AnomalyLab 0.7.0

A Python package for empirical asset pricing analysis.

AnomalyLab

Authors

Chen Haiwei, Deng Haotian

Overview

This Python package implements various empirical methods from the book Empirical Asset Pricing: The Cross Section of Stock Returns by Turan G. Bali, Robert F. Engle, and Scott Murray. The package includes functionality for:

• Summary statistics
• Correlation analysis
• Persistence analysis
• Portfolio analysis
• Fama-MacBeth regression (FM regression)

Additionally, we have added several extra features, such as:

• Missing value imputation
• Data normalization
• Leading and lagging variables
• Winsorization/truncation
• Transition matrix calculation
• Formatting output tables

Installation

The package can be installed via:

pip install anomalylab

Usage

This package provides a comprehensive suite of tools for empirical asset pricing analysis. Below are key functions with explanations and example usage to help you get started.

Importing Data

import pandas as pd
from pandas import DataFrame

from anomalylab import Panel, TimeSeries, pp
from anomalylab.datasets import DataSet

df: DataFrame = DataSet.get_panel_data()
ts: DataFrame = DataSet.get_time_series_data()

# Specifying Factor Models:
Models: dict[str, list[str]] = {
    "CAPM": ["MKT(3F)"],  # Capital Asset Pricing Model with Market Factor
    "FF3": ["MKT(3F)", "SMB(3F)", "HML(3F)"],  # Fama-French 3 Factor Model
    "FF5": ["MKT(5F)", "SMB(5F)", "HML(5F)", "RMW(5F)", "CMA(5F)"],  # Fama-French 5 Factor Model
}

# Creating Panel and Time Series Objects:
panel = Panel(
    df,
    name="Stocks",
    id="permno",
    time="date",
    frequency="M",
    ret="return",
    classifications="industry",
    drop_all_chars_missing=True,
    is_copy=False,
)
time_series: TimeSeries = TimeSeries(
    df=ts, name="Factor Series", time="date", frequency="M", is_copy=False
)
pp(panel)

Preprocessing Data

Several preprocessing functions are available for handling missing values, normalizing data, shifting variables, and winsorizing data.

# Filling Data:
# Filling Group Columns
panel.fill_group_column(group_column="industry", value="Other")
# Filling Missing Values
panel.fillna(method="mean", value=0, group_columns="date")

# Normalizing Data:
# panel.normalize(method="zscore", group_columns="date")

# Shifting Data:
# panel.shift(periods=1, drop_original=False)

# Winsorizing Data:
panel.winsorize(method="winsorize", group_columns="date")
pp(panel)

Summary statistics

You can compute summary statistics for your dataset using the summary() function:

summary = panel.summary()
pp(summary)

Correlation analysis

The correlation() function computes the correlations between different variables in the panel data:

correlation = panel.correlation()
pp(correlation)

Persistence analysis

Persistence analysis helps you understand the stability of certain variables over time. The persistence() function computes persistence for a given set of periods to analyze the stability of a variable. The transition_matrix() function calculates the transition matrix to evaluate how a variable moves between different states (e.g., deciles) over time.

persistence = panel.persistence(periods=[1, 3, 6, 12, 36, 60])
pp(persistence)
pp(
    panel.transition_matrix(
        var="MktCap",
        group=10,
        lag=12,
        draw=False,
        # path="...",
        decimal=2,
    )
)

Portfolio analysis

You can group data, and perform univariate and bivariate portfolio analyses based on factors.

# Grouping
group_result = panel.group("return", "MktCap", "Illiq", 10)

# Univariate portfolio analysis
uni_ew, uni_vw = panel.univariate_analysis(
    "return", "MktCap", "Illiq", 10, Models, time_series, factor_return=False
)
pp(uni_ew)
pp(uni_vw)

# Bivariate portfolio analysis
bi_ew, bi_vw = panel.bivariate_analysis(
    "return",
    "MktCap",
    "Illiq",
    "IdioVol",
    5,
    5,
    Models,
    time_series,
    True,
    False,
    "dependent",
    factor_return=False,
)
pp(bi_ew)
pp(bi_vw)

Fama-MacBeth regression

You can run Fama-MacBeth regressions with multiple independent variables:

fm_result = panel.fm_reg(
    regs=[
        ["return", "MktCap"],
        ["return", "Illiq"],
        ["return", "IdioVol"],
        ["return", "MktCap", "Illiq", "IdioVol"],
    ],
    exog_order=["MktCap", "Illiq", "IdioVol"],
    weight="MktCap",
    industry="industry",
    industry_weighed_method="value",
    is_winsorize=False,
    is_normalize=True,
)
pp(fm_result)

Formatting results

Finally, you can save and format the results to an Excel file:

output_file_path = "..."
with pd.ExcelWriter(output_file_path) as writer:
    summary.to_excel(writer, sheet_name="summary")
    correlation.to_excel(writer, sheet_name="correlation")
    persistence.to_excel(writer, sheet_name="persistence")
    uni_ew.to_excel(writer, sheet_name="uni_ew")
    uni_vw.to_excel(writer, sheet_name="uni_vw")
    bi_ew.to_excel(writer, sheet_name="bi_ew")
    bi_vw.to_excel(writer, sheet_name="bi_vw")
    fm_result.to_excel(writer, sheet_name="fm_result")

panel.format_excel(
    output_file_path,
    align=True,
    line=True,
    convert_brackets=False,
    adjust_col_widths=True,
)