import pandas as pd
from concurrent.futures import ProcessPoolExecutor
from functools import partial
from scipy.optimize import minimize
from core.utility import draw_sample, sortino
""" Find the best weights of instruments in a portfolio """
def bootstrap(portfolio, n=1500, costs=True, **kw):
data = portfolio.curve(portfolio_weights=1,capital=10E7).returns()
sample_length = 300
samples = [draw_sample(data, length=sample_length).index for k in range(0,n)]
weights = list(mp_optimize_weights(samples, data, **kw))
weights_buffer = pd.DataFrame([x for x in weights if type(x) == pd.Series])
print(len(weights_buffer),"samples")
weights_buffer.mean().plot.bar()
return weights_buffer
def mp_optimize_weights(samples, data, **kw):
return ProcessPoolExecutor().map(partial(optimize_weights, data), samples)
def optimize_weights(data, sample):
data = data.loc[sample].dropna(axis=1, how='all')
if data.shape[1] < 2:
return
guess = [1] * data.shape[1]
bounds = [(0.0, 10.0)] * data.shape[1]
def function(w, data, sample):
wr = (w*data.loc[sample]).sum(axis=1)
return -sortino(wr)
result = minimize(function, guess, (data, sample),\
method = 'SLSQP',\
bounds = bounds,\
tol = 0.0001,\
constraints = {'type': 'eq', 'fun': lambda x: data.shape[1] - sum(x)},\
options = {'eps': 1e-1},
)
return pd.Series(result.x, index=data.columns)
