PyMarkowitz

Purpose

Provide utility to create visually appealing and customisable graphic representation of markowitz's modern portfolio theory.

Usage

For this you will need two files:

• A layout file (see below)

• A SQLite database or a list of CSV files, specify the loader with -l flag with historical data, specify the header you want to load by providing an argument to -c.

You can use this package as a CLI:

>> python -m markowitz --help

usage: PyMarkowitz [options] LAYOUT INPUT [INPUT...]

Display Assets and Portfolio Graphs from Layout Files

positional arguments:
  layout                layout file path
  input                 data input file(s)

optional arguments:
  -h, --help            show this help message and exit
  -l LOADER, --loader LOADER
                        loader (default: sqlite)
  -c COLUMN, --column COLUMN
                        name of the header (default: clot)
  --debug               activate debug mode
  --style STYLE         matplotlib graph style

Or as individual modules:

• markowitz/graphs for graph points generators
• markowitz/sets for point generators
• markowitz/financial for the assets

Layout Files

Syntax:

!	<- comment marker
&	<- start of a new window
(KEY=VALUE)	<- specify options
[]	<- row
|	<- column delimiter
& NAME (OPTIONS) {[|]}  <- valid window

Example:

! Compare the distribution of two assets
& Compare (precision=100, scale=100) {
	[ Kde(asset1) Kde(asset2) ]
	[ NormalGraph(asset1) | NormalGraph(asset2) ]
}

Extending the capabilities

You can extend the available graphs and sets by adding them in the markowitz/graphs and markowitz/sets directories.

• Sets are mathematical vector space point iterators. To implement one you need to subclass markowitz/sets/abstract.py which will force you to write a valid point iterator.

• Graphs are point iterators that map one set (hence markowitz/sets) to a set of points that can be graphed. To implement one you need to subclass markowitz/graphs/abstract.py which will force you to write a points() method which is called when generating the graph, make sure to zip the coordinates together, eg. zip(x, y). Graphs are loaded dynamically at runtime, use your implementation by mentioning the class name in the layout file, and name your implementation file the same way but lowercase. The abstract class holds the default config for a graph, you can override this config and the config passed when parsing layout files in your implementation.

• Loader holds the state of the assets and how they will be loaded, you can implement you own by following the already existing ones.

To be done

• Add better logging
• Add unittests
• Generate documentation
• Split github.io website in two parts: project and documentation
• Add package to pypy
• Add ability to specify other graph obj directory

Credits

• This project is licenced under MIT, see LICENSE.md for more information.
• Produced by Léo Duret as part of his annual project following the CMI EFiQuaS curriculum.