Example7_6.py

# Example 7-6
# # Backtesting the January Effect

import numpy as np
import pandas as pd

onewaytcost = 0.0005

df = pd.read_table("miscFiles/IJR_20080131.txt")
df["Date"] = df["Date"].round().astype("int")
df["Date"] = pd.to_datetime(df["Date"], format="%Y%m%d")
df.set_index("Date", inplace=True)

eoyPrice = df.resample("Y").last()[
    0:-1
]  # End of December prices. Need to remove last date because it isn't really end of year
annret = eoyPrice.pct_change().iloc[1:, :]  # first row has NaN

eojPrice = df.resample("BA-JAN").last()[
    1:-1
]  # End of January prices. Need to remove first date to match the years in lastdayofDec. Need to remove last date because it isn't really end of January.
janret = (eojPrice.values - eoyPrice.values) / eoyPrice.values
janret = janret[
    1:,
]  # match number of rows in annret

for y in range(len(annret)):
    hasData = np.where(np.isfinite(annret.iloc[y, :]))[0]
    sortidx = np.argsort(annret.iloc[y, hasData])
    topN = np.round(len(hasData) / 10)
    portRet = (
        np.nanmean(janret[y, hasData[sortidx.iloc[np.arange(0, topN)]]])
        - np.nanmean(janret[y, hasData[sortidx.iloc[np.arange(-topN + 1, -1)]]])
    ) / 2 - 2 * onewaytcost  # portfolio returns

    print(
        "Last holding date %s: Portfolio return=%f" % (eojPrice.index[y + 1], portRet)
    )


# Last holding date 2006-01-31 00:00:00: Portfolio return=-0.023853
# Last holding date 2007-01-31 00:00:00: Portfolio return=-0.003641
# Last holding date 2008-01-31 00:00:00: Portfolio return=0.088486