def cumulative_returns(returns: pd.DataFrame, period, freq=None):
if not isinstance(period, pd.Timedelta):
period = pd.Timedelta(period)
if freq is None:
freq = returns.index.freq
if freq is None:
freq = pd.tseries.offsets.BDay()
warnings.warn(
"'freq' not set, using business day calendar",
UserWarning
)
trades_idx = returns.index.copy()
returns_idx = utils.add_custom_calendar_timedelta(trades_idx, period)
full_idx = trades_idx.union(returns_idx)
sub_returns = []
while len(trades_idx) > 0:
sub_idx = []
next = trades_idx.min()
while next <= trades_idx.max():
sub_idx.append(next)
next = utils.add_custom_calendar_timedelta(
input=next,
timedelta=period,
)
try:
i = trades_idx.get_loc(next, method="bfill")
next = trades_idx[i]
except KeyError:
break
sub_idx = pd.DatetimeIndex(sub_idx, tz=full_idx.tz)
subret = returns[sub_idx]
subret = subret.reindex(full_idx)
for pret_idx in reversed(sub_idx):
pret = subret[pret_idx]
pret_end_idx = utils.add_custom_calendar_timedelta(pret_idx, period)
slice = subret[(subret.index > pret_idx)
& (subret.index <= pret_end_idx)].index
if pd.isnull(pret):
continue
def rate_of_returns(ret, period):
return ((np.nansum(ret) + 1)**(1.0 / period)) - 1
for slice_idx in slice:
sub_period = utils.diff_custom_calendar_timedeltas(
pret_idx,
slice_idx,
freq
)
def factor_rank_autocorrelation(
factor_data: pd.DataFrame,
period: Union[int,
str] = 1
):
grouper = ["datetime"]
ranks = factor_data.groupby(grouper)["factor"].rank()
asset_factor_rank = ranks.reset_index().pivot(
index="datetime",
columns="code",
values="factor"
)
if isinstance(period, int):
asset_shifted = asset_factor_rank.shift(period)
else:
shifted_idx = utils.add_custom_calendar_timedelta(
asset_factor_rank.index,
-pd.Timedelta(period),
)
asset_shifted = asset_factor_rank.reindex(shifted_idx)
asset_shifted.index = asset_factor_rank.index
autocorr = asset_factor_rank.corrwith(asset_shifted, axis=1)
autocorr.name = period
return autocorr
def quantile_turnover(
quantile_factor: pd.DataFrame,
quantile: int,
period: Union[int,
str] = 1
):
"""
Computes the proportion of names in a factor quantile that were
not in that quantile in the previous period.
Parameters
----------
quantile_factor : pd.Series
DataFrame with date, asset and factor quantile.
quantile : int
Quantile on which to perform turnover analysis.
period: string or int, optional
Period over which to calculate the turnover. If it is a string it must
follow pandas.Timedelta constructor format (e.g. '1 days', '1D', '30m',
'3h', '1D1h', etc).
Returns
-------
quant_turnover : pd.Series
Period by period turnover for that quantile.
"""
quant_names = quantile_factor[quantile_factor == quantile]
quant_name_sets = quant_names.groupby(
level=["datetime"]
).apply(lambda x: set(x.index.get_level_values("code")))
if isinstance(period, int):
name_shifted = quant_name_sets.shift(period)
else:
period = utils.get_period(period)
shifted_idx = utils.add_custom_calendar_timedelta(
quant_name_sets.index,
-pd.Timedelta(period)
)
name_shifted = quant_name_sets.reindex(shifted_idx)
name_shifted.index = quant_name_sets.index
new_names = (quant_name_sets - name_shifted).dropna()
quant_turnover = new_names.apply(lambda x: len(x)
) / quant_name_sets.apply(lambda x: len(x))
quant_turnover.name = quantile
return quant_turnover