See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/236680545

On the singular values of the Hankel matrix with

application in singular spectrum analysis
Article February 2011

CITATIONS

READS

51

2 authors:
Rahim Mahmoudvand

Mohammad Zokaei

Buali Sina University

Shahid Beheshti University

23 PUBLICATIONS 173 CITATIONS

7 PUBLICATIONS 74 CITATIONS

SEE PROFILE

SEE PROFILE

Available from: Rahim Mahmoudvand

Retrieved on: 23 August 2016

Chilean Journal of Statistics

Vol. 3, No. 1, April 2012, 4356

Time Series
Research Paper

On the singular values of the Hankel matrix with

application in singular spectrum analysis
1

Rahim Mahmoudvand1, and Mohammad Zokaei1

Department of Statistics, Shahid Beheshti University, Tehran, Iran
(Received: 21 June 2011 Accepted in final form: 24 August 2011)

Abstract
Hankel matrices are an important family of matrices that play a fundamental role in
diverse fields of study, such as computer science, engineering, mathematics and statistics.
In this paper, we study the behavior of the singular values of the Hankel matrix by
changing its dimension. In addition, as an application, we use the obtained results for
choosing the optimal values of the parameters of singular spectrum analysis, which is a
powerful technique in time series based on the Hankel matrix.

Keywords: Eigenvalues Singular spectrum analysis.

Mathematics Subject Classification: Primary 15A18 Secondary 37M10.

1.

Introduction

A Hankel matrix can be finite or infinite and its (i, j) entry is a function of i+j; see Widom
(1966). In other words, a matrix whose entries are the same along the anti-diagonals is
called the Hankel matrix. Specifically, an L K Hankel matrix H is a rectangular matrix
of the form

h1 h2 . . . hK
h2 h3 . . . hK+1

H = ..
(1)
.. . .
.. ,
.
. .
.
hL hL+1 . . . hN
where K = N L + 1.
Hankel matrices play many roles in diverse areas of mathematics, such as approximation
and interpolation theory, stability theory, system theory, theory of moments and theory
Corresponding author: Rahim Mahmoudvand. Department of Statistics, Shahid Beheshti University, PO. Box
1983963113, Evin, Tehran, Iran. Email: [email protected]

ISSN: 0718-7912 (print)/ISSN: 0718-7920 (online)

c Chilean Statistical Society Sociedad Chilena de Estadstica

http://www.soche.cl/chjs

44

R. Mahmoudvand and M. Zokaei

of orthogonal polynomials, as well as in communication and control engineering, including

filter design, identification, model reduction and broadband matching; for more details, see
Peller (2003). Thus, this type of matrices has been subjected to intensive study with respect to its spectrum (collection of eigenvalues) and many interesting results were derived.
However, closed form computation of eigenvalues is not known and, consequently, the effect of changing the dimension of the matrix on its eigenvalues have not been investigated
in detail.
In recent years, singular spectrum analysis (SSA), a relatively novel, but powerful technique in time series analysis, has been developed and applied to many practical problems;
see, e.g., Golyandina et al. (2001), Hassani et al. (2009), Hassani and Thomakos (2010)
and references therein. The SSA decomposes the original time series into a sum of small
numbers of interpretable components, such as slowly varying trend, oscillatory component
and noise. The basic SSA method consists of two complementary stages: decomposition
and reconstruction; each stage includes two separate steps. At the first stage, we decompose the series and, at the second stage, we reconstruct the noise free series by using the
reconstructed series for forecasting new data points.
A short description of the SSA technique is given in the next section. For more explanations and comparison with other time series analysis techniques, refer to Hassani
(2007).
The whole procedure of the SSA technique depends upon two parameters:
(i) The window length, which is usually denoted by L.
(ii) The number of needed singular values, denoted by r, for reconstruction.
Improper choice values of parameters L or r may yield incomplete reconstruction and
misleading results in forecasting.
Considering a series of length N , Elsner and Tsonis (1996) provided some discussion
and remarked that choosing L = N/4 is a common practice. Golyandina et al. (2001)
recommended that L should be large enough, but not larger than N/2. Large values of
L allow longer period oscillations to be resolved, but choosing L too large leaves too few
observations from which to estimate the covariance matrix of the L variables. It should
be noted that variations in L may influence separability feature of the SSA technique: the
orthogonality and closeness of the singular values. There are some methods for selecting
L. For example, the weighted correlation between the signal and noise component has
been proposed in Golyandina et al. (2001) to determine the suitable value of L in terms
of separability.
Although considerable attempt and various techniques have been taken into account for
selecting the proper value of L, but there is not enough algebraic and theoretical materials
for choosing L and r. The aim of his paper is to obtain some theoretical properties of the
singular values of the Hankel matrix that can be used directly for choosing proper values
of the two parameters of the SSA.
The outline of this paper is as follows. Section 2 describes the SSA technique and also
shows the importance of a Hankel matrix for this technique. Section 3 provides the main
results of the paper. Section 4 discusses some examples and an application of the obtained
results. Section 5 sketches some conclusion of this work.

2.

Singular Spectrum Analysis

In this section, we briefly introduce stages of the SSA method and discuss the importance
of using a Hankel matrix in the development of this technique.

Chilean Journal of Statistics

45

2.1 Stage I: decomposition

1st step: embedding. Embedding is as a mapping that transfers a one-dimensional time

series YN = (y1 , . . . , yN ) into the multi-dimensional series X1 , . . . , XK with vectors Xi =

(yi , . . . , yi+L1 )> RL , where L (2 L N 1) is the window length and K = N L+1.
The result of this step is the trajectory matrix
X = (X1 , . . . , XK ) = (xij )L,K
i,j=1 .

