strand: A framework for investment strategy simulation
Overview
strand
provides a framework for performing discrete (share-level)
simulations of investment strategies. Simulated portfolios optimize
exposure to an input signal subject to constraints such as position size
and factor exposure.
The package
vignette
provides an in-depth discussion of setup and usage. See
vignette("strand")
.
Features
- Straightforward data interfaces.
- Option to load daily data from binary (feather) files for fast access and low memory footprint.
- Share-level bookkeeping.
- YAML-based configuration.
- Position sizing based on portfolio weight and percentage of expected volume.
- Trade sizing based on percentage of expected volume.
- Ability to specify constraints on factor exposure, category exposure, and turnover.
- Automatic loosening of factor and category exposure constraints if no solution is found.
- Realistic trade filling based on percentage of actual volume.
Installation
# Install the latest version from CRAN:
install.packages("strand")
# Install development version from GitHub using remotes:
install.packages("remotes")
remotes::install_github("strand-tech/strand")
Note on solvers
The strand
package uses GLPK as
the default solver for portfolio optimization. As a result, it depends
on package Rglpk
.
It is possible to use SYMPHONY
instead, by setting solver: symphony
in the simulation’s configuration
file and installing the Rsymphony
package. Note that you will need to
install SYMPHONY on your system first, and on OS X perform a few extra
steps
to install Rsymphony
.
Usage
Four ingredients are required to run a simulation using strand
:
Configuration file. A file in yaml format that describes the parameters of the simulation, such as the input signal, risk constraints, trading limits, position limits, the location of data inputs, etc.
Security reference. A listing of all securities allowed in the simulation and any categorical values (such as sector and industry) that can be used in exposure constraints.
Signal, factor, and supplementary data. Data for each day including the input signal (to which exposure is maximized) and any factors that appear in constraints. Supplementary data could include, for example, a daily measure of market capitalization for use in universe construction.
Pricing data. Daily prices, dividends, and trading volume for computing market values and filling orders. Unadjusted prices and accompanying adjustment ratios may be used.
library(strand)
# Load up sample data
data(sample_secref)
data(sample_pricing)
data(sample_inputs)
# Load sample configuration file
config <- example_strategy_config()
# Create the Simulation object and run
sim <- Simulation$new(config,
raw_input_data = sample_inputs,
raw_pricing_data = sample_pricing,
security_reference_data = sample_secref)
sim$run()
# Print overall statistics
sim$overallStatsDf()
## Item Gross Net
## 1 Total P&L 419 -2,507
## 2 Total Return on GMV (%) 0.0 -0.1
## 3 Annualized Return on GMV (%) 0.5 -3.2
## 4 Annualized Vol (%) 0.5 0.7
## 5 Annualized Sharpe 1.12 -4.61
## 6 Max Drawdown (%) -0.1 -0.1
## 7 Avg GMV 1,999,350
## 8 Avg NMV 73
## 9 Avg Count 403
## 10 Avg Daily Turnover 220,439
## 11 Holding Period (months) 0.9
Example shiny application (local)
To run an example shiny application that allows interactively configuring and running a simulation:
library(strand)
example_shiny_app()
Example shiny application (docker)
If you have docker
and docker-compose
installed, you can run the
example shiny application by cloning the github
repository and running the
following commands from the top-level directory:
$ docker-compose build
$ docker-compose up
The application will run by default on port 80. To configure edit
docker-compose.yml
.