UniverseΒΆ

Universe is a set of instruments in portfolio for each period, and the set of instruments changes along the way.

Generally, there are two types of universe

  1. Estimation universe

  2. Model universe

Estimation universe is used for modeling the factor risks, and normally selects the liquid and representative instruments in the entire universe. For example, in U.S. stock market, an estimation universe can be selected by the top 40% free-float market capitalisation equities with actively trading liquidity.

In the meantime, the model universe are selected in a more relaxed manner. It can be all the stocks in the region sometimes, or a list of instruments traded in your portfolio. The risk model of model universe is usually transformed from that generated from estimation universe. In our library, a risk model of estimation universe can be always transformed with method transform and the returns of the model universe

estimation_risk_model = PCA(n_components=5)
estimation_risk_model.fit(X=estimation_returns)

model_risk_model = estimation_risk_model.transform(X=model_returns)

In our benchmark metrics, model universes of most instruments in the region are always used to examine the model forecasting accuracy.

The instrument type is an important selection criteria. For example, composite assets, like ETF and trusts, are highly correlated with a combination of subsets of universe. Secondary listing stocks have little representation in the region but more exposed to the foreign exchange movement. So, it is important to review the instrument list in the universe about their constitution and characteristics.

Finally, we have a separate library, factor-pricing-model-universe to provide functions and pipelines to generate universes. Please refer to its documentation for further details.