[FinRL] CH3. 플랫폼 구성-(1)Data

Financial Reinforcement Learning(FinRL) Platform

Contents

1. Installation
2. Framework Overview
3. Main Component
   • Data
   • Train
   • Backtest
4. Examples

 

CH3. Main Component - (1)Data

Introduction

• 강화 학습을 위해 yahoo finance API를 통해 주가 데이터를 다운로드 후 데이터 가공을 진행 
• API 로드(Part1): 각종 코드에서 사용될 API를 설치하고 Import
• 데이터 다운로드(Part2): yahoo finance API를 통해 OHLCV* 데이터를 다운로드
• 데이터 가공(Part3): OHLCV* 데이터에 MACD, RSI, Turbulence Index를 추가 및 가공   
• 데이터 저장(Part4): 가공된 데이터를 CSV 파일로 저장

*OHLCV: Open, High, Low, Close, Volume

 

GitHub - AI4Finance-Foundation/FinRL-Tutorials: Tutorials. Please star.

---

Stock NeurIPS2018 Part 1. Data¶

This series is a reproduction of paper the process in the paper Practical Deep Reinforcement Learning Approach for Stock Trading.

This is the first part of the NeurIPS2018 series, introducing how to use FinRL to fetch and process data that we need for ML/RL trading.

Other demos can be found at the repo of FinRL-Tutorials).

Part 1. Install Packages¶

In [ ]:

## install required packages
!pip install swig
!pip install wrds
!pip install pyportfolioopt
## install finrl library
!pip install git+https://github.com/AI4Finance-Foundation/FinRL.git

In [ ]:

import pandas as pd
import numpy as np
import datetime
import yfinance as yf

from finrl.meta.preprocessor.yahoodownloader import YahooDownloader
from finrl.meta.preprocessor.preprocessors import FeatureEngineer, data_split
from finrl import config_tickers
from finrl.config import INDICATORS

import itertools

Part 2. Fetch data¶

yfinance is an open-source library that provides APIs fetching historical data form Yahoo Finance. In FinRL, we have a class called YahooDownloader that use yfinance to fetch data from Yahoo Finance.

OHLCV: Data downloaded are in the form of OHLCV, corresponding to open, high, low, close, volume, respectively. OHLCV is important because they contain most of numerical information of a stock in time series. From OHLCV, traders can get further judgement and prediction like the momentum, people's interest, market trends, etc.

Data for a single ticker¶

Here we provide two ways to fetch data with single ticker, let's take Apple Inc. (AAPL) as an example.

Using yfinance¶

In [ ]:

aapl_df_yf = yf.download(tickers = "aapl", start='2023-01-01', end='2023-12-31')

[*********************100%%**********************]  1 of 1 completed

In [ ]:

aapl_df_yf.head()

Out[ ]:

	Open	High	Low	Close	Adj Close	Volume
Date
2023-01-03	130.279999	130.899994	124.169998	125.070000	124.374802	112117500
2023-01-04	126.889999	128.660004	125.080002	126.360001	125.657639	89113600
2023-01-05	127.129997	127.769997	124.760002	125.019997	124.325081	80962700
2023-01-06	126.010002	130.289993	124.889999	129.619995	128.899521	87754700
2023-01-09	130.470001	133.410004	129.889999	130.149994	129.426575	70790800

Using FinRL¶

In FinRL's YahooDownloader, we modified the data frame to the form that convenient for further data processing process. We use adjusted close price instead of close price, and add a column representing the day of a week (0-4 corresponding to Monday-Friday).

In [ ]:

aapl_df_finrl = YahooDownloader(start_date = '2023-01-01',
                                end_date = '2023-12-31',
                                ticker_list = ['aapl']).fetch_data()

[*********************100%%**********************]  1 of 1 completed

Shape of DataFrame:  (250, 8)

In [ ]:

aapl_df_finrl.head()

Out[ ]:

	date	open	high	low	close	volume	tic	day
0	2023-01-03	130.279999	130.899994	124.169998	124.374802	112117500	aapl	1
1	2023-01-04	126.889999	128.660004	125.080002	125.657639	89113600	aapl	2
2	2023-01-05	127.129997	127.769997	124.760002	124.325081	80962700	aapl	3
3	2023-01-06	126.010002	130.289993	124.889999	128.899521	87754700	aapl	4
4	2023-01-09	130.470001	133.410004	129.889999	129.426575	70790800	aapl	0

Data for the chosen tickers¶

In [ ]:

config_tickers.DOW_30_TICKER

Out[ ]:

['AXP',
 'AMGN',
 'AAPL',
 'BA',
 'CAT',
 'CSCO',
 'CVX',
 'GS',
 'HD',
 'HON',
 'IBM',
 'INTC',
 'JNJ',
 'KO',
 'JPM',
 'MCD',
 'MMM',
 'MRK',
 'MSFT',
 'NKE',
 'PG',
 'TRV',
 'UNH',
 'CRM',
 'VZ',
 'V',
 'WBA',
 'WMT',
 'DIS',
 'DOW']

