SlideShare a Scribd company logo
Particle Swarm Optimization for Portfolio Selection.
Giacomo di Tollo
Dipartimento di Economia, University Ca’ Foscari, Venezia
(giacomo.ditollo@unive.it)
Applied Micro Seminar
Graduate School of Economics, Kyoto University
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 1 / 29
Summary
1 Metaheuristics and Exact Methods
2 Particle Swarm Optimisation
3 Our Initialisation
4 Portfolio Selection
5 Experimental Analysis
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 2 / 29
Metaheuristics
Strategies to guide the action of subordinated heuristics.
General principles.
Large scale optimization problems typically require larger
computational resources, and both practitioners and theoreticians
claim for robust methods, often endowed also with theoretical
properties [Nocedal, Wright ’00].
Combining theoretical properties of exact methods and fast
progress of heuristics [Fasano et al. ’14].
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 3 / 29
Metaheuristics
Strategies to guide the action of subordinated heuristics.
General principles.
Large scale optimization problems typically require larger
computational resources, and both practitioners and theoreticians
claim for robust methods, often endowed also with theoretical
properties [Nocedal, Wright ’00].
Combining theoretical properties of exact methods and fast
progress of heuristics [Fasano et al. ’14].
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 3 / 29
Metaheuristics
Strategies to guide the action of subordinated heuristics.
General principles.
Large scale optimization problems typically require larger
computational resources, and both practitioners and theoreticians
claim for robust methods, often endowed also with theoretical
properties [Nocedal, Wright ’00].
Combining theoretical properties of exact methods and fast
progress of heuristics [Fasano et al. ’14].
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 3 / 29
Metaheuristics
Strategies to guide the action of subordinated heuristics.
General principles.
Large scale optimization problems typically require larger
computational resources, and both practitioners and theoreticians
claim for robust methods, often endowed also with theoretical
properties [Nocedal, Wright ’00].
Combining theoretical properties of exact methods and fast
progress of heuristics [Fasano et al. ’14].
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 3 / 29
Classification of Metaheuristics
Trajectory methods (Simulated Annealing, Tabu Search,
Threshold Accepting).
Population methods (Genetic Algorithms, Ant Colony
Optimisation, Particle Swarm Optimisation).
Intensification VS Diversification [Blum and Roli ’03].
Exploitation VS Exploration [di Tollo et al. ’11].
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 4 / 29
Classification of Metaheuristics
Trajectory methods (Simulated Annealing, Tabu Search,
Threshold Accepting).
Population methods (Genetic Algorithms, Ant Colony
Optimisation, Particle Swarm Optimisation).
Intensification VS Diversification [Blum and Roli ’03].
Exploitation VS Exploration [di Tollo et al. ’11].
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 4 / 29
Classification of Metaheuristics
Trajectory methods (Simulated Annealing, Tabu Search,
Threshold Accepting).
Population methods (Genetic Algorithms, Ant Colony
Optimisation, Particle Swarm Optimisation).
Intensification VS Diversification [Blum and Roli ’03].
Exploitation VS Exploration [di Tollo et al. ’11].
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 4 / 29
Classification of Metaheuristics
Trajectory methods (Simulated Annealing, Tabu Search,
Threshold Accepting).
Population methods (Genetic Algorithms, Ant Colony
Optimisation, Particle Swarm Optimisation).
Intensification VS Diversification [Blum and Roli ’03].
Exploitation VS Exploration [di Tollo et al. ’11].
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 4 / 29
Basics on PSO
PSO [Kennedy-Eberhart ’95], for unconstrained global
optimization problem
min
x∈IRn
f(x) (1)
Population-based method.
f(x) is assumed to be nonlinear and non-convex.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 5 / 29
Basics on PSO
PSO [Kennedy-Eberhart ’95], for unconstrained global
optimization problem
min
x∈IRn
f(x) (1)
Population-based method.
f(x) is assumed to be nonlinear and non-convex.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 5 / 29
Basics on PSO
PSO [Kennedy-Eberhart ’95], for unconstrained global
optimization problem
min
x∈IRn
f(x) (1)
Population-based method.
f(x) is assumed to be nonlinear and non-convex.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 5 / 29
Basics on PSO
We have a set of particles. For each particle we define its velocity
and its position.
The random velocity flown the particle through the problem space.
Each particle is attracted to its previous best position.
Each particle is attracted to the global best position.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 6 / 29
Basics on PSO
We have a set of particles. For each particle we define its velocity
and its position.
The random velocity flown the particle through the problem space.
Each particle is attracted to its previous best position.
Each particle is attracted to the global best position.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 6 / 29
Basics on PSO
We have a set of particles. For each particle we define its velocity
and its position.
The random velocity flown the particle through the problem space.
Each particle is attracted to its previous best position.
Each particle is attracted to the global best position.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 6 / 29
Basics on PSO
We have a set of particles. For each particle we define its velocity
and its position.
The random velocity flown the particle through the problem space.
Each particle is attracted to its previous best position.
Each particle is attracted to the global best position.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 6 / 29
Basics on PSO
xk+1
j = xk
j + vk+1
j
vk+1
j = vk
j + α ⊗ (pk
j − xk
j ) + β ⊗ (pk
g − xk
j )
(2)
α = cj
k rk
j
β = cg
k rk
g
cj
k , cg
k ∈ (0, 2.5)
(3)
xk+1
j = xk
j + vk+1
j
vk+1
j = wj
k vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
(4)
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
Basics on PSO
xk+1
j = xk
j + vk+1
j
vk+1
j = vk
j + α ⊗ (pk
j − xk
j ) + β ⊗ (pk
g − xk
j )
(2)
α = cj
k rk
j
β = cg
k rk
g
cj
k , cg
k ∈ (0, 2.5)
(3)
xk+1
j = xk
j + vk+1
j
vk+1
j = wj
k vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
(4)
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
Basics on PSO
xk+1
j = xk
j + vk+1
j
vk+1
j = vk
j + α ⊗ (pk
j − xk
j ) + β ⊗ (pk
g − xk
j )
(2)
α = cj
k rk
j
β = cg
k rk
g
cj
k , cg
k ∈ (0, 2.5)
(3)
xk+1
j = xk
j + vk+1
j
vk+1
j = wj
k vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
(4)
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
Basics on PSO
xk+1
j = xk
j + vk+1
j
vk+1
j = vk
j + α ⊗ (pk
j − xk
j ) + β ⊗ (pk
g − xk
j )
(2)
α = cj
k rk
j
β = cg
k rk
g
cj
k , cg
k ∈ (0, 2.5)
(3)
xk+1
j = xk
j + vk+1
j
vk+1
j = wj
k vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
(4)
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
Basics on PSO
xk+1
j = xk
j + vk+1
j
vk+1
j = vk
j + α ⊗ (pk
j − xk
j ) + β ⊗ (pk
g − xk
j )
(2)
α = cj
k rk
j
β = cg
k rk
g
cj
k , cg
k ∈ (0, 2.5)
(3)
xk+1
j = xk
j + vk+1
j
vk+1
j = wj
k vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
(4)
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
Basics on PSO
xk+1
j = xk
j + vk+1
j
vk+1
j = vk
j + α ⊗ (pk
j − xk
j ) + β ⊗ (pk
g − xk
j )
(2)
α = cj
k rk
j
β = cg
k rk
g
cj
k , cg
k ∈ (0, 2.5)
(3)
xk+1
j = xk
j + vk+1
j
vk+1
j = wj
k vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
(4)
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
Basics on PSO
vk+1
j = χ wj
k
vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
xk+1
j = xk
j + vk+1
j
(5)
vk
j is the velocity (search direction) of the j-th particle at step k.
xk
j is the position of the j-th particle at step k.
f(pk
j ) = min
0≤ℓ≤k
{f(xℓ
j }, j = 1, . . . , P.
f(pk
g) = min
0≤ℓ≤k;j=1,...,P
{f(xℓ
j }.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
Basics on PSO
vk+1
j = χ wj
k
vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
xk+1
j = xk
j + vk+1
j
(5)
vk
j is the velocity (search direction) of the j-th particle at step k.
xk
j is the position of the j-th particle at step k.
f(pk
j ) = min
0≤ℓ≤k
{f(xℓ
j }, j = 1, . . . , P.
f(pk
g) = min
0≤ℓ≤k;j=1,...,P
{f(xℓ
j }.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
Basics on PSO
vk+1
j = χ wj
k
vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
xk+1
j = xk
j + vk+1
j
(5)
vk
j is the velocity (search direction) of the j-th particle at step k.
xk
j is the position of the j-th particle at step k.
f(pk
j ) = min
0≤ℓ≤k
{f(xℓ
j }, j = 1, . . . , P.
f(pk
g) = min
0≤ℓ≤k;j=1,...,P
{f(xℓ
j }.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
Basics on PSO
vk+1
j = χ wj
k
vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
xk+1
j = xk
j + vk+1
j
(5)
vk
j is the velocity (search direction) of the j-th particle at step k.
xk
j is the position of the j-th particle at step k.
f(pk
j ) = min
0≤ℓ≤k
{f(xℓ
j }, j = 1, . . . , P.
f(pk
g) = min
0≤ℓ≤k;j=1,...,P
{f(xℓ
j }.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
Basics on PSO
vk+1
j = χ wj
k
vk
j + ck
j rk
j ⊗ (pk
j − xk
j ) + ck
g rk
g ⊗ (pk
g − xk
j )
xk+1
j = xk
j + vk+1
j
(5)
vk
j is the velocity (search direction) of the j-th particle at step k.
xk
j is the position of the j-th particle at step k.
f(pk
j ) = min
0≤ℓ≤k
{f(xℓ
j }, j = 1, . . . , P.
f(pk
g) = min
0≤ℓ≤k;j=1,...,P
{f(xℓ
j }.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
Our Approach
Ortogonal particles’ initialization in PSO.
Deterministic PSO.
Experimental Analysis to prove the effectiveness of our proposal.
Portfolio Selection Problem.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 9 / 29
Our Approach
Ortogonal particles’ initialization in PSO.
Deterministic PSO.
Experimental Analysis to prove the effectiveness of our proposal.
Portfolio Selection Problem.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 9 / 29
Our Approach
Ortogonal particles’ initialization in PSO.
Deterministic PSO.
Experimental Analysis to prove the effectiveness of our proposal.
Portfolio Selection Problem.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 9 / 29
Our Approach
Ortogonal particles’ initialization in PSO.
Deterministic PSO.
Experimental Analysis to prove the effectiveness of our proposal.
Portfolio Selection Problem.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 9 / 29
Dynamic Systems
xk = A ·xk−1 + bk−1
x1 = A ·x0 + b0
x2 = A ·x1 + b1
x2 = A ·(A·x0 + b0) + b1
x2 = A2 ·x0 + A ·b0 + b1
x3 = A ·x2 + b2
x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2
x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2
xk = Ak ·x0 + k−1
t=0 Ak−t−1·bt
FREE REPONSE FORCED RESPONSE
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
Dynamic Systems
xk = A ·xk−1 + bk−1
x1 = A ·x0 + b0
x2 = A ·x1 + b1
x2 = A ·(A·x0 + b0) + b1
x2 = A2 ·x0 + A ·b0 + b1
x3 = A ·x2 + b2
x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2
x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2
xk = Ak ·x0 + k−1
t=0 Ak−t−1·bt
FREE REPONSE FORCED RESPONSE
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
Dynamic Systems
xk = A ·xk−1 + bk−1
x1 = A ·x0 + b0
x2 = A ·x1 + b1
x2 = A ·(A·x0 + b0) + b1
x2 = A2 ·x0 + A ·b0 + b1
x3 = A ·x2 + b2
x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2
x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2
xk = Ak ·x0 + k−1
t=0 Ak−t−1·bt
FREE REPONSE FORCED RESPONSE
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
Dynamic Systems
xk = A ·xk−1 + bk−1
x1 = A ·x0 + b0
x2 = A ·x1 + b1
x2 = A ·(A·x0 + b0) + b1
x2 = A2 ·x0 + A ·b0 + b1
x3 = A ·x2 + b2
x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2
x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2
xk = Ak ·x0 + k−1
t=0 Ak−t−1·bt
FREE REPONSE FORCED RESPONSE
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
Dynamic Systems
xk = A ·xk−1 + bk−1
x1 = A ·x0 + b0
x2 = A ·x1 + b1
x2 = A ·(A·x0 + b0) + b1
x2 = A2 ·x0 + A ·b0 + b1
x3 = A ·x2 + b2
x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2
x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2
xk = Ak ·x0 + k−1
t=0 Ak−t−1·bt
FREE REPONSE FORCED RESPONSE
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
Dynamic Systems
xk = A ·xk−1 + bk−1
x1 = A ·x0 + b0
x2 = A ·x1 + b1
x2 = A ·(A·x0 + b0) + b1
x2 = A2 ·x0 + A ·b0 + b1
x3 = A ·x2 + b2
x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2
x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2
xk = Ak ·x0 + k−1
t=0 Ak−t−1·bt
FREE REPONSE FORCED RESPONSE
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
Ortogonal Initialisation
Assumption
We assume in (5) that ck
j = c, rk
j = r for any j = 1, ..., P, ck
g = ¯c,
rk
g = ¯r and wk
j = w, for any k ≥ 0.
PSO iteration (5) is equivalent to the following discrete stationary
(time-invariant) system (i.e., X(k + 1) = AX(k) + b(k)):
Xj (k + 1) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In

