Equilibrium expected returns
PortfolioOptimisers.EquilibriumExpectedReturns — Type
struct EquilibriumExpectedReturns{__T_ce, __T_w, __T_l} <: AbstractExpectedReturnsEstimatorComputes the expected excess returns that a set of equilibrium weights implies, by reverse optimisation.
It holds a covariance estimator, the equilibrium weights and the risk aversion parameter. FactorBlackLittermanPrior and AugmentedBlackLittermanPrior build their prior mean with the same expression when their l is set.
The equilibrium risk premium vector $\mathbf{\Pi} = \delta \mathbf{\Sigma} \mathbf{W}$ is step 6 of the Appendix of [59] and of [60], with the market weights $\mathbf{W}$. [61] derives it earlier. It maximises $E_p - \lambda_S V_p$ under full investment, sets the gradient to zero, and finds the implied excess return $E_i - \pi = 2 \lambda_S (\mathbf{\Sigma} \boldsymbol{x})_i$ of every asset held, where $\pi$ is the riskless return. It also finds $\lambda_S = (E_p - \pi) / (2 V_p)$. Equations 5.1 and 5.2 of [12] derive the same result by setting the gradient of the quadratic utility $\boldsymbol{x}^\intercal \boldsymbol{\Pi} - \frac{\delta}{2} \boldsymbol{x}^\intercal \mathbf{\Sigma} \boldsymbol{x}$ to zero, so $\delta = 2 \lambda_S$.
l is the risk aversion of the representative investor. It is the $\lambda$ of the equation below, which [59] and [60] write $\delta$. Both say that $\delta$ is a proportionality constant based on the formulas of [62]. That work gives the universal hedging constant, on page 18, and no formula for $\delta$. This library takes l from the caller, and does not estimate it from the data. The result is linear in l. Twice the l gives twice every return, so a negative return moves down as l grows. An l of zero gives a zero mean.
w is the weight vector that the model assumes the market holds, such as the market capitalisation weights or the weights of a benchmark. The caller supplies it too, because the data hold returns and no holdings. If w is nothing, equilibrium_mu uses equal weights. It reads their number from the columns of the covariance matrix, not from X.
Fields
ce: Covariance estimator.
w: Optional equilibrium weights vectorassets × 1. Ifnothing,equilibrium_muuses equal weights.
l: Risk aversion parameter.
Constructors
EquilibriumExpectedReturns(; ce::StatsBase.CovarianceEstimator = PortfolioOptimisersCovariance(), w::Option{<:VecNum} = nothing, l::Number = 1) -> EquilibriumExpectedReturnsKeywords correspond to the struct's fields.
Validation
ceis not anAbstractVarianceEstimator. A variance estimator has nocovorcor, soassert_covariance_estimatorraises anArgumentErrorthat namesce.- If
wis notnothing,!isempty(w).
Propagated parameters
factory passes its arguments on to these @fprop fields:
ce: Recursively updated throughfactory.
View parameters
port_opt_view takes the view of these @vprop fields at the selected indices:
ce: Recursively viewed throughport_opt_view.w: Sliced to the selected indices throughport_opt_view.
Examples
julia> EquilibriumExpectedReturns()EquilibriumExpectedReturns ce ┼ PortfolioOptimisersCovariance │ ce ┼ Covariance │ │ me ┼ SimpleExpectedReturns │ │ │ w ┴ nothing │ │ ce ┼ GeneralCovariance │ │ │ ce ┼ StatsBase.SimpleCovariance: StatsBase.SimpleCovariance(true) │ │ │ w ┴ nothing │ │ alg ┼ FullMoment() │ │ w ┴ nothing │ mp ┼ MatrixProcessing │ │ pdm ┼ Posdef │ │ │ alg ┼ UnionAll: NearestCorrelationMatrix.Newton │ │ │ kwargs ┴ @NamedTuple{}: NamedTuple() │ │ dn ┼ nothing │ │ dt ┼ nothing │ │ alg ┼ nothing │ │ order ┴ NTuple{4, Symbol}: (:pdm, :dn, :dt, :alg) w ┼ nothing l ┴ Int64: 1Related
AbstractExpectedReturnsEstimatorStatsBase.CovarianceEstimatorStatsBase.AbstractWeightsequilibrium_mufactoryport_opt_view
References
- [61] W. F. Sharpe. Imputing expected security returns from portfolio composition. The Journal of Financial and Quantitative Analysis 9, 463–472 (1974). Sections II and III, pages 465 to 467.
