This project reproduces and extends a momentum factor research workflow based on the Quantopian Alphalens framework.
Since Quantopian has been discontinued, the original workflow is no longer directly executable. In this project, the factor construction pipeline is reimplemented using Yahoo Finance market data, pandas rolling-window calculations, and the Alphalens-reloaded library.
The objective of this project is to reproduce the original momentum factor evaluation workflow and compare multiple momentum factor constructions under a unified evaluation framework.
This project implements and evaluates the following momentum factors:
- Simple Momentum
- Regression Momentum
- Regression Momentum × R²
- Volatility-adjusted Momentum
- 12–1 Momentum
- EWMA Momentum
Each factor is evaluated using Alphalens to analyze its predictive power, portfolio performance, and factor stability.
- Yahoo Finance (yfinance)
- S&P 500 historical daily prices
The following momentum factors are implemented:
| Factor | Description |
|---|---|
| Simple Momentum | Past 12-month cumulative return |
| Regression Momentum | Linear regression slope of log prices |
| Regression Momentum × R² | Annualized regression slope weighted by R² |
| Volatility-adjusted Momentum | Momentum normalized by historical volatility |
| 12–1 Momentum | Excludes the most recent month to reduce short-term reversal |
| EWMA Momentum | Momentum based on exponentially weighted returns |
Each factor is evaluated using Alphalens factor tear sheets, including:
- Long–Short Quantile Return Spread
- Information Coefficient (IC)
- IC t-statistic
- Mean Quantile Turnover
- Rank Autocorrelation
| Factor | 10D Spread | 10D IC Mean | t-stat | 10D Mean Turnover | Notes |
|---|---|---|---|---|---|
| Simple Momentum | 4.743 | 0.053 | 4.702 | 0.171 | Baseline |
| Regression Momentum | 5.555 | 0.064 | 5.831 | 0.074 | Strongest IC |
| Regression Momentum × R² | 1.913 | 0.050 | 4.697 | 0.084 | Penalizes noisy trends |
| Volatility-adjusted Momentum | 3.810 | -0.008 | -0.691 | 0.173 | Risk-adjusted momentum |
| 12–1 Momentum | 1.479 | 0.053 | 4.876 | 0.164 | Classical momentum |
| EWMA Momentum | 4.407 | 0.024 | 2.386 | 0.203 | Recent prices weighted more |
- Successfully reproduced the original Alphalens momentum factor evaluation workflow without relying on the deprecated Quantopian platform.
- Regression Momentum achieved the strongest predictive performance among the reproduced momentum factors.
- Additional momentum factor constructions were implemented to compare different definitions of momentum.
- The comparison demonstrates how different factor construction methods influence predictive power, portfolio turnover, and factor stability.
Alphalens evaluates factor quality from several different perspectives.
Stocks are ranked by their factor values and divided into five equally sized quantiles.
- Top Quantile (Q5): Stocks with the highest factor values.
- Bottom Quantile (Q1): Stocks with the lowest factor values.
- Quantiles 2–4: Intermediate groups sorted by factor value.
If the factor is effective, stocks in the Top Quantile are expected to outperform those in the Bottom Quantile over future holding periods.
The figure below shows the distribution of factor values across the five quantiles.
The Long–Short Spread measures the performance difference between the highest-ranked and lowest-ranked portfolios.
Spread = Return(Q5) − Return(Q1)
A larger positive spread indicates stronger factor predictive power.
The Information Coefficient measures the correlation between today's factor ranking and future stock returns.
- IC > 0 indicates that higher-ranked stocks tend to generate higher future returns.
- Larger IC values imply stronger predictive ability.
The IC t-statistic measures whether the Information Coefficient is statistically significant.
Generally,
- t-stat > 2 indicates statistically significant predictive power.
Turnover measures how frequently stocks move in or out of the same quantile over time.
- Lower turnover indicates a more stable factor.
- Higher turnover implies more frequent portfolio rebalancing and potentially higher transaction costs.
Rank Autocorrelation measures how stable the factor rankings remain over consecutive periods.
- Values close to 1 indicate highly stable rankings.
- Lower values indicate that factor rankings change rapidly over time.
Six momentum factor definitions were compared using identical Alphalens evaluation metrics.
Compared with the original workflow, this project includes the following contributions:
- Reimplemented the original Quantopian workflow using Yahoo Finance data.
- Replaced deprecated Quantopian APIs with pandas rolling-window implementations.
- Reproduced the original Alphalens factor evaluation pipeline.
- Implemented and evaluated additional momentum factors:
- Volatility-adjusted Momentum
- 12–1 Momentum
- EWMA Momentum
- Compared multiple momentum factor constructions under a unified evaluation framework.
Momentum-Factor-Research/
│
├── notebooks/
│ └── Momentum_Factor_Research.ipynb
│
├── figures/
│
├── data/
│
├── README.md
├── requirements.txt
└── .gitignore
pip install -r requirements.txtRun the notebook:
jupyter notebook notebooks/Momentum_Factor_Research.ipynbor open the notebook directly in VS Code.
This project is inspired by the original Alphalens project developed by Quantopian.
Original Repository:
https://github.com/quantopian/alphalens
License:
Apache License 2.0
The original Alphalens project provides the factor evaluation framework that inspired this work.
Since Quantopian has been discontinued, this repository provides an independent implementation using Yahoo Finance data and pandas rolling-window calculations.
All additional momentum factor implementations, comparative experiments, documentation, and project organization were independently developed in this repository.