 Xj (k) +


χ crpk
j + ¯c¯rpk
g
χ crpk
j + ¯c¯rpk
g


where
Xj (k) =


vk
j
xk
j

 ∈ IR2n
, k ≥ 0.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 11 / 29
Ortogonal Initialisation
Assumption
We assume in (5) that ck
j = c, rk
j = r for any j = 1, ..., P, ck
g = ¯c,
rk
g = ¯r and wk
j = w, for any k ≥ 0.
PSO iteration (5) is equivalent to the following discrete stationary
(time-invariant) system (i.e., X(k + 1) = AX(k) + b(k)):
Xj (k + 1) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In

 Xj (k) +


χ crpk
j + ¯c¯rpk
g
χ crpk
j + ¯c¯rpk
g


where
Xj (k) =


vk
j
xk
j

 ∈ IR2n
, k ≥ 0.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 11 / 29
Ortogonal Initialisation
Assumption
We assume in (5) that ck
j = c, rk
j = r for any j = 1, ..., P, ck
g = ¯c,
rk
g = ¯r and wk
j = w, for any k ≥ 0.
PSO iteration (5) is equivalent to the following discrete stationary
(time-invariant) system (i.e., X(k + 1) = AX(k) + b(k)):
Xj (k + 1) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In

 Xj (k) +


χ crpk
j + ¯c¯rpk
g
χ crpk
j + ¯c¯rpk
g


where
Xj (k) =


vk
j
xk
j

 ∈ IR2n
, k ≥ 0.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 11 / 29
PSO reformulation
Using a standard notation for linear systems, we can split Xj(k) into the
free response XjL(k) and the forced response XjF (k), so that
Xj (k) = XjL(k) + XjF (k)
where
XjL(k) = Φ(k)Xj (0), XjF (k) =
k−1
τ=0
H(k − τ)Uj(τ)
and
Φ(k) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In


k
∈ IR2n×2n
H(k − τ) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In


k−τ−1
∈ IR2n×2n
Uj (τ) =


χ crpτ
j + ¯c¯rpτ
g
χ crpτ
j + ¯c¯rpτ
g

 ∈ IR2n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 12 / 29
PSO reformulation
Using a standard notation for linear systems, we can split Xj(k) into the
free response XjL(k) and the forced response XjF (k), so that
Xj (k) = XjL(k) + XjF (k)
where
XjL(k) = Φ(k)Xj (0), XjF (k) =
k−1
τ=0
H(k − τ)Uj(τ)
and
Φ(k) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In


k
∈ IR2n×2n
H(k − τ) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In


k−τ−1
∈ IR2n×2n
Uj (τ) =


χ crpτ
j + ¯c¯rpτ
g
χ crpτ
j + ¯c¯rpτ
g

 ∈ IR2n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 12 / 29
PSO reformulation
Using a standard notation for linear systems, we can split Xj(k) into the
free response XjL(k) and the forced response XjF (k), so that
Xj (k) = XjL(k) + XjF (k)
where
XjL(k) = Φ(k)Xj (0), XjF (k) =
k−1
τ=0
H(k − τ)Uj(τ)
and
Φ(k) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In


k
∈ IR2n×2n
H(k − τ) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In


k−τ−1
∈ IR2n×2n
Uj (τ) =


χ crpτ
j + ¯c¯rpτ
g
χ crpτ
j + ¯c¯rpτ
g

 ∈ IR2n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 12 / 29
PSO reformulation
Using a standard notation for linear systems, we can split Xj(k) into the
free response XjL(k) and the forced response XjF (k), so that
Xj (k) = XjL(k) + XjF (k)
where
XjL(k) = Φ(k)Xj (0), XjF (k) =
k−1
τ=0
H(k − τ)Uj(τ)
and
Φ(k) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In


