Hi, I’d like to check a potential extension before implementing it, to ensure it fits the library scope and would be of interest.
I’m interested in extending the current mean-variance framework to support optimisation relative to an external stochastic process, motivated by insurance / ALM-style portfolio construction.
The idea is to allow portfolio optimisation in terms of surplus dynamics relative to a liability process.
Inputs:
Consistent with the existing design, the user provides:
Assets:
returns: dataframe of asset returns / prices
Liabilities (new input)
liability_returns: dataframe of returns / prices
same time index as asset returns
represents changes in present value of liabilities over time
liabilities are treated as a stochastic return process, like a noninvestable benchmark series.
Define surplus: S(w) = w^T μ - L
where:
μ: asset returns
L: liability return process
w: portfolio weights
A mean variance formulation on surplus: max w^T μ- λ(w^T Σw-2w^T c)
where:
λ: risk aversion
Σ: asset covariance matrix
c: Cov(μ,L), asset-liability covariance matrix
This is mathematically equivalent to MVO on the surplus process itself.
Does this proposal align with the library’s scope, and would a PR of this form be of interest? I wanted to ask first before doing a deep dive into the best approach. Thanks!
Hi, I’d like to check a potential extension before implementing it, to ensure it fits the library scope and would be of interest.
I’m interested in extending the current mean-variance framework to support optimisation relative to an external stochastic process, motivated by insurance / ALM-style portfolio construction.
The idea is to allow portfolio optimisation in terms of surplus dynamics relative to a liability process.
Inputs:
Consistent with the existing design, the user provides:
Assets:
returns: dataframe of asset returns / prices
Liabilities (new input)
liability_returns: dataframe of returns / prices
same time index as asset returns
represents changes in present value of liabilities over time
liabilities are treated as a stochastic return process, like a noninvestable benchmark series.
Define surplus: S(w) = w^T μ - L
where:
μ: asset returns
L: liability return process
w: portfolio weights
A mean variance formulation on surplus: max w^T μ- λ(w^T Σw-2w^T c)
where:
λ: risk aversion
Σ: asset covariance matrix
c: Cov(μ,L), asset-liability covariance matrix
This is mathematically equivalent to MVO on the surplus process itself.
Does this proposal align with the library’s scope, and would a PR of this form be of interest? I wanted to ask first before doing a deep dive into the best approach. Thanks!