% test period. Regularized weighted least squares provides the linear baseline, with a neural network as the comparison. Both modeling branches feed portfolio-weight calculations using a diagonal factor-covariance approximation, stock-specific variances, and the Woodbury identity.

What the report covers

The report develops the weight calculation and presents regularization studies alongside portfolio diagnostics. Its saved plots separate total risk from the stock-specific component and inspect long and short market values. The method and risk pages below let you follow the mathematics into those outputs, rather than judging the models only by a fitted prediction error.

Project architecture

Workflow reconstructed from the public notebook · scroll to explore, use + to zoom

Factor risk & portfolio construction: source-derived architecture overview

A simplified reconstruction of the notebook’s research workflow.

Read the report

Notebook report · 29 pages · scroll to read, use + to zoom

Factor risk & portfolio construction, original report page 11
View a saved result
Saved risk decomposition from the committed report or notebook
Saved risk decomposition · original repository output. Source

Research scope. The input dataset is not included in the public repository. Portfolio diagnostics are historical and pre-cost.

Source: kecheng-shi/Portfolio-Optimization-of-Factor-Based-Risk-Model-using-Machine-Learning · 249ffd8

04 CQF · Academic study

SPY classification study

Features passed through a decision tree for classification

Compared feature selection and tuning methods for classifying SPY returns. I built an XGBoost workflow with saved evaluation results. Its same-day return target limits its use for forecasting.

Pythonscikit-learnXGBoost

GitHub

Details & validation

I built an XGBoost classification study around SPY market data, exploring how features are selected, how model settings are tuned, and how the results are evaluated. The executed notebook labels whether the current day’s return exceeds 0.25%; it therefore studies same-day classification rather than a next-day forecasting target.

How I built it

The workflow downloads daily price and volume history and constructs rolling statistics, technical indicators, lagged returns, and calendar features. Feature exploration combines mutual information, random-forest recursive elimination, and XGBoost importance. A separate seven-feature model set is used for the baseline and tuned classifiers, with a chronological 80/20 split and TimeSeriesSplit for tuning. The feature-selection stage occurs before the final holdout, an important distinction when interpreting the evaluation.

What the report covers

The report walks through preprocessing, selection, and model diagnostics. The saved ROC and historical strategy comparison use baseline probabilities, rather than the tuned model’s predictions. The comparison shifts the trading signal by one day and omits trading costs. Those details are kept visible so the original figures can be read as methodology evidence, with their assumptions intact.

Project architecture

Workflow reconstructed from the public notebook · scroll to explore, use + to zoom

SPY classification study: source-derived architecture overview

Workflow reconstructed from the executed public notebook.

Read the report

Notebook report · 20 pages · scroll to read, use + to zoom

SPY classification study, original report page 4
View a saved result
Saved strategy comparison from the committed report or notebook
Saved strategy comparison · original repository output. Source

Research scope. Same-day labels and feature selection before the holdout limit forecasting claims. Saved comparison results omit transaction costs.

Source: kecheng-shi/Predicting-Positive-SPY-Moves-with-Gradient-Boosting · 612dda6

Experience

Numeraxial LLC Quantitative Research Intern New York, NY

July 2025 – Present

Standardized market data and built reusable research tools to compare market regimes and portfolio allocations under constraints. Project details

Western Securities Co., Ltd. Quantitative Analyst Intern Shanghai, China

January 2024 – July 2024

Tested EMA–ADX and Z-score cryptocurrency strategies, evaluating parameter sensitivity, trading costs, turnover, value at risk (VaR), and drawdowns.

Skills & technologies

systematic trading machine learning portfolio risk portfolio construction backtesting factor models feature engineering Black–Litterman regime analysis paper trading XGBoost data validation drawdown analysis mean–variance regularized regression clustering HMM HRP turnover quantitative research cross-asset research trend following mean reversion VaR trading costs limit orders parameter sensitivity corporate actions portfolio constraints risk contributions neural networks weighted least squares residual returns winsorization feature selection random forests mutual information time-series validation momentum volatility calendar features option pricing numerical methods stochastic processes Python pandas PostgreSQL Docker Git Alpaca EMA ADX Z-score vectorization order execution exposure data freshness order fills NumPy SciPy pseudoinverse Woodbury identity industry factors style factors stock-specific risk scikit-learn Jupyter classification RFE holdout testing cross-validation confusion matrices model tuning SQL Bloomberg equities FX market data Sharpe ratio asset allocation benchmarking trading calendars structured logs probability statistics derivatives time series operational research linear models computational finance

Education

Columbia University M.A. in Mathematics of Finance New York, NY

September 2024 – December 2025

GPA: 3.94/4.0

Teaching assistant · Machine Learning for Finance (MATH GR5430)

Coursework

Machine Learning, Time Series Modeling, Stochastic Process, Programming for Computational Finance, Numerical Methods, Derivatives Trading, Option Pricing.

Low Library and Alma Mater at Columbia University
Low Library · Columbia University

University of Liverpool B.Sc. in Mathematics with Finance Suzhou, China / Liverpool, UK

September 2020 – May 2024

GPA: 3.83/4.0 · First-class honours (WES report)

Coursework

Stochastic Theory, Linear Statistical Models, Mathematical Risk Theory, Applied Probability, Derivative Securities, Numerical Methods, Operational Research, Statistics and Probability, Financial Reporting.

Victoria Building and University Square at the University of Liverpool
Victoria Building · University of Liverpool

Certificates

Background & interests

Work authorization: Currently on OPT · STEM OPT

Where I’ve studied and worked

World map showing the locations described below

New York Based here

I studied Mathematics of Finance at Columbia and now work on quantitative research at Numeraxial.

Liverpool

I completed my B.Sc. in Mathematics with Finance at the University of Liverpool.

Suzhou & Shanghai

I studied in Suzhou and worked as a quantitative analyst intern at Western Securities in Shanghai.

Languages

  • EnglishBilingual
  • ChineseBilingual
  • Spanish
  • French

Outside work