caoruicn / openalpha

An open source equity statistical arbitrage backtest simulator, use the same API as WorldQuant's WebSim https://www.worldquantvrc.com/en/cms/wqc/websim

OpenAlpha

OpenAlpha is an equity statistical arbitrage backtest simulator, use the same API as WorldQuant's WebSim. It targets professional users, support both Python and C++.

Installation Prerequisites

This installation is based on Ubuntu 18.04. We need build boost to support boost numpy of python3. If you have installed boost-dev from apt, please uninstall first.

sudo -i
apt-get install -y \
    g++  \
    make \
    cmake \
    clang-format \
    clang \
    python \
    python3-dev \
    vim \
    exuberant-ctags \
    git \
    wget \
    liblog4cxx-dev \
    python3-tk \
    libhdf5-dev \
  && pip3 install six numpy pandas cython pytest matplotlib tables h5py \
  && wget https://dl.bintray.com/boostorg/release/1.65.1/source/boost_1_65_1.tar.gz \
  && tar xzf boost_1_65_1.tar.gz \
  && cd boost_1_65_1 \
  && ./bootstrap.sh --prefix=/usr/local --with-python-version=3.6 \
  && ./b2 -j `getconf _NPROCESSORS_ONLN` && ./b2 install \
  && cd .. \
  && /bin/rm -rf boost*
  && exit

Build OpenAlpha

cd ~
echo "export LD_LIBRARY_PATH=\$LD_LIBRARY_PATH:/usr/local/lib" >> .bashrc
source .bashrc
git clone https://github.com/opentradesolutions/openalpha
cd openalpha
make

Run sample alpha

It runs sample.py and sampe.cc configured in openalpha.conf.

#download data first
wget -O data.tgz https://www.dropbox.com/s/wdernq2kz3rgcoo/openalpha.tar.xz?dl=0; tar xJf data.tgz
./scripts/data.py -a par2h5 data/symbol.par data/date.par data/adv60.par data/close.par data/industrygroup.par data/industry.par data/industry.par data/sector.par data/subindustry.par

./build/release/openalpha/openalpha

Introduction to data

All data files are stored in hdf5 file format. Please have a look at data files. "data/symbol.h5" defines all instruments. "data/dates.h5" defines all dates. All the other files are 2D arrays. The row is indexed by date, we call it di in our code. The column is indexed by instrument, we call it ii in our code. The transposed version of data file is suffixed with '_t', e.g. transposed 'close.h5' file is named with 'close_t.h5'. There are some help functions in scripts/data.py for data handling.

Parquet format data file

We also support parquet file, please check out parquet branch.

https://github.com/opentradesolutions/openalpha/tree/parquet

In most of cases, HDF5 data format is faster than parquet. If there is no memory limit issue, Please use HDF5.

In parquet branch, array is column major in memory instead of row major in hdf5. So there is C++ api difference, please check out sample c++ file, HDF5 vs Parquet. Python API are the same.

Report

Openalpha has default report, and dump out daily pnl file. You can also use scripts/simsummary.py on the daily pnl file to generate more detailed report, plot, and do correlation calculation. Or you can use ffn.

Python version OpenAlpha

scripts/openalpha.py is a simplified pure-python version of openalpha. The performance can be optimized with cython. You can run it as below.

./scripts/openalpha.py -l 2 sample.py