fipie 0.0.6

A simple portfolio optimiser beyond the mean-variance optimisation

fipie

A simple portfolio optimiser beyond the mean-variance optimisation

Overview

fipie is a simple portfolio optimiser which allows you to compute asset weights with algorithms from Markowitz's mean-variance optimisation to more modern methodologies!

Highlights:

• Extendable: fipie allows you to change the algorithm in a plug-and-play manner. Currently, it supports the following methods
  • Mean-variance optimisation
  • Equal nominal weighting
  • Volatility parity
  • Maximum diversification
  • Minimum variance
  • Equal risk contribution (ERC, aka risk parity)
• Clustering: Clustering is supported for all weighting methods above. If applied, instruments can be grouped by clusters before applying the weighting method

Example

The simplest example is to compute equal nominal weights without clusters.

from fipie import Portfolio, EqualWeight
from fipie.data import load_example_data

# Create a portfolio instance
price = load_example_data()
# Use weekly returns to calculate weights -- but this does not matter for equal weighting.
ret = price.asfreq('w', method='pad').pct_change().dropna()
portfolio = Portfolio(ret)

# Compute the latest weight
portfolio.weight_latest(EqualWeight())

As expected, each instrument gets a weight of 1/7.

node_id
SPY    0.142857
IWM    0.142857
QQQ    0.142857
MDY    0.142857
TLT    0.142857
GLD    0.142857
USO    0.142857
Name: weight, dtype: float64

The portfolio above actually contains similar ETFs (e.g., SPY and IWM which are both US equities). In this scenario, it's more desirable to group similar instruments into clusters first and then compute weights.

from fipie import CorrMatrixDistance

# Compute the latest weight with the clustering algorithm
cluster_algo = CorrMatrixDistance(max_clusters=3)
portfolio.weight_latest(EqualWeight(), cluster_algo)

With clusters, we can see different weights -- SPY and IWM are still equally weighted, but much smaller weights. On the other hand, TLT has got a much larger weight.

node_id
SPY    0.055556
IWM    0.055556
QQQ    0.166667
MDY    0.055556
TLT    0.333333
GLD    0.166667
USO    0.166667
Name: weight, dtype: float64

This is because the instruments are grouped as follows.

portfolio.create_tree(cluster_algo).show()

TLT has its own group, while equity ETFs are grouped into a cluster. There is another cluster for commodities (GLD and USO), resulting in weights somewhat between equities and bonds.

Node(root)
    Node(cluster_0)
        Node(cluster_1)
            Node(SPY)
            Node(IWM)
            Node(MDY)
        Node(QQQ)
    Node(cluster_2)
        Node(GLD)
        Node(USO)
    Node(TLT)

More details can be found here: https://fipie.readthedocs.io/en/latest/

Installation

pip install fipie

Requirements

These are the tested versions but most likely it should work with older versions.

• python >= 3.6
• pandas >= 0.25
• scipy >= 1.0

Licence

This project is licensed under the terms of the MIT license.