The PortSort class enables single, double or triple sorting of entities in portfolios. Construction of portfolios based on firm characteristics is possible with the ff_portfolios() method. Sorts can be conditional or uncondtional or a mix of both in triple sorting.
pip install portsortThe PortSort_Example.py file is used to demonstrate the functionality of the PortSort class and its methods.
A panel dataset of the characteristics of 800 random US domiciled and traded securities for 2018-2020 and a dataset of monthly returns are used for the example.
The 'FirmCharacteristics2018.csv' dataset has 8 columns:
- 'year' : Calendar year
- 'notPERMNO' : Randomized security identifier based on the true CRSP PERMNO
- 'CAP' : Market Capitalization of the last business day of the current year
- 'CAP_W' : Market Capitalization of the last business day of the previous year
- 'RET_total' : Total return (including dividends) of a stock in a year
- 'SPREAD_PC_median' : Daily median of the ratio of the Bid-Ask spread over the closing price of the stock in a year
- 'FF30' : Fama-French 30 industry classification of a security based on its SIC.
- 'EXCHCD' : Market exchange in which the security is traded. 1 is NYSE, 2 is AMEX and 3 is NASDAQ.
The 'STOCKmonthlydata2019.csv' dataset has 4 columns:
- 'date_m': Date in YYYYmm format.
- 'year' : Calendar year
- 'RET' : Returns of securities in monthly frequency (not in percentage).
- 'notPERMNO' : Randomized security identifier based on the true CRSP PERMNO
"""
Examples of the PortSort class and its methods.
"""
import os
import pandas as pd
import numpy as np
# Main directory
wdir = r'C:\Users\ropot\OneDrive\Desktop\PortSort-main'
os.chdir(wdir)
# Import the PortSort class. For more details:
# https://github.com/ioannisrpt/portsort.git
# pip install portsort
from portsort import portsort as ps
# Import FirmCharacteristics table (annual frequency)
ftotype32 = {'year' : np.int32,
'CAP' : np.float32,
'CAP_W' : np.float32,
'RET_total' : np.float32,
'SPREAD_PC_median' : np.float32,
'EXCHCD' : np.int32,
'notPERMNO' : np.int32}
firmchars = pd.read_csv(os.path.join(wdir, 'FirmCharacteristics2018.csv')).astype(ftotype32)
# Import return data (monthly frequency)
ctotype32 = {'RET' : np.float32,
'date_m' : np.int32,
'year' : np.int32,
'notPERMNO' : np.int32}
crspm = pd.read_csv(os.path.join(wdir, 'STOCKmonthlydata2019.csv')).astype(ctotype32)
# Define the PortSort class
portchar = ps.PortSort(df = firmchars,
entity_id = 'notPERMNO',
time_id = 'year',
save_dir = wdir)
# -------------
# single_sort()
# -------------
# Single sort stocks into quintile portfolios based on the market
# capitalization of the last business day of the previous year ('CAP')
portchar.single_sort(firm_characteristic = 'CAP',
lagged_periods = 1,
n_portfolios = 5)
print(portchar.single_sorted.head(20))
# Single sort stocks into 3 portfolios (30%, 40% 30%) based on the market
# capitalization of the last business day of the previous year ('CAP').
# NYSE breakpoints for size are used.
portchar.single_sort(firm_characteristic = 'CAP',
lagged_periods = 1,
n_portfolios = np.array([0, 0.3, 0.7]),
quantile_filter = ['EXCHCD', 1])
print(portchar.single_sorted.head(20))
# -------------
# double_sort()
# -------------
# Double sort stocks unconditionally into 5x2 portfolios based on the market
# capitalization of the last business day of the previous year ('CAP') and the
# total annual return of the past year ('RET_total').
portchar.double_sort(firm_characteristics = ['CAP', 'RET_total'],
lagged_periods = [1,1],
n_portfolios = [5,2])
print(portchar.double_sorted.head(20))
