Weight_Initialization.ipynb

Kernel: finrl

Installation Setup

%load_ext autoreload
%autoreload 2

import numpy as np
import pandas as pd
import dask.dataframe as dd
from scipy import stats
import statsmodels.api as sm
from statsmodels.regression.rolling import RollingOLS
import matplotlib.pyplot as plt
from matplotlib import ticker
import seaborn as sns; sns.set()
from pandas.tseries.offsets import *
from dateutil.relativedelta import *
import datetime as dt
import os
from linearmodels.asset_pricing import TradedFactorModel, LinearFactorModel
from IPython.core.pylabtools import figsize
from IPython.core.interactiveshell import InteractiveShell
from fredapi import Fred
fred = Fred(api_key = 'b0363f9c9d853b92b27e06c4727bc2ea')
import pandas_datareader.data as web

%matplotlib inline 
%pylab inline
pylab.rcParams['figure.figsize'] = (20,10)

%pylab is deprecated, use %matplotlib inline and import the required libraries.
Populating the interactive namespace from numpy and matplotlib

C:\Users\kentw\AppData\Local\Continuum\anaconda3\envs\finrl\lib\site-packages\IPython\core\magics\pylab.py:162: UserWarning: pylab import has clobbered these variables: ['WE', 'MO', 'SU', 'FR', 'SA', 'TH', 'TU']
`%matplotlib` prevents importing * from pylab and numpy
  warn("pylab import has clobbered these variables: %s"  % clobbered +

import pickle
from multiprocessing import Pool 
import random
import json
import sys
import StockPortfolioEnv

import pytz
import itertools
from datetime import datetime as dt
from finrl.meta.preprocessor.yahoodownloader import YahooDownloader
from finrl.meta.preprocessor.preprocessors import FeatureEngineer, data_split
from finrl import config
from finrl import config_tickers
from finrl.config import (
    DATA_SAVE_DIR,
    TRAINED_MODEL_DIR,
    TENSORBOARD_LOG_DIR,
    RESULTS_DIR,
    INDICATORS,
    TRAIN_START_DATE,
    TRAIN_END_DATE,
    TEST_START_DATE,
    TEST_END_DATE,
    TRADE_START_DATE,
    TRADE_END_DATE,
)

if not os.path.exists("./" + config.RESULTS_DIR):
    os.makedirs("./" + config.RESULTS_DIR)

InteractiveShell.ast_node_interactivity = "all"
pd.options.display.float_format = '{:,.3f}'.format
pd.set_option('mode.use_inf_as_na', True)
pd.set_option('display.max_columns', 300)
pd.set_option('display.max_rows', 500)
idx = pd.IndexSlice
import warnings
warnings.filterwarnings('ignore')

Load Data

df_merged = pd.read_csv('data/merged.csv')
df_merged

Weight Initialization

Retail Weights (Rank-based method)

# Compute a set of weights for asset allocation
df_merged['moribvol'] = df_merged.groupby(['date'])['moribvol'].rank(method='dense')
df_merged['moribvol'] = df_merged.groupby('date')['moribvol'].apply(lambda x: x/x.sum())

df_merged 

Mean-Variance Optimization Weights

#TODO: Compute and append mean variance weights to "df_merged" here, say column name as "mean-var"

Data Split

states = ['date', 'open', 'high', 'low', 'close', 'volume', 'tic', 'day', 'macd', 'boll_ub', 'boll_lb', 'rsi_30', 'cci_30', 'dx_30', 'close_30_sma', 'close_60_sma', 'moribvol']

train_data = data_split(df_merged[states], '2007-01-01', '2018-01-01')
trade_data = data_split(df_merged[states], '2018-01-01', '2022-01-01')

# boom bust?

train_data

trade_data

# Last, save both train and trade dfs to csv in data foler
# TODO: add more features in "state" if needed
train_data.to_csv('data/train_data.csv', index=True)  
trade_data.to_csv('data/trade_data.csv', index=True) 

Benchmarks

We compare the performance of different weighting methods on the train period

• Mean Variance

• Equally weighted (Buy and hold)

• Market indexes (NASDAQ and XLK)

• Individual stocks

Environment configuration

A gym-style portfolio allocation environment for agents to interact. It is handy to compare the performances.