k
∈ IR2n×2n
H(k − τ) =


χwIn −χ(cr + ¯c¯r)In
χwIn [1 − χ(cr + ¯c¯r)] In


k−τ−1
∈ IR2n×2n
Uj (τ) =


χ crpτ
j + ¯c¯rpτ
g
χ crpτ
j + ¯c¯rpτ
g

 ∈ IR2n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 13 / 29
Near Orthogonality of particles’ trajectories
The free
response XjL(k) only depends on the initial point Xj(0), and
not on the vectors pτ
j , pτ
g, with τ ≥ 0.
The velocity vk
j of the j-th particle at iteration k may be regarded
as a search direction from the current position xk
j .
lim
k→∞
XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus).
We can enforce diversification / intensification by setting Φ(k) in
an appropriated way:
0 < χw < 1;
0 < χ(cr + ¯c¯r) < 2(χw + 1).
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
Near Orthogonality of particles’ trajectories
The free
response XjL(k) only depends on the initial point Xj(0), and
not on the vectors pτ
j , pτ
g, with τ ≥ 0.
The velocity vk
j of the j-th particle at iteration k may be regarded
as a search direction from the current position xk
j .
lim
k→∞
XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus).
We can enforce diversification / intensification by setting Φ(k) in
an appropriated way:
0 < χw < 1;
0 < χ(cr + ¯c¯r) < 2(χw + 1).
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
Near Orthogonality of particles’ trajectories
The free
response XjL(k) only depends on the initial point Xj(0), and
not on the vectors pτ
j , pτ
g, with τ ≥ 0.
The velocity vk
j of the j-th particle at iteration k may be regarded
as a search direction from the current position xk
j .
lim
k→∞
XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus).
We can enforce diversification / intensification by setting Φ(k) in
an appropriated way:
0 < χw < 1;
0 < χ(cr + ¯c¯r) < 2(χw + 1).
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
Near Orthogonality of particles’ trajectories
The free
response XjL(k) only depends on the initial point Xj(0), and
not on the vectors pτ
j , pτ
g, with τ ≥ 0.
The velocity vk
j of the j-th particle at iteration k may be regarded
as a search direction from the current position xk
j .
lim
k→∞
XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus).
We can enforce diversification / intensification by setting Φ(k) in
an appropriated way:
0 < χw < 1;
0 < χ(cr + ¯c¯r) < 2(χw + 1).
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
Near Orthogonality of particles’ trajectories
The free
response XjL(k) only depends on the initial point Xj(0), and
not on the vectors pτ
j , pτ
g, with τ ≥ 0.
The velocity vk
j of the j-th particle at iteration k may be regarded
as a search direction from the current position xk
j .
lim
k→∞
XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus).
We can enforce diversification / intensification by setting Φ(k) in
an appropriated way:
0 < χw < 1;
0 < χ(cr + ¯c¯r) < 2(χw + 1).
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
Near Orthogonality of particles’ trajectories
The free
response XjL(k) only depends on the initial point Xj(0), and
not on the vectors pτ
j , pτ
g, with τ ≥ 0.
The velocity vk
j of the j-th particle at iteration k may be regarded
as a search direction from the current position xk
j .
lim
k→∞
XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus).
We can enforce diversification / intensification by setting Φ(k) in
an appropriated way:
0 < χw < 1;
0 < χ(cr + ¯c¯r) < 2(χw + 1).
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
Near Orthogonality of particles’ trajectories
We set the initial position and velocity of the particles, so that the
subvectors {νk
j } (first n entries of the free responses) {XjL(k)},
are mutually orthogonal:
We assign 1 to one variable, and 0 to all other variables if P ≤ 2n.
If P > 2n, then set the initial position/velocity of the first 2n
particles as stated, while the other particles may have whatever
initial position/velocity.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 15 / 29
Near Orthogonality of particles’ trajectories
We set the initial position and velocity of the particles, so that the
subvectors {νk
j } (first n entries of the free responses) {XjL(k)},
are mutually orthogonal:
We assign 1 to one variable, and 0 to all other variables if P ≤ 2n.
If P > 2n, then set the initial position/velocity of the first 2n
particles as stated, while the other particles may have whatever
initial position/velocity.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 15 / 29
Basics of Portfolio Selection
Given a set of assets, the aim is to decide in which assets to invest
and by how much in order to optimise some specific criterion.
Minimise the risk given an expected return [Markowitz ’52].
Different risk measures can be used.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 16 / 29
Basics of Portfolio Selection
Given a set of assets, the aim is to decide in which assets to invest
and by how much in order to optimise some specific criterion.
Minimise the risk given an expected return [Markowitz ’52].
Different risk measures can be used.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 16 / 29
Basics of Portfolio Selection
Given a set of assets, the aim is to decide in which assets to invest
and by how much in order to optimise some specific criterion.
Minimise the risk given an expected return [Markowitz ’52].
Different risk measures can be used.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 16 / 29
Markowitz
Information about future prices is contained in historical series.
We are given a set (Universe) of assets {a1 . . . an}. Each asset ai
has associated a mean return ri and a return variance σ2
i .
For each pair of assets (ai, aj ) we know the return covariance σij .
A portfolio is a vector of real values P = x1 . . . xn.
rp = n
i=1 rixi .
σp = n
i=1
n
j=1 σijxi xj .
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
Markowitz
Information about future prices is contained in historical series.
We are given a set (Universe) of assets {a1 . . . an}. Each asset ai
has associated a mean return ri and a return variance σ2
i .
For each pair of assets (ai, aj ) we know the return covariance σij .
A portfolio is a vector of real values P = x1 . . . xn.
rp = n
i=1 rixi .
σp = n
i=1
n
j=1 σijxi xj .
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
Markowitz
Information about future prices is contained in historical series.
We are given a set (Universe) of assets {a1 . . . an}. Each asset ai
has associated a mean return ri and a return variance σ2
i .
For each pair of assets (ai, aj ) we know the return covariance σij .
A portfolio is a vector of real values P = x1 . . . xn.
rp = n
i=1 rixi .
σp = n
i=1
n
j=1 σijxi xj .
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
Markowitz
Information about future prices is contained in historical series.
We are given a set (Universe) of assets {a1 . . . an}. Each asset ai
has associated a mean return ri and a return variance σ2
i .
For each pair of assets (ai, aj ) we know the return covariance σij .
A portfolio is a vector of real values P = x1 . . . xn.
rp = n
i=1 rixi .
σp = n
i=1
n
j=1 σijxi xj .
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
Markowitz
Information about future prices is contained in historical series.
We are given a set (Universe) of assets {a1 . . . an}. Each asset ai
has associated a mean return ri and a return variance σ2
i .
For each pair of assets (ai, aj ) we know the return covariance σij .
A portfolio is a vector of real values P = x1 . . . xn.
rp = n
i=1 rixi .
σp = n
i=1
n
j=1 σijxi xj .
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
Markowitz
Information about future prices is contained in historical series.
We are given a set (Universe) of assets {a1 . . . an}. Each asset ai
has associated a mean return ri and a return variance σ2
i .
For each pair of assets (ai, aj ) we know the return covariance σij .
A portfolio is a vector of real values P = x1 . . . xn.
rp = n
i=1 rixi .
σp = n
i=1
n
j=1 σijxi xj .
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
Markowitz
we impose a minimum required return re
min n
i=1
n
j=1 σij xi xj
n
i=1 ri xi ≥ re
n
i=1 xi = 1
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 18 / 29
Markowitz
we impose a minimum required return re
min n
i=1
n
j=1 σij xi xj
n
i=1 ri xi ≥ re
n
i=1 xi = 1
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 18 / 29
Markowitz
we impose a minimum required return re
min n
i=1
n
j=1 σij xi xj
n
i=1 ri xi ≥ re
n
i=1 xi = 1
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 18 / 29
Markowitz
we impose a minimum required return re
min n
i=1
n
j=1 σij xi xj
n
i=1 ri xi ≥ re
n
i=1 xi = 1
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 18 / 29
Coherence [Artzner et al. ’99]
Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y);
Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments
increase risk);
Positive homogeneity: ρ(λx) = λρ(x) (liquidity);
Translation invariance: ρ(x + αr0) = ρ(x) − α.
Markowitz is not coherent
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
Coherence [Artzner et al. ’99]
Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y);
Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments
increase risk);
Positive homogeneity: ρ(λx) = λρ(x) (liquidity);
Translation invariance: ρ(x + αr0) = ρ(x) − α.
Markowitz is not coherent
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
Coherence [Artzner et al. ’99]
Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y);
Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments
increase risk);
Positive homogeneity: ρ(λx) = λρ(x) (liquidity);
Translation invariance: ρ(x + αr0) = ρ(x) − α.
Markowitz is not coherent
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
Coherence [Artzner et al. ’99]
Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y);
Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments
increase risk);
Positive homogeneity: ρ(λx) = λρ(x) (liquidity);
Translation invariance: ρ(x + αr0) = ρ(x) − α.
Markowitz is not coherent
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
Coherence [Artzner et al. ’99]
Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y);
Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments
increase risk);
Positive homogeneity: ρ(λx) = λρ(x) (liquidity);
Translation invariance: ρ(x + αr0) = ρ(x) − α.
Markowitz is not coherent
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
Coherence [Artzner et al. ’99]
Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y);
Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments
increase risk);
Positive homogeneity: ρ(λx) = λρ(x) (liquidity);
Translation invariance: ρ(x + αr0) = ρ(x) − α.
Markowitz is not coherent
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
Our Optimisation Problem [Chen and Wang ’08]
ρa,p(x) = a (x − E[x])+
1 + (1 − a) (x − E[x])−
p − E[x]
rP ≥ re
n
i=1 xi = 1
Kd ≤ n
i=1 zi ≤ Ku
zi d ≤ xi ≤ zi u, i = 1, . . . , n
zi (zi − 1) = 0, i = 1, . . . , n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
Our Optimisation Problem [Chen and Wang ’08]
ρa,p(x) = a (x − E[x])+
1 + (1 − a) (x − E[x])−
p − E[x]
rP ≥ re
n
i=1 xi = 1
Kd ≤ n
i=1 zi ≤ Ku
zi d ≤ xi ≤ zi u, i = 1, . . . , n
zi (zi − 1) = 0, i = 1, . . . , n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
Our Optimisation Problem [Chen and Wang ’08]
ρa,p(x) = a (x − E[x])+
1 + (1 − a) (x − E[x])−
p − E[x]
rP ≥ re
n
i=1 xi = 1
Kd ≤ n
i=1 zi ≤ Ku
zi d ≤ xi ≤ zi u, i = 1, . . . , n
zi (zi − 1) = 0, i = 1, . . . , n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
Our Optimisation Problem [Chen and Wang ’08]
ρa,p(x) = a (x − E[x])+
1 + (1 − a) (x − E[x])−
p − E[x]
rP ≥ re
n
i=1 xi = 1
Kd ≤ n
i=1 zi ≤ Ku
zi d ≤ xi ≤ zi u, i = 1, . . . , n
zi (zi − 1) = 0, i = 1, . . . , n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
Our Optimisation Problem [Chen and Wang ’08]
ρa,p(x) = a (x − E[x])+
1 + (1 − a) (x − E[x])−
p − E[x]
rP ≥ re
n
i=1 xi = 1
Kd ≤ n
i=1 zi ≤ Ku
zi d ≤ xi ≤ zi u, i = 1, . . . , n
zi (zi − 1) = 0, i = 1, . . . , n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
Our Optimisation Problem [Chen and Wang ’08]
ρa,p(x) = a (x − E[x])+
1 + (1 − a) (x − E[x])−
p − E[x]
rP ≥ re
n
i=1 xi = 1
Kd ≤ n
i=1 zi ≤ Ku
zi d ≤ xi ≤ zi u, i = 1, . . . , n
zi (zi − 1) = 0, i = 1, . . . , n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
Our Optimisation Problem [Chen and Wang ’08]
ρa,p(x) = a (x − E[x])+
1 + (1 − a) (x − E[x])−
p − E[x]
rP ≥ re
n
i=1 xi = 1
Kd ≤ n
i=1 zi ≤ Ku
zi d ≤ xi ≤ zi u, i = 1, . . . , n
zi (zi − 1) = 0, i = 1, . . . , n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
Our Optimisation Problem [Chen and Wang ’08]
ρa,p(x) = a (x − E[x])+
1 + (1 − a) (x − E[x])−
p − E[x]
rP ≥ re
n
i=1 xi = 1
Kd ≤ n
i=1 zi ≤ Ku
zi d ≤ xi ≤ zi u, i = 1, . . . , n
zi (zi − 1) = 0, i = 1, . . . , n
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
Exact Penalty Approach [di Tollo and Roli, 2008]
minx,z ρa,p(x) +
1
ε
7
ℓ=1
pℓ
p1 = max(0, re − rP)
p2 =
n
x=1
xi − 1
p3 + p4 =
n
x=1
max (0, zi · d − xi ) +
n
x=1
max (0, (xi − zi · u))
p5 + p6 = max 0, (Kd −
n
x=1
zi ) + max 0, (
n
x=1
zi − Ku)
p7 =
n
x=1
|zi · (1 − zi )|
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
Exact Penalty Approach [di Tollo and Roli, 2008]
minx,z ρa,p(x) +
1
ε
7
ℓ=1
pℓ
p1 = max(0, re − rP)
p2 =
n
x=1
xi − 1
p3 + p4 =
n
x=1
max (0, zi · d − xi ) +
n
x=1
max (0, (xi − zi · u))
p5 + p6 = max 0, (Kd −
n
x=1
zi ) + max 0, (
n
x=1
zi − Ku)
p7 =
n
x=1
|zi · (1 − zi )|
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
Exact Penalty Approach [di Tollo and Roli, 2008]
minx,z ρa,p(x) +
1
ε
7
ℓ=1
pℓ
p1 = max(0, re − rP)
p2 =
n
x=1
xi − 1
p3 + p4 =
n
x=1
max (0, zi · d − xi ) +
n
x=1
max (0, (xi − zi · u))
p5 + p6 = max 0, (Kd −
n
x=1
zi ) + max 0, (
n
x=1
zi − Ku)
p7 =
n
x=1
|zi · (1 − zi )|
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
Exact Penalty Approach [di Tollo and Roli, 2008]
minx,z ρa,p(x) +
1
ε
7
ℓ=1
pℓ
p1 = max(0, re − rP)
p2 =
n
x=1
xi − 1
p3 + p4 =
n
x=1
max (0, zi · d − xi ) +
n
x=1
max (0, (xi − zi · u))
p5 + p6 = max 0, (Kd −
n
x=1
zi ) + max 0, (
n
x=1
zi − Ku)
p7 =
n
x=1
|zi · (1 − zi )|
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
Exact Penalty Approach [di Tollo and Roli, 2008]
minx,z ρa,p(x) +
1
ε
7
ℓ=1
pℓ
p1 = max(0, re − rP)
p2 =
n
x=1
xi − 1
p3 + p4 =
n
x=1
max (0, zi · d − xi ) +
n
x=1
max (0, (xi − zi · u))
p5 + p6 = max 0, (Kd −
n
x=1
zi ) + max 0, (
n
x=1
zi − Ku)
p7 =
n
x=1
|zi · (1 − zi )|
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
Exact Penalty Approach [di Tollo and Roli, 2008]
minx,z ρa,p(x) +
1
ε
7
ℓ=1
pℓ
p1 = max(0, re − rP)
p2 =
n
x=1
xi − 1
p3 + p4 =
n
x=1
max (0, zi · d − xi ) +
n
x=1
max (0, (xi − zi · u))
p5 + p6 = max 0, (Kd −
n
x=1
zi ) + max 0, (
n
x=1
zi − Ku)
p7 =
n
x=1
|zi · (1 − zi )|
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
Benchmarks
FTSE MIB (32 assets, 1396 days)
DJIA (32 assets, 9312 days)
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 22 / 29
Benchmarks
FTSE MIB (32 assets, 1396 days)
DJIA (32 assets, 9312 days)
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 22 / 29
Best Objective over time
Figure: Experiments with ORTHOinit
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 23 / 29
Best Objective over time
Figure: Experiments without ORTHOinit
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 24 / 29
Results
Exact Solver
re ρ Constraints Violated Computational time (sec)
min(ri ) 0.0046 — 25693
max(ri ) 0.021 — 7064
PSO with ORTHOinit
re ρ Constraints Violated Computational time (sec)
min(ri ) 0.0048 Lower Bound 5375
max(ri ) 0.0238 Lower Bound, Capital, re 7021
PSO without ORTHOinit
re ρ Constraints Violated Computational time (sec)
min(ri ) 0.0042 Lower Bound 6530
max(ri ) 0.0238 Lower Bound, Capital, re 6135
Table: Experimental Results, Instance DJIA, basic PSO
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 25 / 29
Results
ρ Success Ratio
Strategy Min Max Std Mean
PSO-newinit-REVAC 0.00683077 0.00872689 0.00109750 46.6
PSO-standard 0.00691271 0.15525469 0.07204026 23.3
NEOS 0.00658258 0.00658258
Table: Experimental Results, Instance FTSI MIB, enhanced PSO.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 26 / 29
Conclusions
ORTHOinit fosters a better diversification;
Computational times;
The exact penalty approach should be revised, trying different
reformulations of our portfolio selection problem, where possibly
the simple constraints (i.e. linear constraints) are not moved to the
penalty function.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 27 / 29
References
[Artzner et al ’99]: P. Artzner, F. Delbaen, J.M. Eber, D. Heath (1999),
Coherent measures of risk, Mathematical Finance, vol. 9, pp. 203–228.
[Blum and Roli ’03]: C. Blum, A. Roli (2003), Metaheuristics in
combinatorial optimization: Overview and conceptual comparison, ACM
Computing Surveys, vol. 35 (3), pp. 268–308.
[Chen and Wang ’08]: Z. Chen, Y. Wang (2008), Two-sided coherent risk
measures and their application in realistic portfolio optimization, Journal
of Banking & Finance, vol. 32, pp. 2667–2673.
[Corazza et al. ’13]: M.Corazza, G.Fasano, R.Gusso (2013), Particle
Swarm Optimization with non-smooth penalty reformulation for a
complex portfolio selection problem, Applied Mathematics and
Computation, vol. 224, pp. 611–624.
[di Tollo et al. ’11]: G. di Tollo, F. Lardeux, J. Maturana, F. Saubion
(2011), From Adaptive to More Dynamic Control in Evolutionary
Algorithms, in EvoCOP 2011, LNCS 6622 Proceedings.
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 28 / 29
References
[di Tollo and Roli, ’08]: G. di Tollo, A. Roli (2008), Metaheuristics for the
portfolio selection problem, International Journal of Operations
Research, vol. 5 (1), pp. 13–35.
[Fasano et al ’14]: G. Fasano, G. Liuzzi, S. Lucidi, F. Rinaldi (2014), A
Linesearch-based Derivative-free Approach for Nonsmooth Constrained
Optimization , SIAM Journal on Optimization, vol. 24 (3), pp. 959-992.
[Markowitz ’52]: H. Markowitz (1952), Portfolio Selection, The Journal of
Finance, vol. 7 (1), pp. 77–91.
[Nocedal and Wright ’00]: J. Nocedal, S. Wright (2000), Numerical
Optimization (Springer Series in Operations Research and Financial
Engineering).
G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 29 / 29