# Double sort stocks conditionally into 3x2 portfolios based on the market
# capitalization of the last business day of the previous year ('CAP') and the
# total annual return of the past year ('RET_total').
# NYSE breakpoints for size are used.
portchar.double_sort(firm_characteristics = ['CAP', 'RET_total'],
lagged_periods = [1,1],
n_portfolios = [np.array([0, 0.3, 0.7]), 2],
quantile_filters = [['EXCHCD', 1], None],
conditional = True)
print(portchar.double_sorted.head(20))
# -------------
# triple_sort()
# -------------
# Triple Sort stocks unconditionally into 2x2x2 portfolios based on the market
# capitalization of the last business day of the previous year ('CAP'), total
# annual return ('RET_total') and daily median spread percentage
# ('SPREAD_PC_median') of the past year.
# NYSE breakpoitns are used for size and spread percentage
# but not for total return.
portchar.triple_sort(firm_characteristics=['CAP', 'RET_total', 'SPREAD_PC_median'],
lagged_periods = [1,1,1],
n_portfolios = [2,2,2],
quantile_filters = [['EXCHCD', 1], None, ['EXCHCD', 1]])
print(portchar.triple_sorted.head(20))
# Triple Sort stocks into 2x2x2 portfolios based on the market capitalization
# of the last business day of the previous year ('CAP'), total annual return
# ('RET_total') and daily median spread percentage ('SPREAD_PC_median') of the
# past year.
# First stocks are uncondtionally sorted by size and total annual return and
# then within these portfolios they are conditionally sorted by spread.
# If A, B, C are the characteristics in that order and '+', '|' correspond
# to intersection and conditionality of sets, then conditional = [False, True]
# is equivalent to C|(A+B).
# Type help(PortSort.triple_sort) for more details.
# NYSE breakpoitns are used for size and spread percentage
# but not for total return.
portchar.triple_sort(firm_characteristics=['CAP', 'RET_total', 'SPREAD_PC_median'],
lagged_periods = [1,1,1],
n_portfolios = [2,2,2],
quantile_filters = [['EXCHCD', 1], None, ['EXCHCD', 1]],
conditional = [False, True])
print(portchar.triple_sorted.head(20))
# Triple Sort stocks into 2x2x2 portfolios based on the market capitalization
# of the last business day of the previous year ('CAP'), total annual return
# ('RET_total') and daily median spread percentage ('SPREAD_PC_median') of the
# past year. Entities conditional on size, are then sorted into 2x2
# unconditional return and spread portfolios.
# If A, B, C are the characteristics in that exact order and
# '+', '|' correspond to intersection and conditionality of sets,
# then conditional = [True, False] is equivalent to (B + C)| A.
# Type help(PortSort.triple_sort) for more details.
# NYSE breakpoitns are used for size and spread percentage
# but not for total return.
portchar.triple_sort(firm_characteristics=['CAP', 'RET_total', 'SPREAD_PC_median'],
lagged_periods = [1,1,1],
n_portfolios = [2,2,2],
quantile_filters = [['EXCHCD', 1], None, ['EXCHCD', 1]],
conditional = [True, False])
print(portchar.triple_sorted.head(20))
# ---------------------
# augment_last_traded()
# ---------------------
# First we adjust for delisted returns during the calendar
# year. PortSort handles the firm characteristics and the
# return dataframe separately and only merge them together
# at the very end for the calculation of portfolio returns.
# As such, we need to augment the characteristics dataset
# with the data for stocks that are delisted but need to be
# included in the sorting procedure. augment_last_traded()
# method allows for that adjustment while it fills the extra
# rows with the weighting variable 'CAP_W' and the exchange
# market 'EXCHCD'. If we don't adjust for the delistings,
# our results will suffer from look-ahead bias.
portchar.augment_last_traded(ret_data = crspm,
ret_time_id = 'date_m',
col_w='CAP',
col_w_lagged_periods=1,
col_w_suffix = 'W',
fill_cols=['EXCHCD'])
