toraniko_pandas 0.1.2

Pandas implementation of institutional-grade multi-factor risk model (Polars original by 0xfdf/toraniko)

toraniko-pandas

Pandas implementation of institutional-grade multi-factor risk model (Polars original by 0xfdf/toraniko)

Changelog

New Statistical Functions

The following functions in extras.py are inspired by the techniques described in "The Elements of Quantitative Investing" by Giuseppe A. Paleologo:

• Covariance Estimation

  • shrinkage_cov: Implementation of Ledoit & Wolf's optimal shrinkage method for covariance matrices
  • ewma_cov: Exponentially Weighted Moving Average for covariance estimation
  • lagged_covs: Calculation of lag-based covariance matrices for time series data

• Autocorrelation-Consistent Covariance Estimation

  • autocorr_cov: Autocorrelation-consistent estimator for factor covariance
  • nwest_cov: Newey-West estimator for heteroskedasticity and autocorrelation consistent covariance

• CVXPY Optimization

  • estimate_factor_returns_cvxpy: Constrained optimization for factor return estimation
  • _factor_returns_cvxpy: Implementation of the convex optimization approach with sum-zero constraints

These functions provide robust statistical methods for portfolio management, risk modeling, and factor analysis as described in modern quantitative finance literature.

Installation

pip install toraniko_pandas

User Manual

This notebook demonstrates factor analysis using SimFin financial data and custom factor models.

Important

You need free API key from SimFin to run my example. See their Github for more info. Create a .env file and place it like below

# SimFin
SIMFIN_API_KEY = "YOUR_KEY"

Data

import warnings

import matplotlib.pyplot as plt
import statsmodels.api as sm

from simfin_downloader import SimFin
from toraniko_pandas.model import estimate_factor_returns
from toraniko_pandas.styles import factor_mom, factor_sze, factor_val
from toraniko_pandas.utils import top_n_by_group

# Configure warnings
warnings.simplefilter(action="ignore", category=FutureWarning)
warnings.filterwarnings("ignore", message="invalid value encountered in log")
warnings.filterwarnings("ignore", message="divide by zero encountered in log")

simfin = SimFin()
df, sectors, industries = simfin.get_toraniko_data()

# Filter top 3000 companies by market cap
df = top_n_by_group(df.dropna(), 3000, "market_cap", ("date",), True)

# Display sector data
sectors

	sector_basic_materials	sector_business_services	sector_consumer_cyclical	sector_consumer_defensive	sector_energy	sector_financial_services	sector_healthcare	sector_industrials	sector_other	sector_real_estate	sector_technology	sector_utilities
symbol
A	0	0	0	0	0	0	1	0	0	0	0	0
AA	1	0	0	0	0	0	0	0	0	0	0	0
AAC	0	0	0	0	0	1	0	0	0	0	0	0
AACI	0	0	0	0	0	1	0	0	0	0	0	0
AAC_delisted	0	0	0	0	0	0	1	0	0	0	0	0
...	...	...	...	...	...	...	...	...	...	...	...	...
ZWS	0	0	0	0	0	0	0	1	0	0	0	0
ZY	0	0	0	0	0	0	1	0	0	0	0	0
ZYME	0	0	0	0	0	0	1	0	0	0	0	0
ZYNE	0	0	0	0	0	0	1	0	0	0	0	0
ZYXI	0	0	0	0	0	0	1	0	0	0	0	0

5766 rows × 12 columns

# Display industry data
industries

	industry_advertising_and_marketing_services	industry_aerospace_and_defense	industry_agriculture	industry_airlines	industry_alternative_energy_sources_and_other	industry_application_software	industry_asset_management	industry_autos	industry_banks	industry_beverages_-_alcoholic	...	industry_retail_-_defensive	industry_semiconductors	industry_steel	industry_tobacco_products	industry_transportation_and_logistics	industry_travel_and_leisure	industry_truck_manufacturing	industry_utilities_-_independent_power_producers	industry_utilities_-_regulated	industry_waste_management
symbol
A	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
AA	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
AAC	0	0	0	0	0	0	0	0	1	0	...	0	0	0	0	0	0	0	0	0	0
AACI	0	0	0	0	0	0	0	0	1	0	...	0	0	0	0	0	0	0	0	0	0
AAC_delisted	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
ZWS	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
ZY	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
ZYME	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
ZYNE	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0
ZYXI	0	0	0	0	0	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0

