Bootstrap of constraint estimators with application to rank estimation.
Fran¸cois Portier
IRMAR-University of Rennes 1
April 17, 2012
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 1 / 20
Table of contents
1 Bootstrap, hypothesis testing, estimation under constraint
2 Application to rank estimation
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 2 / 20
Introduction to Bootstrap
Goal: To reproduce the asymptotic behavior of some estimators.
Means: Creation of a new sample which “look like” the previous one.
•Bootstrap of Efron [Efron(1982)] :
Suppose that (X1, ...,Xn) i.i.d. with lawP. We draw (X1∗, ...,Xn∗) with respect to the law
Pb=n−1
n
X
i=1
δXi.
Defineθ0=E[X], X =1nPn
i=1Xi and X∗= 1
n
n
X
i=1
Xi∗= 1 n
n
X
i=1
NiXi withNi∼ mult(1/n)
•Another bootstrap method:
X∗=X+1 n
n
X
i=1
iXi= 1 n
n
X
i=1
(i+ 1)Xi
withi any i.i.d. sequence standard random variable.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 3 / 20
Introduction to Bootstrap
Example of results with both previous bootstrap methods
Ifφis continuously differentiable on a neighborhood ofθ0=E[X], ifP has a finite second order moment 2, then
√n(φ(X∗)−φ(X)) bootstrap √
n(φ(X)−φ(θ0)), i.e.
L(√
n(φ(X∗)−φ(X))|bP)n→∞= L(√
n(φ(X)−φ(θ0))|P).
Why the bootstrap ? Alternative to the use of the asymptotic law ([Hall(1992)]) for
Building confidence interval
Hypothesis testing (for the choice of quantile)
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 4 / 20
Test of equal means: classical bootstrap works
Assumeθ0∈R,
H0: θ0=θ against H1: θ06=θ To arbitrate:
k√
n(X−θ)k2is compared to
q∞α a quantile of the limiting distribution qα∗ a quantile of the bootstrap statistic
Level and power
PH0(k√
n(X−θ)k2>q∞α orq∗α) and PH1(k√
n(X −θ)k2>qα∞or qα∗) Forq∞α: OK.
Forq∗α: √
n(X∗−X)
| {z }
do not depend onH0orH1
bootstrap √
n(X−θ0) ⇒OK.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 5 / 20
In general classical bootstrap fails
Assumeθ0∈R2andCis the unit circle,
H0: θ0∈ C against H1: θ0∈ C/
⇒Constraint estimators :
Tbn=n min
g(θ)=0kX−θk2
Does the classical Bootstrap works ?
UnderH0, Tbn=|√
n(φ(X)−φ(θ0))|2 withφ:x → min
g(θ)=0kx−θk Bootstrap candidate:
Tn∗=|√
n(φ(X∗)−φ(X))|2 Can not work becauseφis notC1.
⇒Even if we can bootstrap√
n(X−θ0), it is not clear we are able to bootstrap some constraint estimators.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 6 / 20
From nowθ0∈Rp (parameter of interest), it existsθbsome consistent estimators ofθ0. Define the random function
Qbn(θ) = (bθ−θ)TSb(bθ−θ).
Question
If we can bootstrap√
n(bθ−θ0), does theunder H0-law of
√n(bθc−θ0) with bθc= argmin
g(θ)=0
Qb(θ) can be bootstrapped ?
Applications
Statistics of the kind min
g(θ)=0
Q(θ) to arbitrate between
H0: g(θ0) = 0 and H1:g(θ0)6= 0
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 7 / 20
Intuitively
Define
θc∗= argmin
g(θ)=0
Q∗(θ) and Q∗(θ) = (θ∗−θ)TS∗(θ∗−θ).
As traditional bootstrap: we expect results such as
√n(θ∗c−θbc) bootstrap√
n(bθc−θ0)
The idea : A good choice of θ
∗We try to ”reproduce”H0with
θ∗=θbc+ ”something going to 0 with good speed and variance”
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 8 / 20
θbc = argmin
g(θ)=0
(bθ−θ)TbS(bθ−θ) and θ∗c = argmin
g(θ)=0
(θ∗−θ)TS∗(θ∗−θ) Assumptions :
1 Sb−→P S andS∗−→P S.
2 S is full rank.
3 g :Rp→Rq isC1on a neighborhood ofθ0andJg(θ0) is of full rank.
