getfactormodels

Reliably retrieve data for various multi-factor asset pricing models.

Models

• The 3-factor, 5-factor, and 6-factor models of Fama & French ^{[1] [3] [4]}
• Mark Carhart's 4-factor model ^[2]
• Pastor and Stambaugh's liquidity factors ^[5]
• Mispricing factors of Stambaugh and Yuan^[6]
• The $q$-factor model of Hou, Mo, Xue and Zhang^[7]
• The augmented $q^5$-factor model of Hou, Mo, Xue and Zhang^[8]
• Intermediary Capital Ratio (ICR) of He, Kelly & Manela^[9]
• The DHS behavioural factors of Daniel, Hirshleifer & Sun^[10]
• The HML $^{DEVIL}$ factor of Asness & Frazzini^[11]
• The 6-factor model of Barillas and Shanken^[12]

Thanks to: Kenneth French, Robert Stambaugh, Lin Sun, Zhiguo He, AQR Capital Management (AQR.com) and Hou, Xue and Zhang (global-q.org), for their research and for the datasets they provide.

Installation

getfactormodels requires Python >=3.7

• The easiest way to install getfactormodels is via pip:

  $ pip install getfactormodels

Usage

Important

getfactormodels is new. It was released on December 20, 2023. Don't rely on it for anything.

After installation, import getfactormodels, and call the get_factors() function with the model and frequency parameters.

• For example, to retrieve the monthly ${q}^{5}$ factor model:

   import getfactormodels
  
   data = getfactormodels.get_factors(model='q', frequency='m')

  Trimmed output:

  > print(data)
                Mkt-RF      R_ME      R_IA     R_ROE      R_EG        RF
  date                                                                  
  1967-01-03  0.000778  0.004944  0.001437 -0.007118 -0.008563  0.000187
  1967-01-04  0.001667 -0.003487 -0.000631 -0.002044 -0.000295  0.000187
  1967-01-05  0.012990  0.004412 -0.005688  0.000838 -0.003075  0.000187
  1967-01-06  0.007230  0.006669  0.008897  0.003603  0.002669  0.000187
  1967-01-09  0.008439  0.006315  0.000331  0.004949  0.002979  0.000187
  ...              ...       ...       ...       ...       ...       ...
  2022-12-23  0.005113 -0.001045  0.004000  0.010484  0.003852  0.000161
  2022-12-27 -0.005076 -0.001407  0.010190  0.009206  0.003908  0.000161
  2022-12-28 -0.012344 -0.004354  0.000133 -0.010457 -0.004953  0.000161
  2022-12-29  0.018699  0.008568 -0.008801 -0.012686 -0.002162  0.000161
  2022-12-30 -0.002169  0.001840  0.001011 -0.004151 -0.003282  0.000161
  
  [14096 rows x 6 columns]

• To retrieve the daily data for the Fama-French 3-factor model, since start_date:

  import getfactormodels as gfm
  
  df = gfm.get_factors(model='ff3', frequency='d', start_date=`2006-01-01`)

• To retrieve data for Stambaugh and Yuan's monthly Mispricing factors, between start_date and end_date, and save the data to a file:

  import getfactormodels as gfm
  
  df = gfm.get_factors(model='mispricing', start_date='1970-01-01', end_date=1999-12-31, output='mispricing_factors.csv')

  output can be a filename, directory, or path. If no extension is specified, defaults to .csv (can be one of: .xlsx, .csv, .txt, .pkl, .md)

You can import only the models that you need:

• For example, to import only the ICR and q-factor models:

  from getfactormodels import icr_factors, q_factors
  
  # Passing a model function without params defaults to monthly data.
  df = icr_factors()
  
  # The 'q' models, and the 3-factor model of Fama-French have weekly data available:
  df = q_factors(frequency="W", start_date="1992-01-01, output='.xlsx')

  output allows just a file extension (with the ., else it'll be passed as a filename).

• When using ff_factors(), specify an additional model parameter (this might be changed):

  # To get annual data for the 5-factor model:
  data = ff_factors(model="5", frequency="Y", output=".xlsx")
  
  # Daily 3-factor model data, since 1970 (not specifying an end date
  # will return data up until today):
  data = ff_factors(model="3", frequency="D", start_date="1970-01-01")

There's also the FactorExtractor class (which doesn't do much yet, it's mainly used by the CLI):

from getfactormodels import FactorExtractor

fe = FactorExtractor(model='carhart', start_date='1980-01-01', end_date='1980-05-01)
fe.get_factors()
fe.drop_rf() 
fe.to_file('~/carhart_factors.md')

• The resulting carhart_factors.md file will look like this:

  date	Mkt-RF	SMB	HML	MOM
  1980-01-31 00:00:00	0.0551	0.0162	0.0175	0.0755
  1980-02-29 00:00:00	-0.0122	-0.0185	0.0061	0.0788
  1980-03-31 00:00:00	-0.129	-0.0664	-0.0101	-0.0955
  1980-04-30 00:00:00	0.0397	0.0105	0.0106	-0.0043

.drop_rf() will return the DataFrame without the RF column. You can also drop the Mkt-RF column with .drop_mkt()

