How Much Money We Can Make?

import numpy as np
import pandas as pd
import json
import os

import sys
sys.path.append("../../")

# import our modules
import lstm
import strat
import profit

Load Data

stocks: pd.DataFrame = pd.read_csv("../../data/stocks.csv", index_col=0, parse_dates=True)
stocks

	Company	Sector	Open	High	Low	Close	Volume
Date
2017-11-02	AAPL	Technology	41.650002	42.125000	41.320000	42.027500	165573600
2017-11-03	AAPL	Technology	43.500000	43.564999	42.779999	43.125000	237594400
2017-11-06	AAPL	Technology	43.092499	43.747501	42.930000	43.562500	140105200
2017-11-07	AAPL	Technology	43.477501	43.812500	43.400002	43.702499	97446000
2017-11-08	AAPL	Technology	43.665001	44.060001	43.582500	44.060001	97638000
...	...	...	...	...	...	...	...
2022-10-26	COP	Energy	124.720001	128.179993	124.580002	126.570000	8139100
2022-10-27	COP	Energy	127.699997	129.449997	126.239998	126.639999	8948500
2022-10-28	COP	Energy	128.500000	128.990005	124.010002	127.169998	7293200
2022-10-31	COP	Energy	125.580002	129.990005	125.570000	126.089996	7121000
2022-11-01	COP	Energy	128.740005	129.320007	126.889999	127.779999	5874800

63377 rows × 7 columns

Profit Rate for Each Company

profit_report = {}

companies: list[str] = json.load(open("../../data/companies.json", "r"))
num_days_left_out = 90
for company in companies:
    
    profit_dict = {}
    
    # select stock
    stock = stocks.query(f"Company == '{company}'").drop(columns=["Company", "Sector"])
    
    # load model
    price_predictor = lstm.load_price_predictor(os.path.join("../../models", f"{company}-3-day-predictor.pkl"))
    
    # test data
    stock_test = stock[-num_days_left_out:]
    start_date = stock_test.index[0]
    
    # decide when to buy and sell for SMA
    buy_dates, sell_dates = strat.sma(
        stock,long_period=20
    )
    
    # profit rate
    profit_rate = profit.calc_profit(stock, buy_dates, sell_dates, start_date)
    
    profit_dict["SMA"] = profit_rate
    
    # decide when to buy and sell for prediction-based strategy (LSTM)
    buy_dates, sell_dates = strat.trade_by_pred(
        stock_test, 
        price_predictor, 
        num_days_ahead=2,
    )
    
    # profit rate
    profit_rate = profit.calc_profit(stock, buy_dates, sell_dates, start_date)
    
    profit_dict["LSTM"] = profit_rate
    
    profit_report[company] = profit_dict
    
profit_report = pd.DataFrame(profit_report)
pd.options.display.float_format = "{:.2%}".format

profit_report

	META	TSLA	XOM	BAC	UNH	BA	AAPL
SMA	-19.97%	9.19%	-2.80%	-2.12%	0.77%	0.00%	17.77%
LSTM	9.44%	17.05%	2.06%	11.74%	0.19%	8.91%	15.18%

Note that sometimes applying SMA will lose money even though sometimes SMA has higher profit rate, e.g., UNH, AAPL.

The profit rate for LSTM is always positive for this data, which means we will always earn some money.

Total Profit

weight = strat.determine_portfolio(stocks, companies, num_days_left_out, random_seed=7008)
profit_report.loc["Weight"] = weight
profit_report.loc["Weighted SMA"] = profit_report.loc["Weight"] * profit_report.loc["SMA"]
profit_report.loc["Weighted LSTM"] = profit_report.loc["Weight"] * profit_report.loc["LSTM"]

capital = 10000
summary = profit_report.sum(1).loc[["Weighted SMA", "Weighted LSTM"]]
pd.options.display.float_format = "{:.2f}".format
summary * capital

Weighted SMA    -336.22
Weighted LSTM    510.95
dtype: float64

We will lose $336.22 by applying SMA, while we will earn $510.95 using LSTM!