Theorem
Under H0, if√
n(θ∗−θbc)bootstrap√
n(bθ−θ0)→d X (Gaussian) we have
√n(bθc∗−θbc)bootstrap √
n(bθc−θ0)under H0.
| {z }
(Gaussian limit)
(1.1)
Under H1, we additionally need to assume the existence ofθc such asθbc a.s.→θc
with g(θc) = 0, to get (1.1).
⇒ θ∗=θbc+ (θ∗classical−θ) withb θ∗classical comes from any methods of classical bootstrap.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 9 / 20
What about T
∗?
Corollary
Under the previous set of assumptions under H0 and H1, T∗= argmin
g(θ)=0
(θ∗−θ)TS∗(θ∗−θ) bootstrap Tb under H0
| {z }
weighted Chi-squared limit
Problem
The assumption for convergence underH1: θbc a.s.→ θc need to be check for each case.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 10 / 20
Assumptions :
1 √
n(θ∗−bθc) bootstrap√
n(bθ−θ0)→d X (Gaussian)
2 Sb−→P S andS∗−→P S
3 g :Rp→Rq isC1on a neighborhood ofθ0andJg(θ0) is of full rank.
4 S is full rank.
Corollary 2
The test with null hypothesis
H0: g(θ0) = 0 againstH1: g(θ0)6= 0
and associated statisticTb with bootstrap calculation of quantile is consistent.
For the test procedure, one can draw
T1∗, ...,TB∗ to estimateq∗α and we do not rejectH0ifTb ≤q∗, or rejectH0 if not.
In other words Corollary 2 means:
⇒The asymptotic level of the test isα.
⇒The power of the test goes to 1.
Rk: This kind of test is pivotal (Chi-squared) whenS =Var(X)−1.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 11 / 20
Application to rank estimation
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 12 / 20
Framework and notation
Goal: Estimation of the rank of a matrix Means: Hypothesis testing.
Assumptions
Mb andM are matricesRp×H such that
√n(~(M)b −vec(M))−→ Nd (0,Γ) bΓ−→P Γ
Nothing moreor Γ invertibleorΓ =FFT⊗GGT invertible.
Notations: rank(M) =d0, SVD ofM andM:b
M= (U1U0)
D1 0
0 0
V1T V0T
and Mb = (bU1Ub0) Db1 0 0 Db0
! Vb1T Vb0T
!
P1=U1U1T,Q1=U0U0T,P2=V1V1T,Q2=V0V0T,Pb1,Pb2,Qb1Qb2.
(bλ1, ...,bλp), (resp. (λ1, ..., λp)) singularvalues ofMb (resp. M) in ascending order.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 13 / 20
Short review
Ford= 0, ...,d0, we test
H0: d0=d against H1: d0>d
Some statistics
[Li(1991)] Tb1=n
p−d
X
k=1
bλ2k (=nkvec(Qb1MbQb2)k2) [Bura and Yang(2011)] Tb2=nvec(Qb1MbQb2)TbΓ+vec(Qb1MbQb2) [Cragg and Donald(1997)] Tb3=n min
rank(M)=d
vec(Mb −M)TbΓ−1vec(Mb −M) By noting that
Lemma (From PCA)
Pb1MbPb2= argmin
rank(M)=d
kMb −Mk2F = argmin
rank(M)=d
kvec(Mb −M)k2
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 14 / 20
we get
[Li(1991)] Tb1=n min
rank(M)=dkvec(Mb −M)k2 [Bura and Yang(2011)] Tb2=nvec(Mb −Mbc)TbΓ+vec(Mb −Mbc) [Cragg and Donald(1997)] Tb3=n min
rank(M)=dvec(Mb −M)TbΓ+vec(Mb −M) withMbc = argmin
rank(M)=d
kvecv(Mb −M)k2.
Application of the results
{rank(M) =d}is a smooth submanifold.
Example of sufficient conditions for bootstrap: It existsξ1, ..., ξn i.i.d. with E[kξ1k2F]<+∞such that Mb = 1nPn
i=1ξi.
⇒Example for Tb1andTb2:M∗=Pb1MbPb2+1 n
n
X
i=1
iξi
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 15 / 20
Example under H
1i.e. d < d
0when θ b
cdoes not converge
We need to ensure the a.s. convergence of
θbc = argmin
rank(M)=d
kvec(M)b −vec(M)k2=Pb1MbPb2
⇒problem of convergence of eigenprojectors Riesz formula: Pλ=H
Cλ(Iz−M)−1dz.
