Tutorial: equal weighted equity strategy
This tutorial shows new users to the SigTech platform how to construct a systematic investment strategy where the strategy involves an equal weighted equities portfolio, based on a market capitalisation filtering of the S&P 500 universe.
Caution: this notebook will perform heavy computations that will require an instance running with a minimum configuration of 4 vCPU and 16 GiB. The configuration is performed once you are logged into the SigTech platform. To change configuration of running instances, stop the current research environment and change the configuration.
When creating an investment strategy on the SigTech platform, the workflow tends to follow these steps:
- 1.Set up the environment
- 2.Define the investment universe
- 3.Create the strategy
- 4.Construct the portfolio
- 5.Generate the performance report
Setting up your environment takes three steps:
- Import the relevant internal and external libraries
- Configure the environment parameters
- Initialise the environment
import sigtech.framework as sig
import pandas as pd
import datetime as dtm
import seaborn as sns
from uuid import uuid4
env = sig.init()
Define a range of global parameters used for creating the investment strategy. There are a number of equity indices available on the SigTech platform—in this example the
SPX INDEX is used. Use the Market Data section of the SigTech platform to find other available indices.START_DATE = pd.Timestamp(2021, 1, 4)
END_DATE = pd.Timestamp(2021, 5, 31)
CURRENCY = 'USD'
EQ_INDEX = 'SPX INDEX'
The equity index is filtered by using the
EquityUniverseFilter class. In this case the filter retains the top 5% of the universe, in terms of market capitalisation. The filtered universe is stored in the variable
universe as a Python dictionary.dates = pd.date_range(START_DATE, END_DATE, freq='d')
equity_filter = sig.EquityUniverseFilter(EQ_INDEX)
equity_filter.add('Market Cap', 'top', 5, 'quarterly', percent=True)
universe = equity_filter.apply_filter(dates).to_dict()
The following code block is run to visually show the effect of the universe filtering—the amount of securities, on the y-axis, that the universe contains through time.
pd.Series({d: len(v) for d, v in universe.items()}).plot(title='Universe Size')
The following code block will do the following:
- Unpack the values in the Python dictionary.
- Perform a union of the values.
- Remove all the duplicate values, by turning it into a set.
all_stocks is a set consisting of all securities that are in the filtered investment universe on at least one date.all_stocks = set().union(*universe.values())
The pandas DataFrame
ison indicates which securities are contained within our investment universe through time. The value 1 indicates membership on a given date, whereas the value 0 indicates absence from the investment universe.ison = pd.DataFrame(0, index=universe.keys(), columns=all_stocks)
for d, v in universe.items():
ison.loc[d, v] = 1
ison.columns = [f'{c} REINVSTRAT STRATEGY' for c in ison.columns]
A sub-section of the pandas DataFrame
isonis shown below:ison.iloc[:5, :5]
When single stocks are included in the investment strategy, there's a need to handle all the corporate actions that may arise. To solve this, the building block
ReinvestmentStrategy is used. This generic strategy takes 100% exposure to an underlying single stock, handles corporate actions as they arise, and adjusts the stock and cash positions accordingly. To further speed up runtime when creating the
ReinvestmentStrategy objects, the computations run in parallel. @sig.remote
def get_rs(ticker):
t = f'{ticker} REINVSTRAT'
try:
return sig.obj.get(f'{t} STRATEGY').name
except:
print(f'# Getting reinvestment strategy for {ticker}')
stock = sig.obj.get(ticker)
rs = sig.ReinvestmentStrategy(
# Base currency of the strategy
currency=stock.currency,
# Underlying stock of the strategy
underlyer=ticker,
# Start date of strategy
start_date=max(START_DATE, stock.history().first_valid_index()),
# End date of strategy
end_date=END_DATE,
# Name of strategy
ticker=f'{t}',
# Will trigger order to trade out before expiry, e.g. delisting
final_trade_out=True,
# Place adjustments to minimize exposure slippage
enable_tracking=True,
# No withholding tax used
withholding_tax_override=0.0,
)
_ = rs.history()
return rs.name
Performance time series of strategies that are being computed on the SigTech platform get cached. Using a
try and except block, such as in the above code block, could have a significant impact on runtime, since already calculated strategies can be loaded via the sig.obj.get syntax. Using the list comprehension below, a
ReinvestmentStrategy object is created for all stocks and the sig.calc method allows for running the function in parallel:reinvestment_strategies = sig.calc_remote([get_rs(x) for x in all_stocks])
The Equity Investment Strategy will allocate an equal weight of the portfolio to each of the stocks—and more specifically to those stocks'
ReinvestmentStrategy objects—that have been chosen based on the universe filter. The portfolio will be rebalanced at the end of every month. The
ison pandas DataFrame is transformed to form an equal weight signal:signal = ison.divide(ison.sum(axis=1), axis=0)
The pandas DataFrame gets turned into a
Strategy object with the SignalStrategy building block. There are numerous different parameters the user can utilise. For example, different allocation functions that map the signal to asset allocations in the strategy. ew_strategy = sig.SignalStrategy(
# Base currency of the strategy
currency=CURRENCY,
# Start date of the strategy
start_date=signal.first_valid_index(),
# End date of the strategy
end_date=END_DATE,
# The Signal DataFrame turned into a signal object
# with the help of the signal_library
signal_name=sig.signal_library.from_ts(signal).name,
# Rebalance frequency of the portfolio
rebalance_frequency='EOM',
# On the instruments which have a negative weight,
# refrain from creating a short version of that
# instrument
convert_long_short_weight=False,
# Name of the strategy
ticker=f'EW {str(uuid4())[:5]}'
)
Once the strategy is created, the following code block is used to create a performance report containing information on:
- Metrics
- Rolling performance charts
- Performance heat maps
There are a wide range of different performance views and reports available for the user which might be relevant depending on the specific use case.
sig.PerformanceReport(ew_strategy).report()
Last modified 1mo ago