Strategy Analysis with Pandas TA and AI/ML

This is a Work in Progress and subject to change!
Contributions are welcome and accepted!
Examples below are for educational purposes only.
NOTE: The watchlist module is independent of Pandas TA. To easily use it, copy it from your local pandas_ta installation directory into your project directory.

Required Packages

Uncomment the packages you need to install or are missing

In [1]:

#!pip install numpy
#!pip install pandas
#!pip install mplfinance
#!pip install pandas-datareader
#!pip install requests_cache
#!pip install tqdm
#!pip install alphaVantage-api # Required for Watchlist

In [2]:

%pylab inline
import datetime as dt
import random as rnd
from sys import float_info as sflt

from tqdm import tqdm

import numpy as np
import pandas as pd
pd.set_option("max_rows", 100)
pd.set_option("max_columns", 20)

import mplfinance as mpf
import pandas_ta as ta

from tqdm.notebook import trange, tqdm

from watchlist import colors, Watchlist # Is this failing? If so, copy it locally. See above.

print(f"Numpy v{np.__version__}")
print(f"Pandas v{pd.__version__}")
print(f"mplfinance v{mpf.__version__}")
print(f"\nPandas TA v{ta.version}\nTo install the Latest Version:\n$ pip install -U git+https://github.com/twopirllc/pandas-ta\n")
%matplotlib inline

Populating the interactive namespace from numpy and matplotlib
Numpy v1.20.2
Pandas v1.2.4
mplfinance v0.12.7a12

Pandas TA v0.2.74b0
To install the Latest Version:
$ pip install -U git+https://github.com/twopirllc/pandas-ta

MISC Function(s)

In [3]:

def recent_bars(df, tf: str = "1y"):
    # All Data: 0, Last Four Years: 0.25, Last Two Years: 0.5, This Year: 1, Last Half Year: 2, Last Quarter: 4
    yearly_divisor = {"all": 0, "10y": 0.1, "5y": 0.2, "4y": 0.25, "3y": 1./3, "2y": 0.5, "1y": 1, "6mo": 2, "3mo": 4}
    yd = yearly_divisor[tf] if tf in yearly_divisor.keys() else 0
    return int(ta.RATE["TRADING_DAYS_PER_YEAR"] / yd) if yd > 0 else df.shape[0]

Data Collection

In [4]:

tf = "D"
tickers = ["SPY", "QQQ", "AAPL", "TSLA", "BTC-USD"]
watch = Watchlist(tickers, tf=tf, ds_name="yahoo", timed=True)
# watch.strategy = ta.CommonStrategy # If you have a Custom Strategy, you can use it here.
watch.load(tickers, analyze=True, verbose=False)

[!] Loading All: SPY, QQQ, AAPL, TSLA, BTC-USD
[+] Downloading[yahoo]: SPY[D]
[+] Saving: /Users/kj/av_data/SPY_D.csv
[i] Runtime: 476.9447 ms (0.4769 s)
[+] Downloading[yahoo]: QQQ[D]
[+] Saving: /Users/kj/av_data/QQQ_D.csv
[i] Runtime: 455.1703 ms (0.4552 s)
[+] Downloading[yahoo]: AAPL[D]
[+] Saving: /Users/kj/av_data/AAPL_D.csv
[i] Runtime: 460.0315 ms (0.4600 s)
[+] Downloading[yahoo]: TSLA[D]
[+] Saving: /Users/kj/av_data/TSLA_D.csv
[i] Runtime: 456.2752 ms (0.4563 s)
[+] Downloading[yahoo]: BTC-USD[D]
[+] Saving: /Users/kj/av_data/BTC-USD_D.csv
[i] Runtime: 451.6120 ms (0.4516 s)

Asset Selection

In [5]:

ticker = tickers[0] # change tickers by changing the index
print(f"{ticker} {watch.data[ticker].shape}\nColumns: {', '.join(list(watch.data[ticker].columns))}")

