Duality Methods in Portfolio Allocation with Transaction Constraints ...
Transcript of Duality Methods in Portfolio Allocation with Transaction Constraints ...
![Page 1: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/1.jpg)
BackgroundSpecific ModelsNumerical Tests
Duality Methods in Portfolio Allocation withTransaction Constraints and Uncertainty
Christopher W. Miller
Department of MathematicsUniversity of California, Berkeley
January 9, 2014
C. Miller Duality Methods in Portfolio Allocation
![Page 2: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/2.jpg)
BackgroundSpecific ModelsNumerical Tests
Project Overview
We examine optimal portfolio allocation with transactionconstraints via duality methods
Various models of asset returns and transaction costs
Correlations and uncertainty in estimated parameters
General algorithm to solve dual portfolio allocation problem.
Rapidly approximate optimal allocations with a large number ofassets and uncertain parameters.
C. Miller Duality Methods in Portfolio Allocation
![Page 3: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/3.jpg)
BackgroundSpecific ModelsNumerical Tests
Project Overview
We examine optimal portfolio allocation with transactionconstraints via duality methods
Various models of asset returns and transaction costs
Correlations and uncertainty in estimated parameters
General algorithm to solve dual portfolio allocation problem.
Rapidly approximate optimal allocations with a large number ofassets and uncertain parameters.
C. Miller Duality Methods in Portfolio Allocation
![Page 4: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/4.jpg)
BackgroundSpecific ModelsNumerical Tests
Table of contents
1 BackgroundProblem DescriptionLagrangian Relaxation
2 Specific ModelsBinary ModelTernary ModelCorrelation Model
3 Numerical Tests
C. Miller Duality Methods in Portfolio Allocation
![Page 5: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/5.jpg)
BackgroundSpecific ModelsNumerical Tests
Problem DescriptionLagrangian Relaxation
Problem Description
Single-period investment model with n assets:
Risk-return preference: f = f1 + · · ·+ fn,
Transaction costs: g = g1 + · · ·+ gn,
Investment constraint: wi ∈ [w i ,w i ] =Wi .
Assume that we can rapidly optimize:
minwi∈Wi
fi (wi ) + λgi (wi ).
C. Miller Duality Methods in Portfolio Allocation
![Page 6: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/6.jpg)
BackgroundSpecific ModelsNumerical Tests
Problem DescriptionLagrangian Relaxation
Problem Description
Single-period investment model with n assets:
Risk-return preference: f = f1 + · · ·+ fn,
Transaction costs: g = g1 + · · ·+ gn,
Investment constraint: wi ∈ [w i ,w i ] =Wi .
Assume that we can rapidly optimize:
minwi∈Wi
fi (wi ) + λgi (wi ).
C. Miller Duality Methods in Portfolio Allocation
![Page 7: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/7.jpg)
BackgroundSpecific ModelsNumerical Tests
Problem DescriptionLagrangian Relaxation
Lagrangian Relaxation
In general, we consider an investment problem of the form:
p∗ = minw∈W
{f (w) : g(w) ≤ τ} .
In this project, we consider the related problem:
d∗ = maxλ≥0
minw∈W
L(w , λ)
where L(w , λ) = f (w) + λ (g(w)− τ) .
C. Miller Duality Methods in Portfolio Allocation
![Page 8: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/8.jpg)
BackgroundSpecific ModelsNumerical Tests
Problem DescriptionLagrangian Relaxation
Lagrangian Relaxation
In general, we consider an investment problem of the form:
p∗ = minw∈W
{f (w) : g(w) ≤ τ} .
In this project, we consider the related problem:
d∗ = maxλ≥0
minw∈W
L(w , λ)
where L(w , λ) = f (w) + λ (g(w)− τ) .
C. Miller Duality Methods in Portfolio Allocation
![Page 9: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/9.jpg)
BackgroundSpecific ModelsNumerical Tests
Problem DescriptionLagrangian Relaxation
Lagrangian Relaxation
Proposition
With the dual problem defined as above, p∗ ≥ d∗.
Proof.
Let w∗ ∈ W such that g(w∗) ≤ τ and f (w∗) = p∗. For all λ ≥ 0,
p∗ ≥ f (w∗) + λ (g(w∗)− τ)
≥ minw∈W
{f (w) + λ (g(w)− τ)} .
Thenp∗ ≥ max
λ≥0minw∈W
L(w , λ) = d∗.
C. Miller Duality Methods in Portfolio Allocation
![Page 10: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/10.jpg)
BackgroundSpecific ModelsNumerical Tests
Problem DescriptionLagrangian Relaxation
Lagrangian Relaxation
Proposition
With the dual problem defined as above, p∗ ≥ d∗.
Proof.
Let w∗ ∈ W such that g(w∗) ≤ τ and f (w∗) = p∗.
