Description
There are conceptually two types of linear constraint.
There are linear constraints on categorical variables — for example, constraints on sectors or industries or countries. If the constraint is on countries, then the linear constraint will specify a minimum and maximum value for each country (separately).
There are also linear constraints on numeric variables. The leading example of this is to specify a range for the beta of the portfolio given that we have a beta value for each of the assets. Duration in a bond portfolio is another example of a numeric linear constraint.
The constraints on variance partitions are — in the larger picture — not actually linear constraints at all. But they are linear given that the risk fractions for the individual assets have been computed.
Implementation
A representation of the variables to be constrained is given as the lin.constraints argument. The bounds for the constraints are given in lin.bounds. There are several related arguments that control the specifics of the constraints.
Each constraint can be bounding:
- weight
- fraction of portfolio variance
- monetary value
- value of portfolio variance
Each constraint can be (with some limitations):
- on the portfolio
- on the trade
Each constraint can be (with some limitations):
- on the net
- on the gross
Each constraint can be:
- on the long-side only
- on the short-side only
- on all assets