SPY (7108, 12)
Columns: open, high, low, close, volume, dividends, split, SMA_10, SMA_20, SMA_50, SMA_200, VOL_SMA_20

Trim it

In [6]:

duration = "1y"
asset = watch.data[ticker]
recent = recent_bars(asset, duration)
asset.columns = asset.columns.str.lower()
asset.drop(columns=["dividends", "split"], errors="ignore", inplace=True)
asset = asset.copy().tail(recent)
asset

Out [6]:

	open	high	low	close	volume	sma_10	sma_20	sma_50	sma_200	vol_sma_20
date
2020-04-22	273.956662	276.564830	272.539388	274.694824	93524600	274.489117	263.075163	277.084857	293.063143	167661845.0
2020-04-23	276.062859	279.458417	274.350332	274.675110	104709700	274.986145	264.664179	276.017570	292.996336	157925815.0
2020-04-24	276.299118	279.222243	274.104305	278.503754	85166000	275.455618	265.735499	274.984586	292.941791	149302475.0
2020-04-27	280.619817	283.720094	280.127709	282.519348	77896600	276.576642	267.390459	274.038959	292.903932	141980245.0
2020-04-28	286.426668	286.800675	280.895376	281.220184	105270000	276.767578	268.575457	273.056781	292.853083	138675270.0
...	...	...	...	...	...	...	...	...	...	...
2021-04-15	413.739990	416.160004	413.690002	415.869995	60229800	409.150995	400.666098	392.662884	356.918501	83970505.0
2021-04-16	417.250000	417.910004	415.730011	417.260010	82004500	410.815997	402.018999	393.396016	357.480658	82315105.0
2021-04-19	416.260010	416.739990	413.790009	415.209991	78498500	411.700998	403.305498	394.001630	358.021889	80601140.0
2021-04-20	413.910004	415.089996	410.589996	412.170013	81851800	412.306000	404.284499	394.516144	358.539468	81004800.0
2021-04-21	411.510010	416.290009	411.359985	416.070007	66345300	413.254001	405.612999	395.052841	359.052723	79787735.0

252 rows × 10 columns

Trend Creation

A Trend is the result of some calculation or condition of one or more indicators. For simplicity, a Trend is either True or 1 and No Trend is False or 0. Using the Hello World of Trends, the Golden/Death Cross, it's Trend is Long when long = ma(close, 50) > ma(close, 200) and Short when short = ma(close, 50) < ma(close, 200) .

In [7]:

# Example Long Trends
# long = ta.sma(asset.close, 50) > ta.sma(asset.close, 200) # SMA(50) > SMA(200) "Golden/Death Cross"
# long = ta.sma(asset.close, 10) > ta.sma(asset.close, 20) # SMA(10) > SMA(20)
long = ta.ema(asset.close, 8) > ta.ema(asset.close, 21) # EMA(8) > EMA(21)
# long = ta.increasing(ta.ema(asset.close, 50))
# long = ta.macd(asset.close).iloc[:,1] > 0 # MACD Histogram is positive
# long = ta.amat(asset.close, 50, 200).AMATe_LR_2  # Long Run of AMAT(50, 200) with lookback of 2 bars

# long &= ta.increasing(ta.ema(asset.close, 50), 2) # Uncomment for further long restrictions, in this case when EMA(50) is increasing/sloping upwards
# long = 1 - long # uncomment to create a short signal of the trend

asset.ta.ema(length=8, sma=False, append=True)
asset.ta.ema(length=21, sma=False, append=True)
asset.ta.ema(length=50, sma=False, append=True)
asset.ta.percent_return(append=True)
print("TA Columns Added:")
asset[asset.columns[5:]].tail()

Out [7]:

TA Columns Added:

	sma_10	sma_20	sma_50	sma_200	vol_sma_20	EMA_8	EMA_21	EMA_50	PCTRET_1
date
2021-04-15	409.150995	400.666098	392.662884	356.918501	83970505.0	410.171168	402.758751	393.431087	0.010742
2021-04-16	410.815997	402.018999	393.396016	357.480658	82315105.0	411.746466	404.077047	394.365555	0.003342
2021-04-19	411.700998	403.305498	394.001630	358.021889	80601140.0	412.516138	405.089133	395.182984	-0.004913
2021-04-20	412.306000	404.284499	394.516144	358.539468	81004800.0	412.439222	405.732850	395.849142	-0.007322
2021-04-21	413.254001	405.612999	395.052841	359.052723	79787735.0	413.246063	406.672591	396.642117	0.009462

Trend Signals

Given a Trend, Trend Signals returns the Trend, Trades, Entries and Exits as boolean integers. When asbool=True, it returns Trends, Entries and Exits as boolean values which is helpful when combined with the vectorbt backtesting package.

In [8]:

trendy = asset.ta.tsignals(long, asbool=False, append=True)
trendy.tail()

Out [8]:

	TS_Trends	TS_Trades	TS_Entries	TS_Exits
date
2021-04-15	1	0	0	0
2021-04-16	1	0	0	0
2021-04-19	1	0	0	0
2021-04-20	1	0	0	0
2021-04-21	1	0	0	0

Trend Entries & Exits & Trade Table

This is a simple way to reduce the Asset DataFrame to a Trade Table with Dates, Signals, and Entries and Exits. Gives you an idea what to expect before running through a backtester such as vectorbt.

In [9]:

entries = trendy.TS_Entries * asset.close
entries = entries[~np.isclose(entries, 0)]
entries.dropna(inplace=True)
entries.name = "Entry"

exits = trendy.TS_Exits * asset.close
exits = exits[~np.isclose(exits, 0)]
exits.dropna(inplace=True)
exits.name = "Exit"

total_trades = trendy.TS_Trades.abs().sum()
rt_trades = int(trendy.TS_Trades.abs().sum() // 2)

all_trades = trendy.TS_Trades.copy().fillna(0)
all_trades = all_trades[all_trades != 0]

trades = pd.DataFrame({
    "Signal": all_trades,
    entries.name: entries,
    exits.name: exits
})

# Show some stats if there is an active trade (when there is an odd number of round trip trades)
if total_trades % 2 != 0:
    unrealized_pnl = asset.close.iloc[-1] - entries.iloc[-1]
    unrealized_pnl_pct_change = 100 * ((asset.close.iloc[-1] / entries.iloc[-1]) - 1)
    print("Current Trade:")
    print(f"Price Entry | Last:\t{entries.iloc[-1]:.4f} | {asset.close.iloc[-1]:.4f}")
    print(f"Unrealized PnL | %:\t{unrealized_pnl:.4f} | {unrealized_pnl_pct_change:.4f}%")
print(f"\nTrades Total | Round Trip:\t{total_trades} | {rt_trades}")
print(f"Trade Coverage: {100 * asset.TS_Trends.sum() / asset.shape[0]:.2f}%")

trades

Out [9]:

Current Trade:
Price Entry | Last:	385.9061 | 416.0700
Unrealized PnL | %:	30.1639 | 7.8164%

Trades Total | Round Trip:	7 | 3
Trade Coverage: 81.35%

	Signal	Entry	Exit
date
2020-05-19	1	287.361664	NaN
2020-09-10	-1	NaN	330.066162
2020-10-02	1	331.337799	NaN
2020-10-27	-1	NaN	335.684937
2020-11-04	1	340.965088	NaN
2021-03-02	-1	NaN	385.278137
2021-03-09	1	385.906067	NaN

In [ ]:

Visualization

Chart Display Strings

In [10]:

extime = ta.get_time(to_string=True)
first_date, last_date = asset.index[0], asset.index[-1]
f_date = f"{first_date.day_name()} {first_date.month}-{first_date.day}-{first_date.year}"
l_date = f"{last_date.day_name()} {last_date.month}-{last_date.day}-{last_date.year}"
last_ohlcv = f"Last OHLCV: ({asset.iloc[-1].open:.4f}, {asset.iloc[-1].high:.4f}, {asset.iloc[-1].low:.4f}, {asset.iloc[-1].close:.4f}, {int(asset.iloc[-1].volume)})"
ptitle = f"\n{ticker} [{tf} for {duration}({recent} bars)] from {f_date} to {l_date}\n{last_ohlcv}\n{extime}"

Trade Chart

In [11]:

# chart = asset["close"] #asset[["close", "SMA_10", "SMA_20", "SMA_50", "SMA_200"]]
# chart = asset[["close", "SMA_10", "SMA_20"]]
chart = asset[["close", "EMA_8", "EMA_21", "EMA_50"]]
chart.plot(figsize=(16, 10), color=colors("BkGrOrRd"), title=ptitle, grid=True)

Out [11]:

<AxesSubplot:title={'center':'\nSPY [D for 1y(252 bars)] from Wednesday 4-22-2020 to Wednesday 4-21-2021\nLast OHLCV: (411.5100, 416.2900, 411.3600, 416.0700, 66345300)\nWednesday April 21, 2021, NYSE: 17:08:21, Local: 21:08:21 PDT, Day 111/365 (30.00%)'}, xlabel='date'>

Long and Short Trends

Trends are either a Trend (1) or No Trend (0) depending on the Trend passed into *Trend Signals

In [12]:

long_trend = trendy.TS_Trends
short_trend = 1 - long_trend

long_trend.plot(figsize=(16, 0.85), kind="area", stacked=True, color=colors()[0], alpha=0.25) # Green Area
short_trend.plot(figsize=(16, 0.85), kind="area", stacked=True, color=colors()[1], alpha=0.25) # Red Area

Out [12]:

<AxesSubplot:xlabel='date'>

Trades or Trade Signals

The Trades are either Enter (1) or Exit (-1) or No Position/Action (0). These are based on the Trend passed into Trend Signals whether they are Long or Short Trends.

In [13]:

trendy.TS_Trades.plot(figsize=(16, 1.5), color=colors("BkBl")[0], grid=True)

Out [13]:

<AxesSubplot:xlabel='date'>

Active Returns

Active Returns are returns made during the course of the Trend. They are simply the product of the returns and the Trend

In [14]:

asset["ACTRET_1"] = trendy.TS_Trends * asset.PCTRET_1
asset[["PCTRET_1", "ACTRET_1"]].plot(figsize=(16, 3), color=colors("GyOr"), alpha=1, grid=True).axhline(0, color="black")

Out [14]:

<matplotlib.lines.Line2D at 0x12d86cbb0>

Buy and Hold Returns (PCTRET_1) vs. Cum. Active Returns (ACTRET_1)

In [15]:

asset[["PCTRET_1", "ACTRET_1"]].cumsum().plot(figsize=(16, 3), kind="area", stacked=False, color=colors("GyOr"), title="B&H vs. Cum. Active Returns", alpha=.4, grid=True).axhline(0, color="black")

Out [15]:

<matplotlib.lines.Line2D at 0x12da79790>

Disclaimer

All investments involve risk, and the past performance of a security, industry, sector, market, financial product, trading strategy, or individual’s trading does not guarantee future results or returns. Investors are fully responsible for any investment decisions they make. Such decisions should be based solely on an evaluation of their financial circumstances, investment objectives, risk tolerance, and liquidity needs.
Any opinions, news, research, analyses, prices, or other information offered is provided as general market commentary, and does not constitute investment advice. I will not accept liability for any loss or damage, including without limitation any loss of profit, which may arise directly or indirectly from use of or reliance on such information.

268 KiB Raw Blame History Unescape Escape