Neural Market Trends

The Contrarian Case for Getting Active

Right before the New Year I posted an article on Medium titled “Is Passive Investing Going to Kill Us?” It centers around one particular chart where the amount of money flowing into passive investing is staggering.

While I’m a big fan of Passive Investing and I’ve been riding its success, I can’t help but wonder if I was just plain lucky. I know that markets love to correct imbalances but the question is when? Perhaps it might be prudent to think of active strategies again.

Enter Python

I’ve been using the heck out of Python lately and I discovered a Forex library (it’s V20 compatible) that lets me connect to Oanda easily. I did some more research and found a simple momentum trading bot in about 100 lines of python. Of course, this blog post was old and it didn’t work out of the box. On top of that it used a deprecated library to connect to Oanda. Luckily I was able to find a brave soul that did a lot of the updates to get it 90% of the way there.

I got the sucker up and running and wouldn’t you know it, it sucks. It loses a money when I trade it using a ‘paper trading’ account. When you think about it, this is the right way of doing it. Test it out using a paper trading account and see how it performs.

Disclosure: Some of the links below are affiliate links and at no additional cost to you, I’ll earn a commission. When you purchase a product or service using one of my affiliate links, the company compensates me, which helps me run this blog and keep my content free of charge to you.

Trading Bot Building Process

A customer of mine suggested reading “Advances in Financial Machine Learning” by Marcos Lopez de Prado. It’s quite a heavy read but Marcos shares A TON of good stuff in there. Most importantly in my opinion is how to build these trading bots or systems starting in Chapter 1.

It all starts with the Data Curators, the ones that pull in all the tick data and other stuff. They’re the ones that augment, compile, aggregate, and curate the data for downstream use. This is a crucial step, if this is not done well, your models will suffer.

Next it’s the Feature Analysts. They’re the ones that try to extract interesting features from the curated data set. This is also a crucial step because in my experience, good features can make you and bad features can break you.

The third step is where the Strategists come in, they are the ones that are responsible for taking the Data and Features and devising a strategy to use in the markets. When they’ve come up with a model, they pass this off to the next group, the Back Testers. The Back Testers do just that, they take the new strategy and test it in various market conditions to see if it actually produces alpha.

If the new strategy produces alpha, then the Back Testers move this model over to the Deployment Team. This team is responsible for putting it into production, whether it uses CPU or GPU power, how it’s called, etc. The Deployment Team is also responsible for taking this strategy OUT of production once the market has figured out this strategy and you’re no longer making money with it.

The last step is Portfolio Oversight. This is where the back tested and deployed strategy is carefully monitored. First from a ‘paper trading’ implementation, then to live trading. This is where it should start producing returns and where it gets taken out if the returns start faltering.

The book is full of more great information but as I wrote above, it’s not for the feint hearted. It’s invaluable IMHO if you want to build Trading Bots in a serious manner.

The Python Code

Without much further ado, here’s my cobbled together Momentum Trading Bot. Note, the code is not elegant as I’m still working through stuff and it’s very simplistic. Works with Python 3.6+ and you’ll need to add an account.txt file (the account number from Oanda) and a token.txt file (the V20 API token from Oanda).

#!/usr/bin/env python
# coding: utf-8

# In[1]:


import json
import oandapyV20 as opy
import oandapyV20.endpoints.instruments as instruments
from oandapyV20.contrib.factories import InstrumentsCandlesFactory

import pandas as pd
from pandas.io.json import json_normalize

from oandapyV20.exceptions import V20Error, StreamTerminated
from oandapyV20.endpoints.transactions import TransactionsStream
from oandapyV20.endpoints.pricing import PricingStream
from oandapyV20.contrib.requests import TrailingStopLossOrderRequest

import datetime
from dateutil import parser

import numpy as np


# In[2]:


def exampleAuth():
    accountID, token = None, None
    with open("./oanda_account/account.txt") as I:
        accountID = I.read().strip()
    with open("./oanda_account/token.txt") as I:
        token = I.read().strip()
    return accountID, token


# In[3]:


instrument = "AUD_USD"

#Set time functions to offset chart
today = datetime.datetime.today()
two_years_ago = today - datetime.timedelta(days=720)

current_time = datetime.datetime.now()
twentyfour_hours_ago = current_time - datetime.timedelta(days=2)

print (current_time)
print (twentyfour_hours_ago)


# In[4]:


#Create time parameter for Oanada call
ct = current_time.strftime("%Y-%m-%dT%H:%M:%SZ")
tf = twentyfour_hours_ago.strftime("%Y-%m-%dT%H:%M:%SZ")


# In[5]:


#Connect to tokens
accountID, access_token = exampleAuth()
client = opy.API(access_token=access_token)


# In[6]:


# The v20 api handles from times a little differently - be careful of the timezone
params={"from": tf,
        "to": ct,
        "granularity":'M1',
        "price":'A'}
r = instruments.InstrumentsCandles(instrument=instrument,params=params)
#Do not use client from above
data = client.request(r)
results= [{"time":x['time'],"closeAsk":float(x['ask']['c'])} for x in data['candles']]
df = pd.DataFrame(results).set_index('time')

df.index = pd.DatetimeIndex(df.index)

df.info()
df.head()


# In[7]:


df['returns'] = np.log(df['closeAsk'] / df['closeAsk'].shift(1))

cols = []

for momentum in [15, 30, 60, 120]:
    col = 'position_%s' % momentum
    df[col] = np.sign(df['returns'].rolling(momentum).mean())
    cols.append(col)