More Related Content

Recently uploaded

Compexometric titration/Chelatorphy titration/chelating titration
Compexometric titration/Chelatorphy titration/chelating titrationCompexometric titration/Chelatorphy titration/chelating titration
Compexometric titration/Chelatorphy titration/chelating titration
Vandana Devesh Sharma
 
Sexuality - Issues, Attitude and Behaviour - Applied Social Psychology - Psyc...
Sexuality - Issues, Attitude and Behaviour - Applied Social Psychology - Psyc...Sexuality - Issues, Attitude and Behaviour - Applied Social Psychology - Psyc...
Sexuality - Issues, Attitude and Behaviour - Applied Social Psychology - Psyc...
PsychoTech Services
 
Eukaryotic Transcription Presentation.pptx
Eukaryotic Transcription Presentation.pptxEukaryotic Transcription Presentation.pptx
Eukaryotic Transcription Presentation.pptx
RitabrataSarkar3
 
aziz sancar nobel prize winner: from mardin to nobel
aziz sancar nobel prize winner: from mardin to nobelaziz sancar nobel prize winner: from mardin to nobel
aziz sancar nobel prize winner: from mardin to nobel
İsa Badur
 
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
Sérgio Sacani
 
Micronuclei test.M.sc.zoology.fisheries.
Micronuclei test.M.sc.zoology.fisheries.Micronuclei test.M.sc.zoology.fisheries.
Micronuclei test.M.sc.zoology.fisheries.
Aditi Bajpai
 
Gadgets for management of stored product pests_Dr.UPR.pdf
Gadgets for management of stored product pests_Dr.UPR.pdfGadgets for management of stored product pests_Dr.UPR.pdf
Gadgets for management of stored product pests_Dr.UPR.pdf
PirithiRaju
 
HOW DO ORGANISMS REPRODUCE?reproduction part 1
HOW DO ORGANISMS REPRODUCE?reproduction part 1HOW DO ORGANISMS REPRODUCE?reproduction part 1
HOW DO ORGANISMS REPRODUCE?reproduction part 1
Shashank Shekhar Pandey
 
Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...
Leonel Morgado
 
Sciences of Europe journal No 142 (2024)
Sciences of Europe journal No 142 (2024)Sciences of Europe journal No 142 (2024)
Sciences of Europe journal No 142 (2024)
Sciences of Europe
 
The cost of acquiring information by natural selection
The cost of acquiring information by natural selectionThe cost of acquiring information by natural selection
The cost of acquiring information by natural selection
Carl Bergstrom
 
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills MN
 
AJAY KUMAR NIET GreNo Guava Project File.pdf
AJAY KUMAR NIET GreNo Guava Project File.pdfAJAY KUMAR NIET GreNo Guava Project File.pdf
AJAY KUMAR NIET GreNo Guava Project File.pdf
AJAY KUMAR
 
Applied Science: Thermodynamics, Laws & Methodology.pdf
Applied Science: Thermodynamics, Laws & Methodology.pdfApplied Science: Thermodynamics, Laws & Methodology.pdf
Applied Science: Thermodynamics, Laws & Methodology.pdf
University of Hertfordshire
 
The debris of the ‘last major merger’ is dynamically young
The debris of the ‘last major merger’ is dynamically youngThe debris of the ‘last major merger’ is dynamically young
The debris of the ‘last major merger’ is dynamically young
Sérgio Sacani
 
The binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defectsThe binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defects
Sérgio Sacani
 
Farming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptxFarming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptx
Frédéric Baudron
 
Basics of crystallography, crystal systems, classes and different forms
Basics of crystallography, crystal systems, classes and different formsBasics of crystallography, crystal systems, classes and different forms
Basics of crystallography, crystal systems, classes and different forms
MaheshaNanjegowda
 
Direct Seeded Rice - Climate Smart Agriculture
Direct Seeded Rice - Climate Smart AgricultureDirect Seeded Rice - Climate Smart Agriculture
Direct Seeded Rice - Climate Smart Agriculture
International Food Policy Research Institute- South Asia Office
 
11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf
PirithiRaju
 

Recently uploaded (20)

Compexometric titration/Chelatorphy titration/chelating titration
Compexometric titration/Chelatorphy titration/chelating titrationCompexometric titration/Chelatorphy titration/chelating titration
Compexometric titration/Chelatorphy titration/chelating titration
 
Sexuality - Issues, Attitude and Behaviour - Applied Social Psychology - Psyc...
Sexuality - Issues, Attitude and Behaviour - Applied Social Psychology - Psyc...Sexuality - Issues, Attitude and Behaviour - Applied Social Psychology - Psyc...
Sexuality - Issues, Attitude and Behaviour - Applied Social Psychology - Psyc...
 
Eukaryotic Transcription Presentation.pptx
Eukaryotic Transcription Presentation.pptxEukaryotic Transcription Presentation.pptx
Eukaryotic Transcription Presentation.pptx
 
aziz sancar nobel prize winner: from mardin to nobel
aziz sancar nobel prize winner: from mardin to nobelaziz sancar nobel prize winner: from mardin to nobel
aziz sancar nobel prize winner: from mardin to nobel
 
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
EWOCS-I: The catalog of X-ray sources in Westerlund 1 from the Extended Weste...
 
Micronuclei test.M.sc.zoology.fisheries.
Micronuclei test.M.sc.zoology.fisheries.Micronuclei test.M.sc.zoology.fisheries.
Micronuclei test.M.sc.zoology.fisheries.
 
Gadgets for management of stored product pests_Dr.UPR.pdf
Gadgets for management of stored product pests_Dr.UPR.pdfGadgets for management of stored product pests_Dr.UPR.pdf
Gadgets for management of stored product pests_Dr.UPR.pdf
 
HOW DO ORGANISMS REPRODUCE?reproduction part 1
HOW DO ORGANISMS REPRODUCE?reproduction part 1HOW DO ORGANISMS REPRODUCE?reproduction part 1
HOW DO ORGANISMS REPRODUCE?reproduction part 1
 
Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...Authoring a personal GPT for your research and practice: How we created the Q...
Authoring a personal GPT for your research and practice: How we created the Q...
 
Sciences of Europe journal No 142 (2024)
Sciences of Europe journal No 142 (2024)Sciences of Europe journal No 142 (2024)
Sciences of Europe journal No 142 (2024)
 
The cost of acquiring information by natural selection
The cost of acquiring information by natural selectionThe cost of acquiring information by natural selection
The cost of acquiring information by natural selection
 
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
Travis Hills of MN is Making Clean Water Accessible to All Through High Flux ...
 