5766 rows × 74 columns

# Asset returns was calculated by taking the percentage change of the adjusted close prices
ret_df = df[["symbol", "asset_returns"]]
ret_df

	symbol	asset_returns
date
2019-04-01	EOSS	0.041667
2019-04-01	YTEN	0.017544
2019-04-01	GIDYL	0.000000
2019-04-01	VISL	-0.028571
2019-04-01	ASLE	-0.001020
...	...	...
2024-02-16	NLY	-0.012099
2024-02-16	AAPL	-0.008461
2024-02-16	MSFT	-0.006159
2024-02-16	WPC	0.001669
2024-02-16	RENT	-0.045675

3406422 rows × 2 columns

# Market cap is used to calculate the size factor and later to select the top N by market cap
cap_df = df[["symbol", "market_cap"]]
cap_df

	symbol	market_cap
date
2019-04-01	EOSS	1.600000e+05
2019-04-01	YTEN	2.939904e+05
2019-04-01	GIDYL	3.113633e+05
2019-04-01	VISL	3.423120e+05
2019-04-01	ASLE	3.623279e+05
...	...	...
2024-02-16	NLY	2.291316e+12
2024-02-16	AAPL	2.865508e+12
2024-02-16	MSFT	3.016308e+12
2024-02-16	WPC	3.098176e+12
2024-02-16	RENT	8.506228e+12

3406422 rows × 2 columns

# To calculate the sector scores, we merge the sectors dataframe with the main dataframe
sector_scores = (
    df.reset_index()
    .merge(sectors, on="symbol")
    .set_index("date")[["symbol"] + sectors.columns.tolist()]
)
sector_scores

	symbol	sector_basic_materials	sector_business_services	sector_consumer_cyclical	sector_consumer_defensive	sector_energy	sector_financial_services	sector_healthcare	sector_industrials	sector_other	sector_real_estate	sector_technology	sector_utilities
date
2019-04-01	EOSS	0	0	0	1	0	0	0	0	0	0	0	0
2019-04-01	YTEN	1	0	0	0	0	0	0	0	0	0	0	0
2019-04-01	VISL	0	0	0	0	0	0	0	0	0	0	1	0
2019-04-01	ASLE	0	0	0	0	0	0	0	1	0	0	0	0
2019-04-01	ALBO	0	0	0	0	0	0	1	0	0	0	0	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...
2024-02-16	NLY	0	0	0	0	0	1	0	0	0	0	0	0
2024-02-16	AAPL	0	0	0	0	0	0	0	0	0	0	1	0
2024-02-16	MSFT	0	0	0	0	0	0	0	0	0	0	1	0
2024-02-16	WPC	0	0	0	0	0	0	0	0	0	1	0	0
2024-02-16	RENT	0	0	1	0	0	0	0	0	0	0	0	0

3393141 rows × 13 columns

# Size factor calculation
size_df = factor_sze(df, lower_decile=0.2, upper_decile=0.8)

# Momentum factor calculation
mom_df = factor_mom(df, trailing_days=252, winsor_factor=0.01)

# Value factor calculation
value_df = factor_val(df, winsor_factor=0.05)

# Momentum distribution
mom_df["mom_score"].hist(bins=100)
plt.title("Momentum Factor Distribution")
plt.show()

# Value distribution
value_df["val_score"].hist(bins=100)
plt.title("Value Factor Distribution")
plt.show()

# %%  All of which we merge to get the style scores
style_scores = (
    value_df.merge(mom_df, on=["symbol", "date"])
    .merge(size_df, on=["symbol", "date"])
    .dropna()
)
style_scores

