pca_portfolio.py

import time
import yfinance
import dryscrape
from bs4 import BeautifulSoup
import numpy as np
import pandas as pd
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt

def get_market_data(tickers, start, end):
    market_data = yfinance.download(
        tickers = tickers,
        start=start,
        end=end,
        interval = '1d',
        treads = False
    )

    # Fills missing values. Some stocks are recent IPOs
    return (market_data['Close']
            .fillna(method='backfill')
            .fillna(method='ffill'))

def plot_pca_vs_ibov(log_ts, idx):
    rs_df = pd.concat([log_ts, idx], 1)
    rs_df.columns = ["PCA Portfolio", "IBOV"]

    crs_df = rs_df.cumsum().apply(np.exp)
    crs_df.plot()
    plt.show()

def plot_pca_vs_ibov_l10(idx, pcal10, ibovl10):
    rs_df = pd.concat([idx, pcal10, ibovl10], 1)
    rs_df.columns = ["IBOV", "PCA-L10", "IBOV-L10"]

    crs_df = rs_df.cumsum().apply(np.exp)
    crs_df.plot()
    plt.show()


# Loads Ibovespa tickers from file
composition = pd.read_pickle('composition.df')
w_ibov = composition['Part. (%)']

# Get market data
tickers = composition.index.tolist()
data_ts = get_market_data(tickers, '2020-09-23', '2021-09-23')
ibov_ts = get_market_data(['^BVSP'], '2020-09-23', '2021-09-23')

# Get normalized log-return
log_data = data_ts.apply(np.log).diff(1)[1:]
log_ibov = ibov_ts.apply(np.log).diff(1)[1:]

# Plot market returns
acc = log_data.cumsum().apply(np.exp)
acc.plot()
plt.show()

# PCA 1
pca = PCA(1).fit(log_data)
pc1 = pd.Series(index=log_data.columns, data=pca.components_[0])
weights = abs(pc1)/sum(abs(pc1))

# Compare Ibov vs PCA Portfolio
w_ibov = composition['Part. (%)']
print(weights.nlargest(10))
print(weights.nsmallest(10))
print(w_ibov.nlargest(10))
print(w_ibov.nsmallest(10))

# 
pca_ts = (weights*log_data).sum(1)
pcal10 = log_data[pc1.nlargest(10).index].mean(1)
il10 = w_ibov.nlargest(10) / sum(w_ibov.nlargest(10))
ibovl10 = (log_data[il10.index] * il10).sum(1)
idx = pd.Series(index=ibov_ts.index, data=log_ibov)

# Plot Ibov vs PCA Portfolio
plot_pca_vs_ibov(pca_ts, idx)
plot_pca_vs_ibov_l10(idx, pcal10, ibovl10)