# ---------------
# ff_portfolios()
# ---------------
# Monthly returns of 10 value-weighted portfolios on size ('CAP').
# NYSE breakpoints are used.
portchar.ff_portfolios(ret_data = crspm,
ret_time_id = 'date_m',
ff_characteristics = ['CAP'],
ff_lagged_periods = [1],
ff_n_portfolios = [10],
ff_quantile_filters = [['EXCHCD',1]],
weight_col = 'CAP_W',
return_col = 'RET',
ff_save = True)
print(portchar.portfolios.head(30))
print(portchar.num_stocks)
# Monthly returns of 3x2 value-weighted portfolios on size ('CAP') and
# liquidity ('SPREAD_PC_median').
# The sort is unconditional and NYSE breakpoints are used for size.
# By specifying the market_cap_cols, the portfolio turnover is also returned.
# market_cap_cols is a list =
# [capitalization of the stock at the end of the previous period,
# capitalization of the stock at the ned of the current period]
portchar.ff_portfolios(ret_data = crspm,
ret_time_id = 'date_m',
ff_characteristics = ['CAP', 'SPREAD_PC_median'],
ff_lagged_periods = [1, 1],
ff_n_portfolios = [np.array([0, 0.3, 0.7]),2],
ff_quantile_filters = [['EXCHCD',1], None],
ff_conditional = [False],
weight_col = 'CAP_W',
return_col = 'RET',
market_cap_cols = ['CAP_W', 'CAP'],
ff_save = True)
print(portchar.portfolios.head(30))
print(portchar.num_stocks)
print(portchar.turnover)
print('Acess the explicit portfolio weights of the stocks: \n')
print(portchar.turnover_raw.head(20))
# Monthly returns of 2x2x2 value-weighted portfolios on size ('CAP'),
# liquidity ('SPREAD_PC_median') and annual returns ('RET_total') of the
# previous year.
# The sorts are all conditional (the order matters).
# NYSE breakpoints are used only for size.
portchar.ff_portfolios(ret_data = crspm,
ret_time_id = 'date_m',
ff_characteristics = ['CAP', 'SPREAD_PC_median', 'RET_total'],
ff_lagged_periods = [1,1,1],
ff_n_portfolios = [2,2,2],
ff_quantile_filters = [['EXCHCD',1], None, None],
ff_conditional = [True, True],
weight_col = 'CAP_W',
return_col = 'RET',
market_cap_cols = ['CAP_W', 'CAP'],
ff_save = True)
print(portchar.portfolios.head(30))
print(portchar.num_stocks)
print(portchar.turnover)
# Monthly returns of 2x2x2 value-weighted portfolios on size ('CAP'),
# liquidity ('SPREAD_PC_median') and annual returns ('RET_total') of the
# previous year.
# Entities conditional on size, are then sorted into 2x2 unconditional return
# and spread portfolios.
# If A, B, C are the characteristics in that exact order and
# '+', '|' correspond to intersection and conditionality of sets,
# then conditional = [True, False] is equivalent to (B + C)| A.
# Type help(PortSort.ff_portfolios) for more details.
# NYSE breakpoints are used for size.
portchar.ff_portfolios(ret_data = crspm,
ret_time_id = 'date_m',
ff_characteristics=['CAP', 'SPREAD_PC_median', 'RET_total'],
ff_lagged_periods = [1,1,1],
ff_n_portfolios = [2,2,2],
ff_quantile_filters = [['EXCHCD',1], None, None],
ff_conditional = [True, False],
weight_col = 'CAP_W',
return_col = 'RET',
market_cap_cols = ['CAP_W', 'CAP'],
ff_save = True)
print(portchar.portfolios.head(30))
print(portchar.num_stocks)
print(portchar.turnover)