Fama-French 与 AQR 因子数据

笔记本《交易机器学习》

总结

本文介绍公开的 Fama-French 和 AQR 因子收益数据集，说明其在基准比较、风险归因和因子研究中的用途。文中介绍 Fama-French 数据库中的市场、规模、价值、盈利能力、投资和动量因子，以及涵盖质量、低贝塔和替代价值概念的 AQR 序列。数据支持月度频率；加载工具也支持日度 Fama-French 数据和日期范围筛选。

工作流程涵盖下载和缓存数据集、加载与分析数据概况，以及根据月度收益计算年化均值、波动率和夏普率。文中所述 Fama-French 数据覆盖始于 1926，而 AQR 的覆盖范围因因子而异；数据提供方自行定义序列和格式。本资料是数据访问与探索指南，并非任何因子能持续获得收益的证据。研究人员在将因子用于风险调整或业绩评估时，必须考虑提供方的定义、更新时间以及汇总统计的局限。

核心观点

Fama-French 数据集提供市场、规模、价值、盈利能力、投资和动量收益，可用于研究基准。
AQR 数据集包含质量、低贝塔和替代价值因子序列，各因子的覆盖范围不同。
可以加载月度和日度频率数据，并将分析限制在指定日期范围内。
年化均值、波动率和夏普率提供基础汇总，但无法证明因子未来的表现。
因子定义和数据格式由提供方决定，因此结果取决于所选序列及其覆盖范围。

# Academic Factor Data Dataset


# Academic Factor Data Dataset

Fama-French and AQR factor returns for benchmarking and risk adjustment.

| Property | Value |
|----------|-------|
| **Provider** | Ken French Library, AQR |
| **Asset Class** | Factor Returns |
| **Frequency** | Monthly (daily available) |
| **Factors** | FF3, FF5, Momentum, QMJ, BAB |
| **Coverage** | 1926-present (FF), varies (AQR) |
| **Size** | ~5 MB |
| **API Key** | None (free) |
| **Loader** | `load_ff_factors()`, `load_aqr_factors()` |

```python
"""Academic Factor Data - download, explore, and update workflow."""

from pathlib import Path

import polars as pl
```

## 1. Configuration

Academic factor data is **provider-defined** (no local config file). Each provider
maintains their own factor definitions and data format.

```python
print("=== Academic Factor Configuration ===")
print("\nFama-French (Ken French Library):")
print("  - FF3: Mkt-RF, SMB, HML")
print("  - FF5: FF3 + RMW, CMA")
print("  - Momentum: MOM")
print("  - Coverage: 1926-present")
print("\nAQR Research:")
print("  - QMJ: Quality Minus Junk")
print("  - BAB: Betting Against Beta")
print("  - VME: Value Minus Everything")
print("  - HML Devil: Industry-adjusted value")
print("  - Coverage: varies by factor")
```

## 2. API Key Setup

**No API key required.** Both Ken French Library and AQR provide free public access.

```python
print("Ken French Library: Free, no API key required")
print("  URL: https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html")
print("\nAQR Research: Free, no API key required")
print("  URL: https://www.aqr.com/Insights/Datasets")
```

## 3. Download Data

The `ml4t-data` library handles downloading and caching factor data.

```python
def download_ff_factors(
    datasets: list[str] | None = None, frequency: str = "monthly", dry_run: bool = False
):
    """Download Fama-French factor data.

    Args:
        datasets: Specific datasets to download (default: core factors)
        frequency: "monthly" or "daily"
        dry_run: If True, show what would be downloaded
    """
    from ml4t.data.providers.fama_french import FamaFrenchProvider

    from utils import ML4T_DATA_PATH

    output_dir = ML4T_DATA_PATH / "factors" / "fama-french"

    # Default core datasets
    if datasets is None:
        datasets = ["ff3", "ff5", "mom"]

    print("=== Fama-French Download ===")
    print(f"Datasets: {datasets}")
    print(f"Frequency: {frequency}")
    print(f"Output: {output_dir}")

    if dry_run:
        print("\n[DRY RUN] Would download:")
        for ds in datasets:
            print(f"  - {ds}")
        return

    output_dir.mkdir(parents=True, exist_ok=True)
    provider = FamaFrenchProvider(cache_path=output_dir, use_cache=True)

    print(f"\nDownloading {len(datasets)} datasets...")
    for dataset in datasets:
        print(f"  {dataset}...", end=" ", flush=True)
        try:
            df = provider.fetch(dataset, frequency=frequency)
            print(f"OK ({len(df):,} rows)")
        except Exception as e:
            print(f"ERROR: {e}")

    print("\n=== Complete ===")
    print(f"Data saved to: {output_dir}")


def download_aqr_factors(datasets: list[str] | None = None, dry_run: bool = False):
    """Download AQR factor data.

    Args:
        datasets: Specific datasets to download (default: core factors)
        dry_run: If True, show what would be downloaded
    """
    from ml4t.data.providers.aqr import AQRProvider

    from utils import ML4T_DATA_PATH

    output_dir = ML4T_DATA_PATH / "factors" / "aqr"

    # Default core datasets
    if datasets is None:
        datasets = ["qmj", "bab"]

    print("=== AQR Download ===")
    print(f"Datasets: {datasets}")
    print(f"Output: {output_dir}")

    if dry_run:
        print("\n[DRY RUN] Would download:")
        for ds in datasets:
            print(f"  - {ds}")
        return

    output_dir.mkdir(parents=True, exist_ok=True)
    provider = AQRProvider(cache_path=output_dir)

    print(f"\nDownloading {len(datasets)} datasets...")
    for dataset in datasets:
        print(f"  {dataset}...", end=" ", flush=True)
        try:
            df = provider.fetch(dataset)
            print(f"OK ({len(df):,} rows)")
        except Exception as e:
            print(f"ERROR: {e}")

    print("\n=== Complete ===")
    print(f"Data saved to: {output_dir}")
```

