Portfolio Aligned Universes

In this notebook we’ll demonstrate how to define universes based on existing portfolio data.

For this purpose we will first upload some sample portfolio holdings and subsequently use them to define a universe filter.

import io

import polars as pl

from bayesline.api.equity import (
    PortfolioOrganizerSettings,
    UniverseSettings,
)
from bayesline.apiclient import BayeslineApiClient

bln = BayeslineApiClient.new_client(
    endpoint="https://[ENDPOINT]",
    api_key="[API-KEY]",
)

Uploading Portfolios

We’ll upload two portfolios PORT_1 and PORT_2 which contain assets Apple (IC83A1B819), Microsoft (ICF982536B) and Alphabet (ICA17F00B9).

portfolio_uploaders = bln.equity.uploaders.get_data_type("portfolios")

portfolio_uploader = portfolio_uploaders.create_or_replace_dataset("US-Portfolios")

data_csv = """
portfolio_id	date	asset_id	asset_id_type	value
PORT_1	2025-01-01	IC83A1B819	bayesid	.3
PORT_1	2025-01-01	ICA17F00B9	bayesid	.7
PORT_2	2025-01-01	IC83A1B819	bayesid	.4
PORT_2	2025-01-01	ICF982536B	bayesid	.6
PORT_1	2025-02-01	IC83A1B819	bayesid	1
PORT_2	2025-02-01	ICF982536B	bayesid	1
"""

portfolios_df = pl.read_csv(io.StringIO(data_csv.strip()), separator="\t", try_parse_dates=True)

portfolios_df.pivot(index=["date", "portfolio_id"], on="asset_id", values="value").fill_null(0)

shape: (4, 5)

date	portfolio_id	IC83A1B819	ICA17F00B9	ICF982536B
date	str	f64	f64	f64
2025-01-01	"PORT_1"	0.3	0.7	0.0
2025-01-01	"PORT_2"	0.4	0.0	0.6
2025-02-01	"PORT_1"	1.0	0.0	0.0
2025-02-01	"PORT_2"	0.0	0.0	1.0

portfolio_uploader.fast_commit(portfolios_df, mode="append")

UploadCommitResult(version=1, committed_names=[])

Creating the Universe

Note that below we could still use other filters in addition to the portfolio filter. This way we could for instance express the energy sector within the Russell 3000.

universe_settings = UniverseSettings(
    dataset="Bayesline-US-All-1y",
    portfolio_filter=PortfolioOrganizerSettings(
        # filters against the point in time superset 
        # of all portfolios contained in `US-Portfolios`
        enabled_portfolios="US-Portfolios"
    ),
)

universe_api = bln.equity.universes.load(universe_settings)

When obtaining the universe data note how even though we only specified holdings for 2025-01-01 and 2025-02-01 (which could be rebalance dates) the resulting universe contains entries for each day. Underneath the holdings are forward filled and delisted assets are dropped.

universe_api.get().to_pandas().assign(value=1.).set_index(["date", "bayesid"]).unstack()

	value
bayesid	IC83A1B819	ICA17F00B9	ICF982536B
date
2025-01-01	1.0	1.0	1.0
2025-01-02	1.0	1.0	1.0
2025-01-03	1.0	1.0	1.0
2025-01-04	1.0	1.0	1.0
2025-01-05	1.0	1.0	1.0
...	...	...	...
2025-08-21	1.0	NaN	1.0
2025-08-22	1.0	NaN	1.0
2025-08-23	1.0	NaN	1.0
2025-08-24	1.0	NaN	1.0
2025-08-25	1.0	NaN	1.0

237 rows × 3 columns

# asset counts over
universe_api.counts().to_pandas().set_index("date")["count"].sort_index().plot()

<Axes: xlabel='date'>