- [59] F. Black and R. Litterman. Global asset allocation with equities, bonds, and currencies. Goldman, Sachs & Co., Fixed Income Research (1991). Appendix, step 6.
- [60] F. Black and R. Litterman. Global portfolio optimization. Financial Analysts Journal 48, 28–43 (1992). Appendix, step 6, page 42.
- [62] F. Black. Universal hedging: Optimizing currency risk and reward in international equity portfolios. Financial Analysts Journal 45, 16–22 (1989). The Universal Hedging Formula, page 18.
- [12] D. Cajas. Advanced Portfolio Optimization: A Cutting-edge Quantitative Approach (Springer Nature Switzerland, 2025). Section 5.1.1, Equations 5.1 and 5.2.
Statistics.mean — Method
Statistics.mean(me::EquilibriumExpectedReturns, X::MatNum; dims::Int = 1, kwargs...)Compute equilibrium expected returns from a covariance estimator, weights, and risk aversion.
The method returns λ * Σ * w. λ is the risk aversion parameter me.l, Σ is the covariance matrix that me.ce estimates from X, and w is the vector of equilibrium weights me.w. equilibrium_mu computes the expression, and uses equal weights if me.w is nothing.
The result is an excess return. Reverse optimisation implies a risk premium, so the result holds no risk-free rate, and the method subtracts none.
Mathematical definition
\[\begin{align} \hat{\boldsymbol{\mu}}_{\text{eq}} &= \lambda \, \hat{\mathbf{\Sigma}} \, \boldsymbol{w}\,. \end{align}\]
Where:
- $\lambda$: The risk aversion parameter
me.l. - $\hat{\mathbf{\Sigma}}$: The
N × Ncovariance matrix thatme.ceestimates from the data. - $\boldsymbol{w}$: The
N × 1equilibrium portfolio weightsme.w, or equal weights ifme.wisnothing.
Algorithm
- Estimate the covariance matrix of
Xwithme.ce, throughlibrary_covariance_estimator. The result issigma. - Pass
me.l,sigmaandme.wtoequilibrium_mu. It selects the equal-weight fallback whenme.wisnothing. Its result ismu.
Arguments
me: The equilibrium expected returns estimator.X: The data matrix,observations × assets.dims: Dimension along which the computation runs.kwargs...: More keyword arguments, which the method passes to the covariance estimator.
Validation
dims in (1, 2). This method does not checkdims. The covariance estimatorme.ceraises theDomainError.
Returns
mu::VecNum: The equilibrium expected returns, a vector of lengthN. The method returns a vector for both values ofdims, where most expected returns estimators return a(1, N)or an(N, 1)matrix.equilibrium_mumultiplies the covariance matrix by the weight vector, and the product is a vector.dimsreaches the covariance estimator alone, and the covariance matrix isN × Nfor both values ofdims.
Related
References
- [12]
- D. Cajas. Advanced Portfolio Optimization: A Cutting-edge Quantitative Approach (Springer Nature Switzerland, 2025).
- [59]
- F. Black and R. Litterman. Global asset allocation with equities, bonds, and currencies (Goldman, Sachs & Co., Fixed Income Research, Oct 1991).
- [60]
- F. Black and R. Litterman. Global portfolio optimization. Financial Analysts Journal 48, 28–43 (1992).
- [61]
- W. F. Sharpe. Imputing expected security returns from portfolio composition. The Journal of Financial and Quantitative Analysis 9, 463–472 (1974).
- [62]
- F. Black. Universal hedging: Optimizing currency risk and reward in international equity portfolios. Financial Analysts Journal 45, 16–22 (1989).