train = train_data
trade = trade_data

stock_dimension = len(train.tic.unique())
state_space = stock_dimension
tech_indicator_list = ['macd', 'rsi_30', 'cci_30', 'dx_30']
feature_dimension = len(tech_indicator_list)

print(f"Stock Dimension: {stock_dimension}, State Space: {state_space}")
print(f"Feature Dimension: {feature_dimension}")

env_kwargs = {
    "hmax": 100, 
    "initial_amount": 1000000, 
    "transaction_cost_pct": 0, 
    "state_space": state_space, 
    "stock_dim": stock_dimension, 
    "tech_indicator_list": tech_indicator_list, 
    "action_space": stock_dimension, 
    "reward_scaling": 1e-1
}

e_train_gym = StockPortfolioEnv.StockPortfolioEnv(df = train, **env_kwargs)
e_trade_gym = StockPortfolioEnv.StockPortfolioEnv(df = trade, **env_kwargs)

Stock Dimension: 11, State Space: 11
Feature Dimension: 4

Sampling

retail_train = StockPortfolioEnv.sample_from_env(i=0, env=e_train_gym, weights=train['moribvol'])

=================================
begin_total_asset:1000000
end_total_asset:4904419.075973194
Sharpe:  0.6729768484476762
=================================

# TODO: to be changed to excecute like the above function

# Mean variance
mean_var_df = pd.read_csv('../data/mean_var_weight.csv')
mean_var_df = pd.concat([mean_var_df.iloc[[0]], mean_var_df], ignore_index=True)
mean_var_df = pd.concat([mean_var_df, mean_var_df.iloc[[len(mean_var_df)-1]]], ignore_index=True)
mean_var_df.loc[0, 'date'] = '2007-01-03'
mean_var_df.loc[len(mean_var_df)-1, 'date'] = '2017-12-29'
mean_var_df = mean_var_df.drop(mean_var_df.columns[0], axis=1)

mean_var_dataset = StockPortfolioEnv.sample_from_env(i=0, env=e_train_gym, weights=mean_var_df.values)
mean_var_cum_ret = pd.DataFrame({'ret': 1000000 + np.insert(mean_var_dataset['rewards'].cumsum(), 0, 0, axis=0)})
mean_var_cum_ret['ret'] /= 1000000

=================================
begin_total_asset:1000000
end_total_asset:2954456.9590735934
Sharpe:  0.58838920208855
=================================

# XLK
yf_xlk = YahooDownloader(start_date = start,
                         end_date = end,
                         ticker_list = ["XLK"]).fetch_data()

yf_xlk = fe.preprocess_data(yf_xlk)
yf_xlk = yf_xlk.copy()
yf_xlk = yf_xlk.fillna(0)
yf_xlk = yf_xlk.replace(np.inf,0)
# yf_xlk['date'] = pd.to_datetime(yf_xlk['date'])
yf_xlk = data_split(yf_xlk, '2007-01-01', '2018-01-01')  
yf_xlk['ret'] = yf_xlk['open'] / yf_xlk['open'].iloc[0]

[*********************100%***********************]  1 of 1 completed
Shape of DataFrame:  (4074, 8)
Successfully added technical indicators
Successfully added turbulence index

# individual
grouped = df_merged.groupby('tic')
selected_cum_ret = grouped.apply(lambda x: x['open']/x['open'].iloc[0])
selected_cum_ret = selected_cum_ret.reset_index()
selected_cum_ret = selected_cum_ret.set_index('level_1')
selected_cum_ret = selected_cum_ret.join(df_merged[['date']])
selected_cum_ret = data_split(selected_cum_ret, '2007-01-01', '2018-01-01')

Performance Comparison

fig, ax = plt.subplots()

# Equal
equal_cum_ret = pd.read_csv("./results/equal_cumulative_reward.csv", names=['ret'])
equal_cum_ret['date'] = selected_cum_ret['date'].unique()
ax = equal_cum_ret.plot(ax=ax, kind='line', x='date', y='ret', label="Equal")

# Mean Variance
mean_var_cum_ret['date'] = selected_cum_ret['date'].unique()
ax = mean_var_cum_ret.plot(ax=ax, kind='line', x='date', y='ret', label="Mean Var")

# Retail
retail_cum_ret = pd.read_csv("results/retail_cumulative_reward.csv", names=['ret'])
retail_cum_ret['date'] = selected_cum_ret['date'].unique()
ax = retail_cum_ret.plot(ax=ax, kind='line', x='date', y='ret', label="Retail")

# XLK
ax = yf_xlk.plot(ax=ax, kind='line', x='date', y='ret', label="XLK")

plt.show()

fig, ax = plt.subplots()

# 个股
for key, grp in selected_cum_ret.groupby(['tic']):
    ax = grp.plot(ax=ax, kind='line', x='date', y='open', label=key)
    
plt.show()