AJAY KUMAR NIET GreNo Guava Project File.pdf
AJAY KUMAR NIET GreNo Guava Project File.pdfAJAY KUMAR NIET GreNo Guava Project File.pdf
AJAY KUMAR NIET GreNo Guava Project File.pdf
 
Applied Science: Thermodynamics, Laws & Methodology.pdf
Applied Science: Thermodynamics, Laws & Methodology.pdfApplied Science: Thermodynamics, Laws & Methodology.pdf
Applied Science: Thermodynamics, Laws & Methodology.pdf
 
The debris of the ‘last major merger’ is dynamically young
The debris of the ‘last major merger’ is dynamically youngThe debris of the ‘last major merger’ is dynamically young
The debris of the ‘last major merger’ is dynamically young
 
The binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defectsThe binding of cosmological structures by massless topological defects
The binding of cosmological structures by massless topological defects
 
Farming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptxFarming systems analysis: what have we learnt?.pptx
Farming systems analysis: what have we learnt?.pptx
 
Basics of crystallography, crystal systems, classes and different forms
Basics of crystallography, crystal systems, classes and different formsBasics of crystallography, crystal systems, classes and different forms
Basics of crystallography, crystal systems, classes and different forms
 
Direct Seeded Rice - Climate Smart Agriculture
Direct Seeded Rice - Climate Smart AgricultureDirect Seeded Rice - Climate Smart Agriculture
Direct Seeded Rice - Climate Smart Agriculture
 
11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf11.1 Role of physical biological in deterioration of grains.pdf
11.1 Role of physical biological in deterioration of grains.pdf
 

Featured

2024 State of Marketing Report – by Hubspot
2024 State of Marketing Report – by Hubspot2024 State of Marketing Report – by Hubspot
2024 State of Marketing Report – by Hubspot
Marius Sescu
 
Everything You Need To Know About ChatGPT
Everything You Need To Know About ChatGPTEverything You Need To Know About ChatGPT
Everything You Need To Know About ChatGPT
Expeed Software
 
Product Design Trends in 2024 | Teenage Engineerings
Product Design Trends in 2024 | Teenage EngineeringsProduct Design Trends in 2024 | Teenage Engineerings
Product Design Trends in 2024 | Teenage Engineerings
Pixeldarts
 
How Race, Age and Gender Shape Attitudes Towards Mental Health
How Race, Age and Gender Shape Attitudes Towards Mental HealthHow Race, Age and Gender Shape Attitudes Towards Mental Health
How Race, Age and Gender Shape Attitudes Towards Mental Health
ThinkNow
 
AI Trends in Creative Operations 2024 by Artwork Flow.pdf
AI Trends in Creative Operations 2024 by Artwork Flow.pdfAI Trends in Creative Operations 2024 by Artwork Flow.pdf
AI Trends in Creative Operations 2024 by Artwork Flow.pdf
marketingartwork
 
Skeleton Culture Code
Skeleton Culture CodeSkeleton Culture Code
Skeleton Culture Code
Skeleton Technologies
 
PEPSICO Presentation to CAGNY Conference Feb 2024
PEPSICO Presentation to CAGNY Conference Feb 2024PEPSICO Presentation to CAGNY Conference Feb 2024
PEPSICO Presentation to CAGNY Conference Feb 2024
Neil Kimberley
 
Content Methodology: A Best Practices Report (Webinar)
Content Methodology: A Best Practices Report (Webinar)Content Methodology: A Best Practices Report (Webinar)
Content Methodology: A Best Practices Report (Webinar)
contently
 
How to Prepare For a Successful Job Search for 2024
How to Prepare For a Successful Job Search for 2024How to Prepare For a Successful Job Search for 2024
How to Prepare For a Successful Job Search for 2024
Albert Qian
 
Social Media Marketing Trends 2024 // The Global Indie Insights
Social Media Marketing Trends 2024 // The Global Indie InsightsSocial Media Marketing Trends 2024 // The Global Indie Insights
Social Media Marketing Trends 2024 // The Global Indie Insights
Kurio // The Social Media Age(ncy)
 
Trends In Paid Search: Navigating The Digital Landscape In 2024
Trends In Paid Search: Navigating The Digital Landscape In 2024Trends In Paid Search: Navigating The Digital Landscape In 2024
Trends In Paid Search: Navigating The Digital Landscape In 2024
Search Engine Journal
 
5 Public speaking tips from TED - Visualized summary
5 Public speaking tips from TED - Visualized summary5 Public speaking tips from TED - Visualized summary
5 Public speaking tips from TED - Visualized summary
SpeakerHub
 
ChatGPT and the Future of Work - Clark Boyd
ChatGPT and the Future of Work - Clark Boyd ChatGPT and the Future of Work - Clark Boyd
ChatGPT and the Future of Work - Clark Boyd
Clark Boyd
 
Getting into the tech field. what next
Getting into the tech field. what next Getting into the tech field. what next
Getting into the tech field. what next
Tessa Mero
 
Google's Just Not That Into You: Understanding Core Updates & Search Intent
Google's Just Not That Into You: Understanding Core Updates & Search IntentGoogle's Just Not That Into You: Understanding Core Updates & Search Intent
Google's Just Not That Into You: Understanding Core Updates & Search Intent
Lily Ray
 
How to have difficult conversations
How to have difficult conversations How to have difficult conversations
How to have difficult conversations
Rajiv Jayarajah, MAppComm, ACC
 
Introduction to Data Science
Introduction to Data ScienceIntroduction to Data Science
Introduction to Data Science
Christy Abraham Joy
 
Time Management & Productivity - Best Practices
Time Management & Productivity -  Best PracticesTime Management & Productivity -  Best Practices
Time Management & Productivity - Best Practices
Vit Horky
 
The six step guide to practical project management
The six step guide to practical project managementThe six step guide to practical project management
The six step guide to practical project management
MindGenius
 
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
RachelPearson36
 

Featured (20)

2024 State of Marketing Report – by Hubspot
2024 State of Marketing Report – by Hubspot2024 State of Marketing Report – by Hubspot
2024 State of Marketing Report – by Hubspot
 
Everything You Need To Know About ChatGPT
Everything You Need To Know About ChatGPTEverything You Need To Know About ChatGPT
Everything You Need To Know About ChatGPT
 
Product Design Trends in 2024 | Teenage Engineerings
Product Design Trends in 2024 | Teenage EngineeringsProduct Design Trends in 2024 | Teenage Engineerings
Product Design Trends in 2024 | Teenage Engineerings
 
How Race, Age and Gender Shape Attitudes Towards Mental Health
How Race, Age and Gender Shape Attitudes Towards Mental HealthHow Race, Age and Gender Shape Attitudes Towards Mental Health
How Race, Age and Gender Shape Attitudes Towards Mental Health
 
AI Trends in Creative Operations 2024 by Artwork Flow.pdf
AI Trends in Creative Operations 2024 by Artwork Flow.pdfAI Trends in Creative Operations 2024 by Artwork Flow.pdf
AI Trends in Creative Operations 2024 by Artwork Flow.pdf
 
Skeleton Culture Code
Skeleton Culture CodeSkeleton Culture Code
Skeleton Culture Code
 
PEPSICO Presentation to CAGNY Conference Feb 2024
PEPSICO Presentation to CAGNY Conference Feb 2024PEPSICO Presentation to CAGNY Conference Feb 2024
PEPSICO Presentation to CAGNY Conference Feb 2024
 
Content Methodology: A Best Practices Report (Webinar)
Content Methodology: A Best Practices Report (Webinar)Content Methodology: A Best Practices Report (Webinar)
Content Methodology: A Best Practices Report (Webinar)
 
How to Prepare For a Successful Job Search for 2024
How to Prepare For a Successful Job Search for 2024How to Prepare For a Successful Job Search for 2024
How to Prepare For a Successful Job Search for 2024
 
Social Media Marketing Trends 2024 // The Global Indie Insights
Social Media Marketing Trends 2024 // The Global Indie InsightsSocial Media Marketing Trends 2024 // The Global Indie Insights
Social Media Marketing Trends 2024 // The Global Indie Insights
 
Trends In Paid Search: Navigating The Digital Landscape In 2024
Trends In Paid Search: Navigating The Digital Landscape In 2024Trends In Paid Search: Navigating The Digital Landscape In 2024
Trends In Paid Search: Navigating The Digital Landscape In 2024
 
5 Public speaking tips from TED - Visualized summary
5 Public speaking tips from TED - Visualized summary5 Public speaking tips from TED - Visualized summary
5 Public speaking tips from TED - Visualized summary
 
ChatGPT and the Future of Work - Clark Boyd
ChatGPT and the Future of Work - Clark Boyd ChatGPT and the Future of Work - Clark Boyd
ChatGPT and the Future of Work - Clark Boyd
 
Getting into the tech field. what next
Getting into the tech field. what next Getting into the tech field. what next
Getting into the tech field. what next
 
Google's Just Not That Into You: Understanding Core Updates & Search Intent
Google's Just Not That Into You: Understanding Core Updates & Search IntentGoogle's Just Not That Into You: Understanding Core Updates & Search Intent
Google's Just Not That Into You: Understanding Core Updates & Search Intent
 
How to have difficult conversations
How to have difficult conversations How to have difficult conversations
How to have difficult conversations
 
Introduction to Data Science
Introduction to Data ScienceIntroduction to Data Science
Introduction to Data Science
 
Time Management & Productivity - Best Practices
Time Management & Productivity -  Best PracticesTime Management & Productivity -  Best Practices
Time Management & Productivity - Best Practices
 
The six step guide to practical project management
The six step guide to practical project managementThe six step guide to practical project management
The six step guide to practical project management
 
