mpt.py

import numpy as np
from scipy.sparse import rand   # many 0s in our weight matrix (smaller portfolios)
from create_dataset import DatasetCreator
import pandas as pd
import matplotlib.pyplot as plt


if __name__ == '__main__':
    div_cash_flow_period = 5
    div_growth_period = 5   # valid values = 3, 5, 10

    # takes a long time, only mark true if paramters above have changed OR the mpt_data.csv dataset is missing
    create_dataset = False

    data = None
    if create_dataset:
        creator = DatasetCreator(div_cash_flow_period, div_growth_period)
        data = creator.create_dataset()
        data.to_csv("./mpt_data.csv", index=False)
    else:
        data = pd.read_csv('./mpt_data.csv')

    number_portfolios = 50000
    portfolio_returns = []  # traditionally price changes
    portfolio_risk = []     # traditionally volatility
    portfolio_weights = []

    ind_returns = data.expected_return
    ind_risk = data.risk

    # rather than work with percentages, this should give us ~number_stocks_in_port stocks
    # in each portfolio
    number_stocks_in_port = 10
    weight_density = number_stocks_in_port / len(data) 
    print("Generating Portfolios...")
    for portfolio in range(number_portfolios):
        weights = rand(len(data), 1, density=weight_density)     # generates a vector of mostly 0s
        weights = weights.A
        weights = weights / np.sum(weights) # normalize weights (sum = 1)
        portfolio_weights.append(weights)
        returns = np.dot(weights.T, ind_returns)  # each stock weight * expected return, summed up
        portfolio_returns.append(returns[0])

        # since volatility isn't used, the risk is just the calculated risk (based on payout ratio
        # and company solvency, i.e. dividend safety) mulitplied by the stock's weight
        risk = np.dot(weights.T, ind_risk)
        portfolio_risk.append(risk[0])

    portfolio_data = {'return': portfolio_returns, 'risk': portfolio_risk}
    for counter, symbol in enumerate(data.ticker.to_list()):
        portfolio_data[symbol + ' weight'] = [w[counter][0] for w in portfolio_weights]
    portfolios = pd.DataFrame(portfolio_data)
    portfolios.to_csv('./portfolios.csv', index=False)  # for future processing

    # save lowest risk portfolios (just selected stocks)
    portfolios = portfolios.sort_values(by=["risk"], ascending=True)
    portfolios.head(5).to_csv("./lowest_risk.csv", index=False)

    # save highest return (just selected stocks)
    portfolios = portfolios.sort_values(by=["return"], ascending=False)
    portfolios.head(5).to_csv("./highest_return.csv", index=False)

    portfolios.plot.scatter(x="risk", y="return", marker='o', color='g', s=15, alpha=0.5, grid=True)
    plt.xlabel("Risk (Dividend Safety)")
    plt.ylabel(f"Return ({div_cash_flow_period} year cash flow)")
    plt.title(f"{number_portfolios} Portfolios")
    plt.show()