	symbol	val_score	mom_score	sze_score
date
2020-04-28	ALBO	2.732799	-0.308748	3.418444
2020-04-28	HSDT	2.441402	-1.777434	3.316239
2020-04-28	ASLE	1.547939	0.836804	3.259800
2020-04-28	VISL	2.732799	-1.571045	3.042578
2020-04-28	KNWN	1.570320	-0.325302	2.970678
...	...	...	...	...
2024-02-16	NLY	-2.076270	0.015212	-2.906904
2024-02-16	AAPL	-1.156765	0.344759	-2.996094
2024-02-16	MSFT	-0.963887	0.876672	-3.016549
2024-02-16	WPC	-2.877029	-0.090848	-3.027231
2024-02-16	RENT	-2.327684	-2.079591	-3.430058

2479098 rows × 4 columns

ddf = (
    ret_df.merge(cap_df, on=["date", "symbol"])
    .merge(sector_scores, on=["date", "symbol"])
    .merge(style_scores, on=["date", "symbol"])
    .dropna()
    .astype({"symbol": "category"})
)
returns_df = ddf[["symbol", "asset_returns"]]
mkt_cap_df = ddf[["symbol", "market_cap"]]
sector_df = ddf[["symbol"] + sectors.columns.tolist()]
style_df = ddf[style_scores.columns]

fac_df, eps_df = estimate_factor_returns(
    returns_df,
    mkt_cap_df,
    sector_df,
    style_df,
    winsor_factor=0.1,
    residualize_styles=False,
)

factor_cols = ["market","mom_score", "val_score", "sze_score"]

fac_df[factor_cols].plot(subplots=True, figsize=(15, 10))
plt.title("Factor Returns")
plt.show()

(fac_df[factor_cols] + 1).cumprod().plot(figsize=(15, 10))
plt.title("Factor Returns Cumulative")
plt.show()

y = returns_df.query("symbol == 'AAPL'")["asset_returns"]
X = fac_df[["market", "sector_technology", "mom_score", "val_score", "sze_score"]]
X = sm.add_constant(X)

model = sm.OLS(y, X)
results = model.fit().get_robustcov_results()
results.summary()

OLS Regression Results

Dep. Variable:	asset_returns	R-squared:	0.610
Model:	OLS	Adj. R-squared:	0.608
Method:	Least Squares	F-statistic:	259.8
Date:	Fri, 21 Feb 2025	Prob (F-statistic):	3.77e-175
Time:	10:42:32	Log-Likelihood:	2914.3
No. Observations:	959	AIC:	-5817.
Df Residuals:	953	BIC:	-5787.
Df Model:	5
Covariance Type:	HC1

	coef	std err	t	P>|t|	[0.025	0.975]
const	0.0004	0.000	0.955	0.340	-0.000	0.001
market	-1.3258	0.175	-7.594	0.000	-1.668	-0.983
sector_technology	0.3238	0.087	3.738	0.000	0.154	0.494
mom_score	0.4925	0.095	5.211	0.000	0.307	0.678
val_score	-1.4278	0.233	-6.140	0.000	-1.884	-0.971
sze_score	-4.6804	0.389	-12.030	0.000	-5.444	-3.917

Omnibus:	241.840	Durbin-Watson:	1.897
Prob(Omnibus):	0.000	Jarque-Bera (JB):	2493.520
Skew:	0.841	Prob(JB):	0.00
Kurtosis:	10.718	Cond. No.	1.14e+03

Notes:
[1] Standard Errors are heteroscedasticity robust (HC1)
[2] The condition number is large, 1.14e+03. This might indicate that there are
strong multicollinearity or other numerical problems.

The strong multicollinearity is due to the estimated market factor derived from the sectors is almost correlated 1:1 with the sze factor.