# In[8]:


get_ipython().run_line_magic('matplotlib', 'inline')
import seaborn as sns; sns.set()

strats = ['returns']

for col in cols:
    strat = 'strategy_%s' % col.split('_')[1]
    df[strat] = df[col].shift(1) * df['returns']
    strats.append(strat)


# In[9]:


df[strats].dropna().cumsum().apply(np.exp).plot()


# In[10]:


from oandapyV20.endpoints.pricing import PricingStream
import oandapyV20.endpoints.orders as orders
from oandapyV20.contrib.requests import MarketOrderRequest
from oandapyV20.exceptions import V20Error, StreamTerminated
import oandapyV20.endpoints.trades as trades
from oandapyV20.contrib.requests import TrailingStopLossOrderRequest

class MomentumTrader(PricingStream):
    def __init__(self, momentum, *args, **kwargs):
        PricingStream.__init__(self, *args, **kwargs)
        self.ticks = 0
        self.position = 0
        self.df = pd.DataFrame()
        self.momentum = momentum
        self.units = 1000
        self.connected = False
        #self.client = API(access_token=config['oanda']['access_token'])
        self.client = opy.API(access_token=access_token)

    def create_order(self, units):
        #order = orders.OrderCreate(accountID=config['oanda']['account_id'], data=MarketOrderRequest(instrument="EUR_USD", units=units).data)
        order = orders.OrderCreate(accountID=accountID, data=MarketOrderRequest(instrument=instrument, units=units).data)
        response = self.client.request(order)
        print('\t', response)

        #create a stop might need it's own fuction and called under the trading stragety part
        #stp = pd.Series(response, name='stop')
        #stp_id = stp.orderFillTransaction.get('tradeOpened').get('tradeID')
        #print ('\t The tradeID is ', stp_id)
        #stp_vwap = (stp.orderFillTransaction.get('fullVWAP') - 0.00010
        #print ('\t The stop loss value is ', stp_vwap)
        #stp_order = orders.OrderCreate(accountID=accountID, data=StopLossOrderRequest(tradeID=stp_id, price=stp_vwap))
        #response_stp_order = self.client.request(stp_order)

        #create a trailing stop
        #This below is for a trailing stop that works but the distance is key, this should be set for breakouts
        #tsl = pd.Series(response, name='junk')
        #tsl_id = tsl.orderFillTransaction.get('tradeOpened').get('tradeID')
        #print('The trade ID is: ', tsl_id)
        #tsl_order = orders.OrderCreate(accountID=accountID, data=TrailingStopLossOrderRequest(tradeID=tsl_id, distance=5).data)
        #response_tsl = self.client.request(tsl_order)
        #print('\t', response_tsl)


    def on_success(self, data):
        self.ticks += 1
        print("ticks=",self.ticks)
        # print(self.ticks, end=', ')

        # appends the new tick data to the DataFrame object
        self.df = self.df.append(pd.DataFrame([{'time': data['time'],'closeoutAsk':data['closeoutAsk']}],
                                 index=[data["time"]]))
        # transforms the time information to a DatetimeIndex object
        self.df.index = pd.DatetimeIndex(self.df["time"])

        # Convert items back to numeric (Why, OANDA, why are you returning strings?)
        self.df['closeoutAsk'] = pd.to_numeric(self.df["closeoutAsk"],errors='ignore')

        # resamples the data set to a new, homogeneous interval, set this from '5s' to '1m'
        dfr = self.df.resample('60s').last().bfill()

        #print ('\t this is the dfr', dfr)

        # calculates the log returns
        dfr['returns'] = np.log(dfr['closeoutAsk'] / dfr['closeoutAsk'].shift(1))

        #print('/t returns: ', dfr['returns'])

        # derives the positioning according to the momentum strategy
        dfr['position'] = np.sign(dfr['returns'].rolling(self.momentum).mean())

        print("position=",dfr['position'].iloc[-1])

        if dfr['position'].iloc[-1] == 1:
            print("go long")
            if self.position == 0:
                self.create_order(self.units)
            elif self.position == -1:
                self.create_order(self.units * 2)
            self.position = 1

        elif dfr['position'].iloc[-1] == -1:
            print("go short")
            if self.position == 0:
                self.create_order(-self.units)
            elif self.position == 1:
                self.create_order(-self.units * 2)
            self.position = -1

        if self.ticks == 25000:
            print("close out the position")
            if self.position == 1:
                self.create_order(-self.units)
            elif self.position == -1:
                self.create_order(self.units)
            self.disconnect()

    def disconnect(self):
        self.connected=False

    def rates(self, account_id, instruments, **params):
        self.connected = True
        params = params or {}
        ignore_heartbeat = None
        if "ignore_heartbeat" in params:
            ignore_heartbeat = params['ignore_heartbeat']
        while self.connected:
            response = self.client.request(self)
            for tick in response:
                if not self.connected:
                    break
                if not (ignore_heartbeat and tick["type"]=="HEARTBEAT"):
                    print(tick)
                    self.on_success(tick)


# In[11]:


# Set momentum to be the number of previous 60 second intervals to calculate against
# For two hours, momentum = 120. For 15 minutes, set to 15
mt = MomentumTrader(momentum=15,accountID=accountID,params={'instruments': instrument})
print (mt)


# In[12]:


mt.rates(account_id=accountID, instruments=instrument, ignore_heartbeat=True)

---

---