Suppose thatM andMb are symetric withMb a.s.→M, then Pb=H
Cb(Iz−M)b −1dz
ifλp−d+16=λp−d
=
from a certain rank
H
C(Iz−M)b −1dz Ifλp−d+1=λp−d thenP does not exists. Rk: Application of the previous results toMbMbT andMbTM.b
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 16 / 20
Example under H
1i.e. d < d
0when θ b
cdoes not converge
We need to ensure the a.s. convergence of
θbc = argmin
rank(M)=d
kvec(M)b −vec(M)k2=Pb1MbPb2
⇒problem of convergence of eigenprojectors Riesz formula: Pλ=H
Cλ(Iz−M)−1dz.
Suppose thatM andMb are symetric withMb a.s.→M, then Pb=H
Cb(Iz−M)b −1dz ifλp−d+1=6=λp−d
from a certain rank
H
C(Iz−M)b −1dz
Ifλp−d+1=λp−d thenP does not exists. Rk: Application of the previous results toMbMbT andMbTM.b
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 16 / 20
Example under H
1i.e. d < d
0when θ b
cdoes not converge
We need to ensure the a.s. convergence of
θbc = argmin
rank(M)=d
kvec(M)b −vec(M)k2=Pb1MbPb2
⇒problem of convergence of eigenprojectors Riesz formula: Pλ=H
Cλ(Iz−M)−1dz.
Suppose thatM andMb are symetric withMb a.s.→M, then Pb=H
Cb(Iz−M)b −1dz ifλp−d+1=6=λp−d
from a certain rank
H
C(Iz−M)b −1dz Ifλp−d+1=λp−d thenP does not exists.
Rk: Application of the previous results toMbMbT andMbTM.b
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 16 / 20
Conclusion
Concluding remarks
We provide a general bootstrap procedure for constraint estimator associate to a quadratic function.
The test procedure associate is consistent.
Large application thanks to hypothesis testing.
As an example, it can easily be applied to rank estimation.
Work in progress
Alleviate the underH1 assumptionθca.s.→θc for theTb stat.
Possibility to extend such results toM−estimator,Z−estimator.
Simulation study : bootstrap vs asymptotic,
and also constraint bootstrap vs traditional bootstrap.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 17 / 20
E. Bura and J. Yang.
Dimension estimation in sufficient dimension reduction: a unifying approach.
J. Multivariate Anal., 102(1):130–142, 2011.
John G. Cragg and Stephen G. Donald.
Inferring the rank of a matrix.
J. Econometrics, 76(1-2):223–250, 1997.
Bradley Efron.
The jackknife, the bootstrap and other resampling plans, volume 38 of CBMS-NSF Regional Conference Series in Applied Mathematics.
Society for Industrial and Applied Mathematics (SIAM), Philadelphia, Pa., 1982.
Peter Hall.
The bootstrap and Edgeworth expansion.
Springer Series in Statistics. Springer-Verlag, New York, 1992.
Ker-Chau Li.
Sliced inverse regression for dimension reduction.
J. Amer. Statist. Assoc., 86(414):316–342, 1991.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 18 / 20
SIR
Sufficient dimension reduction (SDR) introduit par [Li(1991)]: on suppose le mod`ele de r´egression suivant,
Y =g(PZ, ε), Z ⊥⊥ε
o`uY ∈R,Z ∈Rp, P est un projecteur orthogonal de rangd0etg est inconnue.
But de la SDR : Estimation de P .
Enjeux : Obtenir une meilleur vitesse lors de l’estimation de g .
L’inf´erence surP se base sur
E[Z|Y]∈Im(P) p.s.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 18 / 20
SIR
On partitionne l’image deY enH tranches appel´eesI(h)
Enjeux de SIR
Estimer l’espace engendr´e par les vecteurs
E[Z|Y ∈I(1)], . . . ,E[Z|Y ∈I(H)]
Procedure de SIR:
1/ Estimation de
Ch=E[Z1{Y∈I(h)}]∈Ec pour h= 1, ...,H.
2/ Extraire une base de span(bC1, ...,CbH) : Elements propres de la matrice MbSIR =X
h
bp−1h CbhCbhT avecph=P(Y ∈I(h)).
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 19 / 20
Trouver la dimension
En notantηb1, ...,ηbp les vecteurs propres deMbSIR dans l’ordre croissant des v.p., on peut estimerPde mani`ere consistante par
Pb =
d0
X
k=1
ηbkbηkT,
maisd0 est inconnu.
Importance de bien estimer d
0Perte dans la valeur explicative du mod`ele.
Vitesse non-param´etrique mauvaise.
Fran¸cois Portier (IRMAR) Bootstrap of constraint estimators April 17, 2012 20 / 20