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
 

Portfolio Selection via Particle Swarm

  • 1. Particle Swarm Optimization for Portfolio Selection. Giacomo di Tollo Dipartimento di Economia, University Ca’ Foscari, Venezia (giacomo.ditollo@unive.it) Applied Micro Seminar Graduate School of Economics, Kyoto University G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 1 / 29
  • 2. Summary 1 Metaheuristics and Exact Methods 2 Particle Swarm Optimisation 3 Our Initialisation 4 Portfolio Selection 5 Experimental Analysis G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 2 / 29
  • 3. Metaheuristics Strategies to guide the action of subordinated heuristics. General principles. Large scale optimization problems typically require larger computational resources, and both practitioners and theoreticians claim for robust methods, often endowed also with theoretical properties [Nocedal, Wright ’00]. Combining theoretical properties of exact methods and fast progress of heuristics [Fasano et al. ’14]. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 3 / 29
  • 4. Metaheuristics Strategies to guide the action of subordinated heuristics. General principles. Large scale optimization problems typically require larger computational resources, and both practitioners and theoreticians claim for robust methods, often endowed also with theoretical properties [Nocedal, Wright ’00]. Combining theoretical properties of exact methods and fast progress of heuristics [Fasano et al. ’14]. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 3 / 29
  • 5. Metaheuristics Strategies to guide the action of subordinated heuristics. General principles. Large scale optimization problems typically require larger computational resources, and both practitioners and theoreticians claim for robust methods, often endowed also with theoretical properties [Nocedal, Wright ’00]. Combining theoretical properties of exact methods and fast progress of heuristics [Fasano et al. ’14]. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 3 / 29
  • 6. Metaheuristics Strategies to guide the action of subordinated heuristics. General principles. Large scale optimization problems typically require larger computational resources, and both practitioners and theoreticians claim for robust methods, often endowed also with theoretical properties [Nocedal, Wright ’00]. Combining theoretical properties of exact methods and fast progress of heuristics [Fasano et al. ’14]. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 3 / 29
  • 7. Classification of Metaheuristics Trajectory methods (Simulated Annealing, Tabu Search, Threshold Accepting). Population methods (Genetic Algorithms, Ant Colony Optimisation, Particle Swarm Optimisation). Intensification VS Diversification [Blum and Roli ’03]. Exploitation VS Exploration [di Tollo et al. ’11]. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 4 / 29
  • 8. Classification of Metaheuristics Trajectory methods (Simulated Annealing, Tabu Search, Threshold Accepting). Population methods (Genetic Algorithms, Ant Colony Optimisation, Particle Swarm Optimisation). Intensification VS Diversification [Blum and Roli ’03]. Exploitation VS Exploration [di Tollo et al. ’11]. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 4 / 29
  • 9. Classification of Metaheuristics Trajectory methods (Simulated Annealing, Tabu Search, Threshold Accepting). Population methods (Genetic Algorithms, Ant Colony Optimisation, Particle Swarm Optimisation). Intensification VS Diversification [Blum and Roli ’03]. Exploitation VS Exploration [di Tollo et al. ’11]. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 4 / 29
  • 10. Classification of Metaheuristics Trajectory methods (Simulated Annealing, Tabu Search, Threshold Accepting). Population methods (Genetic Algorithms, Ant Colony Optimisation, Particle Swarm Optimisation). Intensification VS Diversification [Blum and Roli ’03]. Exploitation VS Exploration [di Tollo et al. ’11]. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 4 / 29
  • 11. Basics on PSO PSO [Kennedy-Eberhart ’95], for unconstrained global optimization problem min x∈IRn f(x) (1) Population-based method. f(x) is assumed to be nonlinear and non-convex. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 5 / 29
  • 12. Basics on PSO PSO [Kennedy-Eberhart ’95], for unconstrained global optimization problem min x∈IRn f(x) (1) Population-based method. f(x) is assumed to be nonlinear and non-convex. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 5 / 29
  • 13. Basics on PSO PSO [Kennedy-Eberhart ’95], for unconstrained global optimization problem min x∈IRn f(x) (1) Population-based method. f(x) is assumed to be nonlinear and non-convex. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 5 / 29
  • 14. Basics on PSO We have a set of particles. For each particle we define its velocity and its position. The random velocity flown the particle through the problem space. Each particle is attracted to its previous best position. Each particle is attracted to the global best position. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 6 / 29
  • 15. Basics on PSO We have a set of particles. For each particle we define its velocity and its position. The random velocity flown the particle through the problem space. Each particle is attracted to its previous best position. Each particle is attracted to the global best position. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 6 / 29
  • 16. Basics on PSO We have a set of particles. For each particle we define its velocity and its position. The random velocity flown the particle through the problem space. Each particle is attracted to its previous best position. Each particle is attracted to the global best position. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 6 / 29
  • 17. Basics on PSO We have a set of particles. For each particle we define its velocity and its position. The random velocity flown the particle through the problem space. Each particle is attracted to its previous best position. Each particle is attracted to the global best position. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 6 / 29
  • 18. Basics on PSO xk+1 j = xk j + vk+1 j vk+1 j = vk j + α ⊗ (pk j − xk j ) + β ⊗ (pk g − xk j ) (2) α = cj k rk j β = cg k rk g cj k , cg k ∈ (0, 2.5) (3) xk+1 j = xk j + vk+1 j vk+1 j = wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) (4) G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
  • 19. Basics on PSO xk+1 j = xk j + vk+1 j vk+1 j = vk j + α ⊗ (pk j − xk j ) + β ⊗ (pk g − xk j ) (2) α = cj k rk j β = cg k rk g cj k , cg k ∈ (0, 2.5) (3) xk+1 j = xk j + vk+1 j vk+1 j = wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) (4) G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
  • 20. Basics on PSO xk+1 j = xk j + vk+1 j vk+1 j = vk j + α ⊗ (pk j − xk j ) + β ⊗ (pk g − xk j ) (2) α = cj k rk j β = cg k rk g cj k , cg k ∈ (0, 2.5) (3) xk+1 j = xk j + vk+1 j vk+1 j = wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) (4) G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
  • 21. Basics on PSO xk+1 j = xk j + vk+1 j vk+1 j = vk j + α ⊗ (pk j − xk j ) + β ⊗ (pk g − xk j ) (2) α = cj k rk j β = cg k rk g cj k , cg k ∈ (0, 2.5) (3) xk+1 j = xk j + vk+1 j vk+1 j = wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) (4) G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
  • 22. Basics on PSO xk+1 j = xk j + vk+1 j vk+1 j = vk j + α ⊗ (pk j − xk j ) + β ⊗ (pk g − xk j ) (2) α = cj k rk j β = cg k rk g cj k , cg k ∈ (0, 2.5) (3) xk+1 j = xk j + vk+1 j vk+1 j = wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) (4) G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
  • 23. Basics on PSO xk+1 j = xk j + vk+1 j vk+1 j = vk j + α ⊗ (pk j − xk j ) + β ⊗ (pk g − xk j ) (2) α = cj k rk j β = cg k rk g cj k , cg k ∈ (0, 2.5) (3) xk+1 j = xk j + vk+1 j vk+1 j = wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) (4) G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 7 / 29
  • 24. Basics on PSO vk+1 j = χ wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) xk+1 j = xk j + vk+1 j (5) vk j is the velocity (search direction) of the j-th particle at step k. xk j is the position of the j-th particle at step k. f(pk j ) = min 0≤ℓ≤k {f(xℓ j }, j = 1, . . . , P. f(pk g) = min 0≤ℓ≤k;j=1,...,P {f(xℓ j }. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
  • 25. Basics on PSO vk+1 j = χ wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) xk+1 j = xk j + vk+1 j (5) vk j is the velocity (search direction) of the j-th particle at step k. xk j is the position of the j-th particle at step k. f(pk j ) = min 0≤ℓ≤k {f(xℓ j }, j = 1, . . . , P. f(pk g) = min 0≤ℓ≤k;j=1,...,P {f(xℓ j }. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
  • 26. Basics on PSO vk+1 j = χ wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) xk+1 j = xk j + vk+1 j (5) vk j is the velocity (search direction) of the j-th particle at step k. xk j is the position of the j-th particle at step k. f(pk j ) = min 0≤ℓ≤k {f(xℓ j }, j = 1, . . . , P. f(pk g) = min 0≤ℓ≤k;j=1,...,P {f(xℓ j }. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
  • 27. Basics on PSO vk+1 j = χ wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) xk+1 j = xk j + vk+1 j (5) vk j is the velocity (search direction) of the j-th particle at step k. xk j is the position of the j-th particle at step k. f(pk j ) = min 0≤ℓ≤k {f(xℓ j }, j = 1, . . . , P. f(pk g) = min 0≤ℓ≤k;j=1,...,P {f(xℓ j }. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
  • 28. Basics on PSO vk+1 j = χ wj k vk j + ck j rk j ⊗ (pk j − xk j ) + ck g rk g ⊗ (pk g − xk j ) xk+1 j = xk j + vk+1 j (5) vk j is the velocity (search direction) of the j-th particle at step k. xk j is the position of the j-th particle at step k. f(pk j ) = min 0≤ℓ≤k {f(xℓ j }, j = 1, . . . , P. f(pk g) = min 0≤ℓ≤k;j=1,...,P {f(xℓ j }. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 8 / 29
  • 29. Our Approach Ortogonal particles’ initialization in PSO. Deterministic PSO. Experimental Analysis to prove the effectiveness of our proposal. Portfolio Selection Problem. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 9 / 29
  • 30. Our Approach Ortogonal particles’ initialization in PSO. Deterministic PSO. Experimental Analysis to prove the effectiveness of our proposal. Portfolio Selection Problem. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 9 / 29
  • 31. Our Approach Ortogonal particles’ initialization in PSO. Deterministic PSO. Experimental Analysis to prove the effectiveness of our proposal. Portfolio Selection Problem. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 9 / 29
  • 32. Our Approach Ortogonal particles’ initialization in PSO. Deterministic PSO. Experimental Analysis to prove the effectiveness of our proposal. Portfolio Selection Problem. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 9 / 29
  • 33. Dynamic Systems xk = A ·xk−1 + bk−1 x1 = A ·x0 + b0 x2 = A ·x1 + b1 x2 = A ·(A·x0 + b0) + b1 x2 = A2 ·x0 + A ·b0 + b1 x3 = A ·x2 + b2 x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2 x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2 xk = Ak ·x0 + k−1 t=0 Ak−t−1·bt FREE REPONSE FORCED RESPONSE G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
  • 34. Dynamic Systems xk = A ·xk−1 + bk−1 x1 = A ·x0 + b0 x2 = A ·x1 + b1 x2 = A ·(A·x0 + b0) + b1 x2 = A2 ·x0 + A ·b0 + b1 x3 = A ·x2 + b2 x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2 x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2 xk = Ak ·x0 + k−1 t=0 Ak−t−1·bt FREE REPONSE FORCED RESPONSE G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