For all λ ≥ 0,
p∗ ≥ f (w∗) + λ (g(w∗)− τ)
≥ minw∈W
{f (w) + λ (g(w)− τ)} .
Thenp∗ ≥ max
λ≥0minw∈W
L(w , λ) = d∗.
C. Miller Duality Methods in Portfolio Allocation
![Page 11: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/11.jpg)
BackgroundSpecific ModelsNumerical Tests
Problem DescriptionLagrangian Relaxation
Lagrangian Relaxation
Proposition
With the dual problem defined as above, p∗ ≥ d∗.
Proof.
Let w∗ ∈ W such that g(w∗) ≤ τ and f (w∗) = p∗. For all λ ≥ 0,
p∗ ≥ f (w∗) + λ (g(w∗)− τ)
≥ minw∈W
{f (w) + λ (g(w)− τ)} .
Thenp∗ ≥ max
λ≥0minw∈W
L(w , λ) = d∗.
C. Miller Duality Methods in Portfolio Allocation
![Page 12: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/12.jpg)
BackgroundSpecific ModelsNumerical Tests
Problem DescriptionLagrangian Relaxation
Lagrangian Relaxation
Proposition
With the dual problem defined as above, p∗ ≥ d∗.
Proof.
Let w∗ ∈ W such that g(w∗) ≤ τ and f (w∗) = p∗. For all λ ≥ 0,
p∗ ≥ f (w∗) + λ (g(w∗)− τ)
≥ minw∈W
{f (w) + λ (g(w)− τ)} .
Thenp∗ ≥ max
λ≥0minw∈W
L(w , λ) = d∗.
C. Miller Duality Methods in Portfolio Allocation
![Page 13: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/13.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Binary Model
In the binary model, we have fixed transaction costs for purchasesand disallow short-sales:
gi (ξ) =
+∞ if ξ < 0
0 if ξ = 0bi if ξ > 0.
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the binary model in time O(n).
Corollary
There is no duality gap in the binary model with unit transactioncosts and integer-valued τ .
C. Miller Duality Methods in Portfolio Allocation
![Page 14: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/14.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Binary Model
In the binary model, we have fixed transaction costs for purchasesand disallow short-sales:
gi (ξ) =
+∞ if ξ < 0
0 if ξ = 0bi if ξ > 0.
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the binary model in time O(n).
Corollary
There is no duality gap in the binary model with unit transactioncosts and integer-valued τ .
C. Miller Duality Methods in Portfolio Allocation
![Page 15: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/15.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Binary Model
In the binary model, we have fixed transaction costs for purchasesand disallow short-sales:
gi (ξ) =
+∞ if ξ < 0
0 if ξ = 0bi if ξ > 0.
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the binary model in time O(n).
Corollary
There is no duality gap in the binary model with unit transactioncosts and integer-valued τ .
C. Miller Duality Methods in Portfolio Allocation
![Page 16: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/16.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Binary Model
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the binary model in time O(n).
Proof
d∗ = maxλ≥0
minw∈W
{f (w) + λ (g(w)− τ)}
= maxλ≥0
{−λτ +
n∑i=1
minwi∈Wi
fi (wi ) + λgi (wi )
}
= maxλ≥0
{−λτ +
n∑i=1
min
{fi (0), λbi + min
0<wi≤w i
fi (wi )
}}.
C. Miller Duality Methods in Portfolio Allocation
![Page 17: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/17.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Binary Model
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the binary model in time O(n).
Proof
d∗ = maxλ≥0
minw∈W
{f (w) + λ (g(w)− τ)}
= maxλ≥0
{−λτ +
n∑i=1
minwi∈Wi
fi (wi ) + λgi (wi )
}
= maxλ≥0
{−λτ +
n∑i=1
min
{fi (0), λbi + min
0<wi≤w i
fi (wi )
}}.
C. Miller Duality Methods in Portfolio Allocation
![Page 18: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/18.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Binary Model
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the binary model in time O(n).
Proof
d∗ = maxλ≥0
minw∈W
{f (w) + λ (g(w)− τ)}
= maxλ≥0
{−λτ +
n∑i=1
minwi∈Wi
fi (wi ) + λgi (wi )
}
= maxλ≥0
{−λτ +
n∑i=1
min
{fi (0), λbi + min
0<wi≤w i
fi (wi )
}}.
C. Miller Duality Methods in Portfolio Allocation
![Page 19: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/19.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Binary Model
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the binary model in time O(n).
Proof
d∗ = maxλ≥0
minw∈W
{f (w) + λ (g(w)− τ)}
= maxλ≥0
{−λτ +
n∑i=1
minwi∈Wi
fi (wi ) + λgi (wi )
}
= maxλ≥0
{−λτ +
n∑i=1
min
{fi (0), λbi + min
0<wi≤w i
fi (wi )
}}.