CLI

Requires bash >=4.2

• You can also use getfactormodels from the command line. It's very basic at the moment, here's the --help:

  $ getfactormodels -h
  
  usage: getfactormodels [-h] -m MODEL [-f FREQ] [-s START] [-e END] [-o OUTPUT] [--no_rf] [--no_mkt]

• An example of how to use the CLI to retrieve the Fama-French 3-factor model data:

  $ getfactormodels --model ff3 --frequency M --start-date 1960-01-01 --end-date 2020-12-31 --output .csv

• Here's another example that retrieves the annual 5-factor data of Fama-French, without the RF column (using --no[_]rf)

  $ getfactormodels -m ff5 -f Y -s 1960-01-01 -e 2020-12-31 --norf -o ~/some_dir/filename.xlsx

• To return the factors without the risk-free rate RF, or the excess market return Mkt-RF, columns:

  $ getfactormodels -m ff5 -f Y -s 1960-01-01 -e 2020-12-31 --norf --nomkt -o ~/some_dir/filename.xlsx

Data Availability

Model	Start	D	W	M	Q	Y
Fama-French 3	1927-01-03	$\checkmark$	$\checkmark$	$\checkmark$		$\checkmark$
Fama-French 5		$\checkmark$		$\checkmark$		$\checkmark$
Fama-French 6		$\checkmark$		$\checkmark$		$\checkmark$
Carhart 4		$\checkmark$		$\checkmark$		$\checkmark$
DHS		$\checkmark$		$\checkmark$
ICR		$\checkmark$		$\checkmark$	$\checkmark$
Mispricing				$\checkmark$
Liquidity	1962-08-31			$\checkmark$
HML $^{DEVIL}$		$\checkmark$		$\checkmark$
$q$-factors		$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$
Barillas Shanken		$\checkmark$		$\checkmark$

[TODO]

References

1. E. F. Fama and K. R. French, ‘Common risk factors in the returns on stocks and bonds’, Journal of Financial Economics, vol. 33, no. 1, pp. 3–56, 1993. PDF
2. M. Carhart, ‘On Persistence in Mutual Fund Performance’, Journal of Finance, vol. 52, no. 1, pp. 57–82, 1997. PDF
3. E. F. Fama and K. R. French, ‘A five-factor asset pricing model’, Journal of Financial Economics, vol. 116, no. 1, pp. 1–22, 2015. PDF
4. E. F. Fama and K. R. French, ‘Choosing factors’, Journal of Financial Economics, vol. 128, no. 2, pp. 234–252, 2018. PDF
5. L. Pastor and R. Stambaugh, ‘Liquidity Risk and Expected Stock Returns’, Journal of Political Economy, vol. 111, no. 3, pp. 642–685, 2003. PDF
6. R. F. Stambaugh and Y. Yuan, ‘Mispricing Factors’, The Review of Financial Studies, vol. 30, no. 4, pp. 1270–1315, 12 2016. PDF
7. K. Hou, H. Mo, C. Xue, and L. Zhang, ‘Which Factors?’, National Bureau of Economic Research, Inc, 2014. PDF
8. K. Hou, H. Mo, C. Xue, and L. Zhang, ‘An Augmented q-Factor Model with Expected Growth*’, Review of Finance, vol. 25, no. 1, pp. 1–41, 02 2020. PDF
9. Z. He, B. Kelly, and A. Manela, ‘Intermediary asset pricing: New evidence from many asset classes’, Journal of Financial Economics, vol. 126, no. 1, pp. 1–35, 2017. PDF
10. K. Daniel, D. Hirshleifer, and L. Sun, ‘Short- and Long-Horizon Behavioral Factors’, Review of Financial Studies, vol. 33, no. 4, pp. 1673–1736, 2020. PDF
11. C. Asness and A. Frazzini, ‘The Devil in HML’s Details’, The Journal of Portfolio Management, vol. 39, pp. 49–68, 2013. PDF
12. F. Barillas and J. Shanken, ‘Comparing Asset Pricing Models’, Journal of Finance, vol. 73, no. 2, pp. 715–754, 2018. PDF

Data sources:

• K. French, "Data Library," Tuck School of Business at Dartmouth. Link
• R. Stambaugh, "Liquidity" and "Mispricing" factor datasets, Wharton School, University of Pennsylvania. Link
• Z. He, "Intermediary Capital Ratio and Risk Factor" dataset, University of Chicago. Link
• K. Hou, G. Xue, R. Zhang, "The Hou-Xue-Zhang q-factors data library," at global-q.org. Link
• AQR Capital Management's Data Sets.
• Lin Sun, DHS Behavioural factors Link

(back to top)

License

The code in this project is released under the MIT License.

Known issues

• The first hml_devil_factors() retrieval is slow, because the download from aqr.com is slow. It's the only model implementing a cache—daily data expires at the end of the day, and will only re-download when the requested end_date exceeds the cache's latest index date. Similar for monthly, expiring at at the end of the month, and re-downloaded when next needed.

Todo

• Docs
  • Examples
• Tests
• Error handling