  • 35. Dynamic Systems xk = A ·xk−1 + bk−1 x1 = A ·x0 + b0 x2 = A ·x1 + b1 x2 = A ·(A·x0 + b0) + b1 x2 = A2 ·x0 + A ·b0 + b1 x3 = A ·x2 + b2 x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2 x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2 xk = Ak ·x0 + k−1 t=0 Ak−t−1·bt FREE REPONSE FORCED RESPONSE G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
  • 36. Dynamic Systems xk = A ·xk−1 + bk−1 x1 = A ·x0 + b0 x2 = A ·x1 + b1 x2 = A ·(A·x0 + b0) + b1 x2 = A2 ·x0 + A ·b0 + b1 x3 = A ·x2 + b2 x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2 x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2 xk = Ak ·x0 + k−1 t=0 Ak−t−1·bt FREE REPONSE FORCED RESPONSE G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
  • 37. Dynamic Systems xk = A ·xk−1 + bk−1 x1 = A ·x0 + b0 x2 = A ·x1 + b1 x2 = A ·(A·x0 + b0) + b1 x2 = A2 ·x0 + A ·b0 + b1 x3 = A ·x2 + b2 x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2 x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2 xk = Ak ·x0 + k−1 t=0 Ak−t−1·bt FREE REPONSE FORCED RESPONSE G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
  • 38. Dynamic Systems xk = A ·xk−1 + bk−1 x1 = A ·x0 + b0 x2 = A ·x1 + b1 x2 = A ·(A·x0 + b0) + b1 x2 = A2 ·x0 + A ·b0 + b1 x3 = A ·x2 + b2 x3 = A ·(A2 ·x0 + A ·b0 + b1) + b2 x3 = A3 ·x0 + A2 ·b0 + A ·b1 + b2 xk = Ak ·x0 + k−1 t=0 Ak−t−1·bt FREE REPONSE FORCED RESPONSE G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 10 / 29
  • 39. Ortogonal Initialisation Assumption We assume in (5) that ck j = c, rk j = r for any j = 1, ..., P, ck g = ¯c, rk g = ¯r and wk j = w, for any k ≥ 0. PSO iteration (5) is equivalent to the following discrete stationary (time-invariant) system (i.e., X(k + 1) = AX(k) + b(k)): Xj (k + 1) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   Xj (k) +   χ crpk j + ¯c¯rpk g χ crpk j + ¯c¯rpk g   where Xj (k) =   vk j xk j   ∈ IR2n , k ≥ 0. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 11 / 29
  • 40. Ortogonal Initialisation Assumption We assume in (5) that ck j = c, rk j = r for any j = 1, ..., P, ck g = ¯c, rk g = ¯r and wk j = w, for any k ≥ 0. PSO iteration (5) is equivalent to the following discrete stationary (time-invariant) system (i.e., X(k + 1) = AX(k) + b(k)): Xj (k + 1) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   Xj (k) +   χ crpk j + ¯c¯rpk g χ crpk j + ¯c¯rpk g   where Xj (k) =   vk j xk j   ∈ IR2n , k ≥ 0. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 11 / 29
  • 41. Ortogonal Initialisation Assumption We assume in (5) that ck j = c, rk j = r for any j = 1, ..., P, ck g = ¯c, rk g = ¯r and wk j = w, for any k ≥ 0. PSO iteration (5) is equivalent to the following discrete stationary (time-invariant) system (i.e., X(k + 1) = AX(k) + b(k)): Xj (k + 1) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   Xj (k) +   χ crpk j + ¯c¯rpk g χ crpk j + ¯c¯rpk g   where Xj (k) =   vk j xk j   ∈ IR2n , k ≥ 0. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 11 / 29
  • 42. PSO reformulation Using a standard notation for linear systems, we can split Xj(k) into the free response XjL(k) and the forced response XjF (k), so that Xj (k) = XjL(k) + XjF (k) where XjL(k) = Φ(k)Xj (0), XjF (k) = k−1 τ=0 H(k − τ)Uj(τ) and Φ(k) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   k ∈ IR2n×2n H(k − τ) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   k−τ−1 ∈ IR2n×2n Uj (τ) =   χ crpτ j + ¯c¯rpτ g χ crpτ j + ¯c¯rpτ g   ∈ IR2n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 12 / 29
  • 43. PSO reformulation Using a standard notation for linear systems, we can split Xj(k) into the free response XjL(k) and the forced response XjF (k), so that Xj (k) = XjL(k) + XjF (k) where XjL(k) = Φ(k)Xj (0), XjF (k) = k−1 τ=0 H(k − τ)Uj(τ) and Φ(k) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   k ∈ IR2n×2n H(k − τ) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   k−τ−1 ∈ IR2n×2n Uj (τ) =   χ crpτ j + ¯c¯rpτ g χ crpτ j + ¯c¯rpτ g   ∈ IR2n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 12 / 29
  • 44. PSO reformulation Using a standard notation for linear systems, we can split Xj(k) into the free response XjL(k) and the forced response XjF (k), so that Xj (k) = XjL(k) + XjF (k) where XjL(k) = Φ(k)Xj (0), XjF (k) = k−1 τ=0 H(k − τ)Uj(τ) and Φ(k) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   k ∈ IR2n×2n H(k − τ) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   k−τ−1 ∈ IR2n×2n Uj (τ) =   χ crpτ j + ¯c¯rpτ g χ crpτ j + ¯c¯rpτ g   ∈ IR2n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 12 / 29
  • 45. PSO reformulation Using a standard notation for linear systems, we can split Xj(k) into the free response XjL(k) and the forced response XjF (k), so that Xj (k) = XjL(k) + XjF (k) where XjL(k) = Φ(k)Xj (0), XjF (k) = k−1 τ=0 H(k − τ)Uj(τ) and Φ(k) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   k ∈ IR2n×2n H(k − τ) =   χwIn −χ(cr + ¯c¯r)In χwIn [1 − χ(cr + ¯c¯r)] In   k−τ−1 ∈ IR2n×2n Uj (τ) =   χ crpτ j + ¯c¯rpτ g χ crpτ j + ¯c¯rpτ g   ∈ IR2n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 13 / 29
  • 46. Near Orthogonality of particles’ trajectories The free response XjL(k) only depends on the initial point Xj(0), and not on the vectors pτ j , pτ g, with τ ≥ 0. The velocity vk j of the j-th particle at iteration k may be regarded as a search direction from the current position xk j . lim k→∞ XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus). We can enforce diversification / intensification by setting Φ(k) in an appropriated way: 0 < χw < 1; 0 < χ(cr + ¯c¯r) < 2(χw + 1). G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
  • 47. Near Orthogonality of particles’ trajectories The free response XjL(k) only depends on the initial point Xj(0), and not on the vectors pτ j , pτ g, with τ ≥ 0. The velocity vk j of the j-th particle at iteration k may be regarded as a search direction from the current position xk j . lim k→∞ XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus). We can enforce diversification / intensification by setting Φ(k) in an appropriated way: 0 < χw < 1; 0 < χ(cr + ¯c¯r) < 2(χw + 1). G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
  • 48. Near Orthogonality of particles’ trajectories The free response XjL(k) only depends on the initial point Xj(0), and not on the vectors pτ j , pτ g, with τ ≥ 0. The velocity vk j of the j-th particle at iteration k may be regarded as a search direction from the current position xk j . lim k→∞ XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus). We can enforce diversification / intensification by setting Φ(k) in an appropriated way: 0 < χw < 1; 0 < χ(cr + ¯c¯r) < 2(χw + 1). G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
  • 49. Near Orthogonality of particles’ trajectories The free response XjL(k) only depends on the initial point Xj(0), and not on the vectors pτ j , pτ g, with τ ≥ 0. The velocity vk j of the j-th particle at iteration k may be regarded as a search direction from the current position xk j . lim k→∞ XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus). We can enforce diversification / intensification by setting Φ(k) in an appropriated way: 0 < χw < 1; 0 < χ(cr + ¯c¯r) < 2(χw + 1). G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
  • 50. Near Orthogonality of particles’ trajectories The free response XjL(k) only depends on the initial point Xj(0), and not on the vectors pτ j , pτ g, with τ ≥ 0. The velocity vk j of the j-th particle at iteration k may be regarded as a search direction from the current position xk j . lim k→∞ XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus). We can enforce diversification / intensification by setting Φ(k) in an appropriated way: 0 < χw < 1; 0 < χ(cr + ¯c¯r) < 2(χw + 1). G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
  • 51. Near Orthogonality of particles’ trajectories The free response XjL(k) only depends on the initial point Xj(0), and not on the vectors pτ j , pτ g, with τ ≥ 0. The velocity vk j of the j-th particle at iteration k may be regarded as a search direction from the current position xk j . lim k→∞ XjL(k) = 0 when Φ(k) eigenvalues are ≤ 1 (modulus). We can enforce diversification / intensification by setting Φ(k) in an appropriated way: 0 < χw < 1; 0 < χ(cr + ¯c¯r) < 2(χw + 1). G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 14 / 29
  • 52. Near Orthogonality of particles’ trajectories We set the initial position and velocity of the particles, so that the subvectors {νk j } (first n entries of the free responses) {XjL(k)}, are mutually orthogonal: We assign 1 to one variable, and 0 to all other variables if P ≤ 2n. If P > 2n, then set the initial position/velocity of the first 2n particles as stated, while the other particles may have whatever initial position/velocity. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 15 / 29
  • 53. Near Orthogonality of particles’ trajectories We set the initial position and velocity of the particles, so that the subvectors {νk j } (first n entries of the free responses) {XjL(k)}, are mutually orthogonal: We assign 1 to one variable, and 0 to all other variables if P ≤ 2n. If P > 2n, then set the initial position/velocity of the first 2n particles as stated, while the other particles may have whatever initial position/velocity. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 15 / 29
  • 54. Basics of Portfolio Selection Given a set of assets, the aim is to decide in which assets to invest and by how much in order to optimise some specific criterion. Minimise the risk given an expected return [Markowitz ’52]. Different risk measures can be used. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 16 / 29
  • 55. Basics of Portfolio Selection Given a set of assets, the aim is to decide in which assets to invest and by how much in order to optimise some specific criterion. Minimise the risk given an expected return [Markowitz ’52]. Different risk measures can be used. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 16 / 29
  • 56. Basics of Portfolio Selection Given a set of assets, the aim is to decide in which assets to invest and by how much in order to optimise some specific criterion. Minimise the risk given an expected return [Markowitz ’52]. Different risk measures can be used. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 16 / 29
  • 57. Markowitz Information about future prices is contained in historical series. We are given a set (Universe) of assets {a1 . . . an}. Each asset ai has associated a mean return ri and a return variance σ2 i . For each pair of assets (ai, aj ) we know the return covariance σij . A portfolio is a vector of real values P = x1 . . . xn. rp = n i=1 rixi . σp = n i=1 n j=1 σijxi xj . G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
  • 58. Markowitz Information about future prices is contained in historical series. We are given a set (Universe) of assets {a1 . . . an}. Each asset ai has associated a mean return ri and a return variance σ2 i . For each pair of assets (ai, aj ) we know the return covariance σij . A portfolio is a vector of real values P = x1 . . . xn. rp = n i=1 rixi . σp = n i=1 n j=1 σijxi xj . G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
  • 59. Markowitz Information about future prices is contained in historical series. We are given a set (Universe) of assets {a1 . . . an}. Each asset ai has associated a mean return ri and a return variance σ2 i . For each pair of assets (ai, aj ) we know the return covariance σij . A portfolio is a vector of real values P = x1 . . . xn. rp = n i=1 rixi . σp = n i=1 n j=1 σijxi xj . G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