C. Miller Duality Methods in Portfolio Allocation
![Page 20: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/20.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Binary Model
d∗ = maxλ≥0
{−λτ +
n∑i=1
min{
f 0i , λbi + f +
i
}}.
Maximization in O(n) via Quickselect algorithm.
C. Miller Duality Methods in Portfolio Allocation
![Page 21: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/21.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Ternary Model
In the ternary model, we have fixed transaction costs for bothpurchases and sales:
gi (ξ) =
si if ξ < 00 if ξ = 0bi if ξ > 0.
Furthermore, let us assume that we have the restriction on τ that
0 ≤ τ ≤n∑
i=1
max (si , bi ) .
C. Miller Duality Methods in Portfolio Allocation
![Page 22: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/22.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Ternary Model
Proposition
The solution of the dual problem under the ternary model may bewritten
d∗ = 1>f 0 + minu±
{(h+)
>u+ + (h−)
>u− : u± ≥ 0,
u+ + u− ≤ 1,b>u+ + s>u− ≤ τ
}for appropriate vectors f 0, h+, and h−.
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the ternary model in polynomial time.
C. Miller Duality Methods in Portfolio Allocation
![Page 23: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/23.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Ternary Model
Proposition
The solution of the dual problem under the ternary model may bewritten
d∗ = 1>f 0 + minu±
{(h+)
>u+ + (h−)
>u− : u± ≥ 0,
u+ + u− ≤ 1,b>u+ + s>u− ≤ τ
}for appropriate vectors f 0, h+, and h−.
Theorem
We can construct an optimal solution (λ∗,w∗) to the dual problemin the ternary model in polynomial time.
C. Miller Duality Methods in Portfolio Allocation
![Page 24: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/24.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Ternary Model
Proposition
In the ternary model, it is not true that p∗ = d∗ in general.
Proposition
Let (λ∗,w∗) be an optimal solution to the dual problem obtainedfrom the algorithm above. Then the duality gap is bounded by
0 ≤ p∗ − d∗ ≤ λ∗ (τ − g(w∗)) .
C. Miller Duality Methods in Portfolio Allocation
![Page 25: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/25.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Ternary Model
Proposition
In the ternary model, it is not true that p∗ = d∗ in general.
Proposition
Let (λ∗,w∗) be an optimal solution to the dual problem obtainedfrom the algorithm above. Then the duality gap is bounded by
0 ≤ p∗ − d∗ ≤ λ∗ (τ − g(w∗)) .
C. Miller Duality Methods in Portfolio Allocation
![Page 26: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/26.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Correlation Model
In the correlation model, we consider an objective function:
f (w , r̂) =1
2w>Σw − r̂>w ,
where r̂ ∈ R ⊂ Rn contains predicted returns and Σ is an estimateof the covariance matrix.
The corresponding dual problem is:
d∗ = maxλ≥0
minw∈W
maxr̂∈R{f (w , r̂) + λ (g(w)− τ)}
C. Miller Duality Methods in Portfolio Allocation
![Page 27: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/27.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Correlation Model
In the correlation model, we consider an objective function:
f (w , r̂) =1
2w>Σw − r̂>w ,
where r̂ ∈ R ⊂ Rn contains predicted returns and Σ is an estimateof the covariance matrix.
The corresponding dual problem is:
d∗ = maxλ≥0
minw∈W
maxr̂∈R{f (w , r̂) + λ (g(w)− τ)}
C. Miller Duality Methods in Portfolio Allocation
![Page 28: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/28.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Branch-and-Bound Method
C. Miller Duality Methods in Portfolio Allocation
![Page 29: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/29.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Branch-and-Bound Method
C. Miller Duality Methods in Portfolio Allocation
![Page 30: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/30.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Branch-and-Bound Method
C. Miller Duality Methods in Portfolio Allocation
![Page 31: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/31.jpg)
BackgroundSpecific ModelsNumerical Tests
Binary ModelTernary ModelCorrelation Model
Branch-and-Bound Method
C. Miller Duality Methods in Portfolio Allocation
![Page 32: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/32.jpg)
BackgroundSpecific ModelsNumerical Tests
Numerical Tests
Tested under extra condition that∑n
i=1 wi ≤ 1.
C. Miller Duality Methods in Portfolio Allocation
![Page 33: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/33.jpg)
BackgroundSpecific ModelsNumerical Tests
Conclusions
Duality methods provide approximate allocations in difficultportfolio optimization problems:
Usefulness of decomposability
Opportunities for parallelization in branch-and-bound methods
Approximate solution as input to solver for the primal problem.
C. Miller Duality Methods in Portfolio Allocation
![Page 34: Duality Methods in Portfolio Allocation with Transaction Constraints ...](https://reader033.fdocuments.in/reader033/viewer/2022052707/58a03fc31a28abf84d8be05f/html5/thumbnails/34.jpg)
BackgroundSpecific ModelsNumerical Tests
C. Miller Duality Methods in Portfolio Allocation