### Download Fama-French Factors

```python
# Uncomment to download
# download_ff_factors()
```

### Download AQR Factors

```python
# Uncomment to download
# download_aqr_factors()
```

### Dry Run (Preview)

```python
download_ff_factors(dry_run=True)
```

## 4. Load and Explore

Once downloaded, use the loaders throughout the book:

```python
from data import load_aqr_factors, load_ff_factors
```

### Fama-French Factors

```python
# Load Fama-French factors
ff = load_ff_factors()

print(f"Shape: {ff.shape}")
print(f"Columns: {ff.columns}")
print(f"Date range: {ff['timestamp'].min()} to {ff['timestamp'].max()}")
print(f"Memory: {ff.estimated_size('mb'):.1f} MB")
```

```python
# Preview
ff.tail(10)
```

```python
# Factor statistics (annualized)
factor_cols = [c for c in ff.columns if c not in ["timestamp", "date"]]
print("Factor Annualized Statistics (%):")
for col in factor_cols[:6]:
    series = ff[col].drop_nulls()
    mean_annual = series.mean() * 12  # Monthly to annual
    vol_annual = series.std() * (12**0.5)
    sharpe = mean_annual / vol_annual if vol_annual > 0 else 0
    print(f"  {col:8s}: mean={mean_annual:6.2f}, vol={vol_annual:6.2f}, SR={sharpe:.2f}")
```

### AQR Factors

```python
# Load AQR factors
aqr = load_aqr_factors()

print(f"Shape: {aqr.shape}")
print(f"Columns: {aqr.columns}")
print(f"Date range: {aqr['timestamp'].min()} to {aqr['timestamp'].max()}")
```

```python
# Preview
aqr.tail(10)
```

## 5. Data Profile

```python
from ml4t.data.storage.data_profile import get_profile_path, load_profile

from utils import ML4T_DATA_PATH

for provider, subdir in [("Fama-French", "fama-french"), ("AQR", "aqr")]:
    profile_path = get_profile_path(ML4T_DATA_PATH / "factors" / subdir)
    profile = load_profile(profile_path)
    if profile is None:
        print(f"No {provider} profile at {profile_path}")
    else:
        print(f"=== {provider} Profile ===")
        print(f"Written by {profile.source}")
        print(profile.summary())

print(
    "\nff_download.py and aqr_download.py unzip the providers' own CSV releases and do\n"
    "not go through ml4t.data.storage.data_profile, so neither carries a profile today.\n"
    "Nothing in this notebook writes one either."
)
```

## 6. Loader Options

The loaders support filtering by frequency and date range:

```python
# Daily frequency
ff_daily = load_ff_factors(frequency="daily")
print(f"FF daily: {ff_daily.shape}")
```

```python
# Date range
recent_ff = load_ff_factors(start_date="2020-01-01")
print(f"FF 2020+: {recent_ff.shape}")
```

## 7. Documentation

### Fama-French Factors

From Ken French's Data Library:

| Factor | Description |
|--------|-------------|
| Mkt-RF | Market excess return |
| SMB | Small Minus Big (size) |
| HML | High Minus Low (value) |
| RMW | Robust Minus Weak (profitability) |
| CMA | Conservative Minus Aggressive (investment) |
| Mom | Momentum (12-1 month return) |

[Ken French Data Library](https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html)

### AQR Factors

Alternative factors from AQR Capital:

| Factor | Description |
|--------|-------------|
| QMJ | Quality Minus Junk (profitability, growth, safety) |
| BAB | Betting Against Beta (low-beta premium) |
| VME | Value Minus Everything (alternative value) |
| HML Devil | Value with industry adjustment |

[AQR Datasets](https://www.aqr.com/Insights/Datasets)

## 8. Updating Data

To update with the latest data:

```python
# Update Fama-French factors
download_ff_factors()

# Update AQR factors
download_aqr_factors()
```

Factor data is typically updated monthly.

## Summary

| Item | Value |
|------|-------|
| Providers | Ken French, AQR |
| Frequencies | Monthly, Daily |
| Coverage | 1926-present (FF), varies (AQR) |
| API Key | None (free) |
| Loaders | `load_ff_factors()`, `load_aqr_factors()` |

**Primary use**: Risk attribution, alpha measurement, factor investing research.

在遵守原作品许可的前提下，附作者信息全文展示。许可协议: MIT

此摘要由 Stratmill 研究智能体根据原文撰写，并非原文副本。

使用 Fama-French 和 AQR 因子进行风险调整与基准比较

总结

核心观点

标签