  • 60. Markowitz Information about future prices is contained in historical series. We are given a set (Universe) of assets {a1 . . . an}. Each asset ai has associated a mean return ri and a return variance σ2 i . For each pair of assets (ai, aj ) we know the return covariance σij . A portfolio is a vector of real values P = x1 . . . xn. rp = n i=1 rixi . σp = n i=1 n j=1 σijxi xj . G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
  • 61. Markowitz Information about future prices is contained in historical series. We are given a set (Universe) of assets {a1 . . . an}. Each asset ai has associated a mean return ri and a return variance σ2 i . For each pair of assets (ai, aj ) we know the return covariance σij . A portfolio is a vector of real values P = x1 . . . xn. rp = n i=1 rixi . σp = n i=1 n j=1 σijxi xj . G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
  • 62. Markowitz Information about future prices is contained in historical series. We are given a set (Universe) of assets {a1 . . . an}. Each asset ai has associated a mean return ri and a return variance σ2 i . For each pair of assets (ai, aj ) we know the return covariance σij . A portfolio is a vector of real values P = x1 . . . xn. rp = n i=1 rixi . σp = n i=1 n j=1 σijxi xj . G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 17 / 29
  • 63. Markowitz we impose a minimum required return re min n i=1 n j=1 σij xi xj n i=1 ri xi ≥ re n i=1 xi = 1 G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 18 / 29
  • 64. Markowitz we impose a minimum required return re min n i=1 n j=1 σij xi xj n i=1 ri xi ≥ re n i=1 xi = 1 G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 18 / 29
  • 65. Markowitz we impose a minimum required return re min n i=1 n j=1 σij xi xj n i=1 ri xi ≥ re n i=1 xi = 1 G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 18 / 29
  • 66. Markowitz we impose a minimum required return re min n i=1 n j=1 σij xi xj n i=1 ri xi ≥ re n i=1 xi = 1 G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 18 / 29
  • 67. Coherence [Artzner et al. ’99] Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y); Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments increase risk); Positive homogeneity: ρ(λx) = λρ(x) (liquidity); Translation invariance: ρ(x + αr0) = ρ(x) − α. Markowitz is not coherent G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
  • 68. Coherence [Artzner et al. ’99] Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y); Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments increase risk); Positive homogeneity: ρ(λx) = λρ(x) (liquidity); Translation invariance: ρ(x + αr0) = ρ(x) − α. Markowitz is not coherent G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
  • 69. Coherence [Artzner et al. ’99] Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y); Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments increase risk); Positive homogeneity: ρ(λx) = λρ(x) (liquidity); Translation invariance: ρ(x + αr0) = ρ(x) − α. Markowitz is not coherent G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
  • 70. Coherence [Artzner et al. ’99] Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y); Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments increase risk); Positive homogeneity: ρ(λx) = λρ(x) (liquidity); Translation invariance: ρ(x + αr0) = ρ(x) − α. Markowitz is not coherent G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
  • 71. Coherence [Artzner et al. ’99] Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y); Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments increase risk); Positive homogeneity: ρ(λx) = λρ(x) (liquidity); Translation invariance: ρ(x + αr0) = ρ(x) − α. Markowitz is not coherent G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
  • 72. Coherence [Artzner et al. ’99] Monotonicity: x ≤ y implies ρ(x) ≥ ρ(y); Sub-additivity: ρ(x + y) ≤ ρ(x) + ρ(y) (no new investments increase risk); Positive homogeneity: ρ(λx) = λρ(x) (liquidity); Translation invariance: ρ(x + αr0) = ρ(x) − α. Markowitz is not coherent G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 19 / 29
  • 73. Our Optimisation Problem [Chen and Wang ’08] ρa,p(x) = a (x − E[x])+ 1 + (1 − a) (x − E[x])− p − E[x] rP ≥ re n i=1 xi = 1 Kd ≤ n i=1 zi ≤ Ku zi d ≤ xi ≤ zi u, i = 1, . . . , n zi (zi − 1) = 0, i = 1, . . . , n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
  • 74. Our Optimisation Problem [Chen and Wang ’08] ρa,p(x) = a (x − E[x])+ 1 + (1 − a) (x − E[x])− p − E[x] rP ≥ re n i=1 xi = 1 Kd ≤ n i=1 zi ≤ Ku zi d ≤ xi ≤ zi u, i = 1, . . . , n zi (zi − 1) = 0, i = 1, . . . , n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
  • 75. Our Optimisation Problem [Chen and Wang ’08] ρa,p(x) = a (x − E[x])+ 1 + (1 − a) (x − E[x])− p − E[x] rP ≥ re n i=1 xi = 1 Kd ≤ n i=1 zi ≤ Ku zi d ≤ xi ≤ zi u, i = 1, . . . , n zi (zi − 1) = 0, i = 1, . . . , n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
  • 76. Our Optimisation Problem [Chen and Wang ’08] ρa,p(x) = a (x − E[x])+ 1 + (1 − a) (x − E[x])− p − E[x] rP ≥ re n i=1 xi = 1 Kd ≤ n i=1 zi ≤ Ku zi d ≤ xi ≤ zi u, i = 1, . . . , n zi (zi − 1) = 0, i = 1, . . . , n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
  • 77. Our Optimisation Problem [Chen and Wang ’08] ρa,p(x) = a (x − E[x])+ 1 + (1 − a) (x − E[x])− p − E[x] rP ≥ re n i=1 xi = 1 Kd ≤ n i=1 zi ≤ Ku zi d ≤ xi ≤ zi u, i = 1, . . . , n zi (zi − 1) = 0, i = 1, . . . , n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
  • 78. Our Optimisation Problem [Chen and Wang ’08] ρa,p(x) = a (x − E[x])+ 1 + (1 − a) (x − E[x])− p − E[x] rP ≥ re n i=1 xi = 1 Kd ≤ n i=1 zi ≤ Ku zi d ≤ xi ≤ zi u, i = 1, . . . , n zi (zi − 1) = 0, i = 1, . . . , n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
  • 79. Our Optimisation Problem [Chen and Wang ’08] ρa,p(x) = a (x − E[x])+ 1 + (1 − a) (x − E[x])− p − E[x] rP ≥ re n i=1 xi = 1 Kd ≤ n i=1 zi ≤ Ku zi d ≤ xi ≤ zi u, i = 1, . . . , n zi (zi − 1) = 0, i = 1, . . . , n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
  • 80. Our Optimisation Problem [Chen and Wang ’08] ρa,p(x) = a (x − E[x])+ 1 + (1 − a) (x − E[x])− p − E[x] rP ≥ re n i=1 xi = 1 Kd ≤ n i=1 zi ≤ Ku zi d ≤ xi ≤ zi u, i = 1, . . . , n zi (zi − 1) = 0, i = 1, . . . , n G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 20 / 29
  • 81. Exact Penalty Approach [di Tollo and Roli, 2008] minx,z ρa,p(x) + 1 ε 7 ℓ=1 pℓ p1 = max(0, re − rP) p2 = n x=1 xi − 1 p3 + p4 = n x=1 max (0, zi · d − xi ) + n x=1 max (0, (xi − zi · u)) p5 + p6 = max 0, (Kd − n x=1 zi ) + max 0, ( n x=1 zi − Ku) p7 = n x=1 |zi · (1 − zi )| G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
  • 82. Exact Penalty Approach [di Tollo and Roli, 2008] minx,z ρa,p(x) + 1 ε 7 ℓ=1 pℓ p1 = max(0, re − rP) p2 = n x=1 xi − 1 p3 + p4 = n x=1 max (0, zi · d − xi ) + n x=1 max (0, (xi − zi · u)) p5 + p6 = max 0, (Kd − n x=1 zi ) + max 0, ( n x=1 zi − Ku) p7 = n x=1 |zi · (1 − zi )| G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
  • 83. Exact Penalty Approach [di Tollo and Roli, 2008] minx,z ρa,p(x) + 1 ε 7 ℓ=1 pℓ p1 = max(0, re − rP) p2 = n x=1 xi − 1 p3 + p4 = n x=1 max (0, zi · d − xi ) + n x=1 max (0, (xi − zi · u)) p5 + p6 = max 0, (Kd − n x=1 zi ) + max 0, ( n x=1 zi − Ku) p7 = n x=1 |zi · (1 − zi )| G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
  • 84. Exact Penalty Approach [di Tollo and Roli, 2008] minx,z ρa,p(x) + 1 ε 7 ℓ=1 pℓ p1 = max(0, re − rP) p2 = n x=1 xi − 1 p3 + p4 = n x=1 max (0, zi · d − xi ) + n x=1 max (0, (xi − zi · u)) p5 + p6 = max 0, (Kd − n x=1 zi ) + max 0, ( n x=1 zi − Ku) p7 = n x=1 |zi · (1 − zi )| G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
  • 85. Exact Penalty Approach [di Tollo and Roli, 2008] minx,z ρa,p(x) + 1 ε 7 ℓ=1 pℓ p1 = max(0, re − rP) p2 = n x=1 xi − 1 p3 + p4 = n x=1 max (0, zi · d − xi ) + n x=1 max (0, (xi − zi · u)) p5 + p6 = max 0, (Kd − n x=1 zi ) + max 0, ( n x=1 zi − Ku) p7 = n x=1 |zi · (1 − zi )| G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
  • 86. Exact Penalty Approach [di Tollo and Roli, 2008] minx,z ρa,p(x) + 1 ε 7 ℓ=1 pℓ p1 = max(0, re − rP) p2 = n x=1 xi − 1 p3 + p4 = n x=1 max (0, zi · d − xi ) + n x=1 max (0, (xi − zi · u)) p5 + p6 = max 0, (Kd − n x=1 zi ) + max 0, ( n x=1 zi − Ku) p7 = n x=1 |zi · (1 − zi )| G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 21 / 29
  • 87. Benchmarks FTSE MIB (32 assets, 1396 days) DJIA (32 assets, 9312 days) G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 22 / 29
  • 88. Benchmarks FTSE MIB (32 assets, 1396 days) DJIA (32 assets, 9312 days) G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 22 / 29
  • 89. Best Objective over time Figure: Experiments with ORTHOinit G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 23 / 29
  • 90. Best Objective over time Figure: Experiments without ORTHOinit G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 24 / 29
  • 91. Results Exact Solver re ρ Constraints Violated Computational time (sec) min(ri ) 0.0046 — 25693 max(ri ) 0.021 — 7064 PSO with ORTHOinit re ρ Constraints Violated Computational time (sec) min(ri ) 0.0048 Lower Bound 5375 max(ri ) 0.0238 Lower Bound, Capital, re 7021 PSO without ORTHOinit re ρ Constraints Violated Computational time (sec) min(ri ) 0.0042 Lower Bound 6530 max(ri ) 0.0238 Lower Bound, Capital, re 6135 Table: Experimental Results, Instance DJIA, basic PSO G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 25 / 29
  • 92. Results ρ Success Ratio Strategy Min Max Std Mean PSO-newinit-REVAC 0.00683077 0.00872689 0.00109750 46.6 PSO-standard 0.00691271 0.15525469 0.07204026 23.3 NEOS 0.00658258 0.00658258 Table: Experimental Results, Instance FTSI MIB, enhanced PSO. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 26 / 29
  • 93. Conclusions ORTHOinit fosters a better diversification; Computational times; The exact penalty approach should be revised, trying different reformulations of our portfolio selection problem, where possibly the simple constraints (i.e. linear constraints) are not moved to the penalty function. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 27 / 29
  • 94. References [Artzner et al ’99]: P. Artzner, F. Delbaen, J.M. Eber, D. Heath (1999), Coherent measures of risk, Mathematical Finance, vol. 9, pp. 203–228. [Blum and Roli ’03]: C. Blum, A. Roli (2003), Metaheuristics in combinatorial optimization: Overview and conceptual comparison, ACM Computing Surveys, vol. 35 (3), pp. 268–308. [Chen and Wang ’08]: Z. Chen, Y. Wang (2008), Two-sided coherent risk measures and their application in realistic portfolio optimization, Journal of Banking & Finance, vol. 32, pp. 2667–2673. [Corazza et al. ’13]: M.Corazza, G.Fasano, R.Gusso (2013), Particle Swarm Optimization with non-smooth penalty reformulation for a complex portfolio selection problem, Applied Mathematics and Computation, vol. 224, pp. 611–624. [di Tollo et al. ’11]: G. di Tollo, F. Lardeux, J. Maturana, F. Saubion (2011), From Adaptive to More Dynamic Control in Evolutionary Algorithms, in EvoCOP 2011, LNCS 6622 Proceedings. G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 28 / 29
  • 95. References [di Tollo and Roli, ’08]: G. di Tollo, A. Roli (2008), Metaheuristics for the portfolio selection problem, International Journal of Operations Research, vol. 5 (1), pp. 13–35. [Fasano et al ’14]: G. Fasano, G. Liuzzi, S. Lucidi, F. Rinaldi (2014), A Linesearch-based Derivative-free Approach for Nonsmooth Constrained Optimization , SIAM Journal on Optimization, vol. 24 (3), pp. 959-992. [Markowitz ’52]: H. Markowitz (1952), Portfolio Selection, The Journal of Finance, vol. 7 (1), pp. 77–91. [Nocedal and Wright ’00]: J. Nocedal, S. Wright (2000), Numerical Optimization (Springer Series in Operations Research and Financial Engineering). G. di Tollo (et al.) PSO for Portfolio Selection 6th December,2016 29 / 29