In [ ]:

TRAIN_START_DATE = '2009-01-01'
TRAIN_END_DATE = '2020-07-01'
TRADE_START_DATE = '2020-07-01'
TRADE_END_DATE = '2021-10-29'

In [ ]:

df_raw = YahooDownloader(start_date = TRAIN_START_DATE,
                     end_date = TRADE_END_DATE,
                     ticker_list = config_tickers.DOW_30_TICKER).fetch_data()

[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed
[*********************100%%**********************]  1 of 1 completed

Shape of DataFrame:  (94301, 8)

In [ ]:

df_raw.head()

Out[ ]:

	date	open	high	low	close	volume	tic	day
0	2009-01-02	3.067143	3.251429	3.041429	2.747390	746015200	AAPL	4
1	2009-01-02	58.590000	59.080002	57.750000	42.737900	6547900	AMGN	4
2	2009-01-02	18.570000	19.520000	18.400000	15.144921	10955700	AXP	4
3	2009-01-02	42.799999	45.560001	42.779999	33.941090	7010200	BA	4
4	2009-01-02	44.910000	46.980000	44.709999	31.093384	7117200	CAT	4

Part 3: Preprocess Data¶

We need to check for missing data and do feature engineering to convert the data point into a state.

• Adding technical indicators. In practical trading, various information needs to be taken into account, such as historical prices, current holding shares, technical indicators, etc. Here, we demonstrate two trend-following technical indicators: MACD and RSI.
• Adding turbulence index. Risk-aversion reflects whether an investor prefers to protect the capital. It also influences one's trading strategy when facing different market volatility level. To control the risk in a worst-case scenario, such as financial crisis of 2007–2008, FinRL employs the turbulence index that measures extreme fluctuation of asset price.

Hear let's take MACD as an example. Moving average convergence/divergence (MACD) is one of the most commonly used indicator showing bull and bear market. Its calculation is based on EMA (Exponential Moving Average indicator, measuring trend direction over a period of time.)

In [ ]:

fe = FeatureEngineer(use_technical_indicator=True,
                     tech_indicator_list = INDICATORS,
                     use_vix=True,
                     use_turbulence=True,
                     user_defined_feature = False)

processed = fe.preprocess_data(df_raw)

Successfully added technical indicators

[*********************100%%**********************]  1 of 1 completed

Shape of DataFrame:  (3228, 8)
Successfully added vix
Successfully added turbulence index

In [ ]:

list_ticker = processed["tic"].unique().tolist()
list_date = list(pd.date_range(processed['date'].min(),processed['date'].max()).astype(str))
combination = list(itertools.product(list_date,list_ticker))

processed_full = pd.DataFrame(combination,columns=["date","tic"]).merge(processed,on=["date","tic"],how="left")
processed_full = processed_full[processed_full['date'].isin(processed['date'])]
processed_full = processed_full.sort_values(['date','tic'])

processed_full = processed_full.fillna(0)

In [ ]:

processed_full.head()

Out[ ]:

	date	tic	open	high	low	close	volume	day	macd	boll_ub	boll_lb	rsi_30	cci_30	dx_30	close_30_sma	close_60_sma	vix	turbulence
0	2009-01-02	AAPL	3.067143	3.251429	3.041429	2.747390	746015200.0	4.0	0.0	2.969345	2.641386	100.0	66.666667	100.0	2.747390	2.747390	39.189999	0.0
1	2009-01-02	AMGN	58.590000	59.080002	57.750000	42.737900	6547900.0	4.0	0.0	2.969345	2.641386	100.0	66.666667	100.0	42.737900	42.737900	39.189999	0.0
2	2009-01-02	AXP	18.570000	19.520000	18.400000	15.144921	10955700.0	4.0	0.0	2.969345	2.641386	100.0	66.666667	100.0	15.144921	15.144921	39.189999	0.0
3	2009-01-02	BA	42.799999	45.560001	42.779999	33.941090	7010200.0	4.0	0.0	2.969345	2.641386	100.0	66.666667	100.0	33.941090	33.941090	39.189999	0.0
4	2009-01-02	CAT	44.910000	46.980000	44.709999	31.093384	7117200.0	4.0	0.0	2.969345	2.641386	100.0	66.666667	100.0	31.093384	31.093384	39.189999	0.0

Part 4: Save the Data¶

Split the data for training and trading¶

In [ ]:

train = data_split(processed_full, TRAIN_START_DATE,TRAIN_END_DATE)
trade = data_split(processed_full, TRADE_START_DATE,TRADE_END_DATE)
print(len(train))
print(len(trade))

83897
9715

Save data to csv file¶

For Colab users, you can open the virtual directory in colab and manually download the files.

For users running on your local environment, the csv files should be at the same directory of this notebook.

In [ ]:

train.to_csv('train_data.csv')
trade.to_csv('trade_data.csv')