Skip to content

ProbStub/BeautifulPortfolioTools

Repository files navigation

Beautiful Portfolio Tools

What it is

The Beautiful Portfolio Tools are a refactoring of RAutoInvest, an abandoned attempt at a robo-advisor venture six years ago. Since then, I have picked up Python and like to put my investment tools on more solid grounds. Use at your own risk—no investment advice.

Importantly each tool performs one activity only (e.g., load data, compute future returns, compute weights, etc.), Combining each tool with UNIX pipes should allow for more complex portfolio construction tasks. If this is not immediately appealing, you may always run the "main"" tool and specify tasks with command line flags and options (or eventually interactive UI).

How to run

Presently the refactoring is focused on the data load and portfolio construction logic. You will need access to Kubernetes (either local minikube or at your preferred cloud provider) to run PySpark (using Python v3.5 or later) and PostgreSQL. Other than that the following should get you started:

  1. (Optional) create a new conda environment by running conda create -n bpt_conda
  2. Ensure all dependencies are installed pip install -r requirements.txt
  3. PySpark, Postgres & MongoDB needs to be installed locally for code to execute
  4. Insert your database, cloud and API secrets into the .env file
  5. Run each tool with a separate python call such as python source/main.py where "main" can be replaced with any of the tools in the repos root directory (this directory)

If you need more detailed instructions have a look at the manifest README which walks through some steps in more details (but is presently work in progress).

How to contribute

The code is public for a reason, and if you like to contribute, please consider to:

  • Provide test_* files for your contribution
  • Ensure that linting completes without issues

Help with the following would be greatly appreciated:

  • Pretty much everything :-)
  • If you enjoy cloud devops a PR for a better Dockerfile and integrating the container into the CircleCI config.yml would make a big difference
  • In the event that you love APIs feel free to donate a REST API to the mesh that can manage tools requests in order to abstract away the cloud deployment details currently preventing use for but the more technical users
  • Should you be a Spark hero I would welcome help with Spark native covariance implementation into PyPortfolioOpt because unfortunately the covariance matrix did not make it into Pandas API port of Spark 3.2.0

To set up your development environment after cloning the repo please perform the following preparation steps:

  1. Pre-Commit hooks have been established so please run pre-commit install
  2. PyTests are located in the /test directory but excluded from pre-commit hooks. Uncomment the testing hook in .pre-commit-config.yaml if you wish to enable pre-commit tests
  3. Very basic linting is configured in .pylintrc, adapt as needed
  4. Secrets can be maintained in .env and .gitignore is ignoring that file

NOTE: pre-commit hooks to trim whitespaces are defined and may fail at the first commit. Check the logs and try again.

Some background

I have created RAutoInvest initially to demonstrate the feasibility of automated ETF investments with relatively simple tooling. The objective has been to acquire funding for a robo-advisor venture but the project did not succeed. There had been discussions to improve the codebase and donate to an investor protection site. The aim has been to screen portfolio proposals for hidden fees. However, life intervened, and I could not go ahead with that project. The technology looks outdated years later, but I still find myself needing portfolio tooling and having a bit of time on my hands. Hence, I am picking up this refactoring project as a hobby, applying some tricks acquired over the years. That said, it is a hobby; issue response and commits may be intermittent.

Refactoring will initially focus on the following:

  • Change the data load to PySpark pipelines
  • Leverage ML facilities for data correction and imputation
  • Provide a Kubernetes run-time environment
  • Establish a testing and basic CI/CD setup
  • Switch to Python tooling for portfolio construction and performance evaluation

Later objectives will include:

  • Building a user interface beyond the command line interface
  • Provision broker connectivity facilities
  • Extend portfolio construction to emerging ML facilitated strategies

Why only one activity with each tool?

Over the years I have seen many complex convoluted code or data structures. As a user I was more than once unable to use a nice tool because it seemed to have more options than a cat has hairs. I want to empower regular people to build investment portfolios for themselves. If not to implement for investment purpose but to challenge and verify advice and offers from professional advisors. The asymmetric information in investment advisory more often than not leads to an unhappy outcome.

Long story short: One task at a time is far better to understand than multiple.

A secondary objective is to allow single tasks to be combined into more complex scenarios. Very basic UNIX commands such as sed and ls are frequently combined to do much more complex operations. Each command expects specific inputs and outputs and does one operation. When processing large amounts of investment data that is advisable as well. When predicting future returns on two separate markets you should be able to do that without waiting for the whole process of data load, forecasting, portfolio construction and output generation. All you want is the forecast.

About

Portfolio construction tooling with a KISS attitude to abstract away the machine learning and cloud hurdles.

Topics

Resources

Stars

Watchers

Forks

Packages

No packages published