(2)

Note that the matrix given in Equation (2) is a Hankel matrix as defined in Equation (1).
2nd step: singular value decomposition (SVD). In this step, we perform the SVD
of X. Denote by 1 , . . . , L the eigenvalues of XX> arranged in the decreasing order
(1 L 0) and by U1 , . . . , UL the corresponding
eigenvectors. The
SVD of X can

be written as X = X1 + + XL , where Xi = i Ui Vi > and Vi = X> Ui / i (if i = 0

we set Xi = 0).

2.2 Stage II: reconstruction

1st step: grouping. The grouping step corresponds to splitting the elementary matrices
into several groups and summing the matrices within each group. Let I = {i1 , . . . , ip },
for p < L, be a group of indices i1 , . . . , ip . Then, the matrix XI corresponding to the
group I is defined as XI = Xi1 + + Xip . The split of the set of indices {1, . . . , L}
into disjoint subsets I1 , . . . , Im corresponds to the representation X = XI1 + + XIm .
The procedure of choosing the sets I1 , . . . , Im is called the grouping. For a given group
I, the contribution
ofPthe component XI is measured by the share of the corresponding
P
eigenvalues iI i / di=1 i , where d is the rank of X.
2nd step: diagonal averaging. The purpose of diagonal averaging is to transform a

matrix Z to the form of a Hankel matrix HZ, which can be subsequently converted to a
time series. If zij stands for an element of a matrix Z, then the kth term of the resulting
series is obtained by averaging zij for all i, j such that i + j = k + 1. Hankelization HZ is
an optimal procedure, which is nearest to Z with respect to the matrix norm.
3.

Theoretical Results

Along the paper, the matrices to be considered are over the field of the real numbers. In
addition, we consider different values of L, whereas N is supposed to be fixed. Recall that,
for any operator A, the operator AA> is always positive, and its unique positive square
root is denoted by |A|. The eigenvalues of |A| counted with multiplicities are called the
singular values of A. In this section, we provide the main results of the paper.
3.1 On sum of square of the singular values of a Hankel matrix
Let L,N
denote the jth ordered eigenvalue of HH> . Then, for a fixed value of L, the trace
j
of HH> is given by
L,N
TH
= tr(HH> ) =

L
X
j=1

L,N
.
j

(3)

46

R. Mahmoudvand and M. Zokaei

L,N
The behavior of TH
given in Equation (3), with respect to different values of L, is
considered in the following theorem.

Theorem 3.1 Consider the Hankel matrix H as defined in Equation (1). Then,
L,N
TH
=

N
X

wjL,N h2j ,

j=1

where wjL,N = min{min{L, K}, j, L + K j} = wjK,N .

Proof Applying definition of H as given in Equation (1), we have
L,N
TH

L NX
L+i
X
i=1

h2j .

(4)

j=i

Changing the order of the summations in Equation (4), we get

L,N
TH
=

N
X

Cj,L,N h2j ,

j=1

where Cj,L,N = min{j, L} max{1, j N + L} + 1. Therefore, we only need to show that

Cj,L,N = wjL,N , for all j and L. We consider two cases: L K and L > K. For the first
case, we have

1 j L;
j,
L + 1 j K;
Cj,L,N = L,

N j + 1, K + 1 j N,
which is exactly equals to wjL,N . Similarly for the second case, we get

Cj,L,N

1 j K;
j,
K + 1 j L;
= K,

N j + 1, L + 1 j N ;

and again is equal to wjL,N , for L > K.

The weight wjL,N defined in Theorem 3.1 can be written in the functional form
wjL,N

N + 1 N 2+1 L N 2+1 j N 2+1 L N 2+1 j

=

.
2
2
2
2

(5)

Equation (5) shows that

L,N
is a concave function of L for all j, where j {1, . . . , N };
wj
L,N
is a concave function of j for all L, where L {2, . . . , N 1};
wj

L,N
is a symmetric function around line (N + 1)/2 with respect to j and L.
wj
L,N
The above mentioned results imply that the behavior of the quantity TH
is similar on
two intervals 2 L [(N + 1)/2] and [(N + 1)/2] + 1 L N 1, where, as usual, [x]
denotes the integer part of the number x. Therefore, we only need to consider one of these
intervals.

Chilean Journal of Statistics

47

L,N
L,N
Theorem 3.2 Let TH
be defined as in Equation (3). Then, TH
is an increasing function of L on {2, . . . , [(N + 1)/2]}, a decreasing function on {[(N + 1)/2] + 1, . . . , N 1},
and

[ N +1 ],N
L,N
max TH
= TH 2
.
Proof First, we show that wjL,N is an increasing function of L on {2, . . . , [(N + 1)/2]}.
Let L1 and L2 be two arbitrary values, where L1 < L2 [(N + 1)/2]. From the definition
of wjL,N , we have

wjL2 ,N wjL1 ,N

0,
1 j L1 ;

L1 + 1 j L2 ;
j L1 ,
L2 + 1 j N L2 + 1;
= L2 L1 ,

j
+
1

L
,
N
L2 + 2 j N L1 + 1;

0,
N L1 + 2 j N.

Therefore, wjL2 ,N wjL1 ,N 0, for all j, and inequality is strict for some j. Thus,

L2 ,N
L1 ,N
TH
TH
=

X
wjL2 ,N wjL1 ,N h2j > 0.

(6)

j=1
L,N
This confirms that TH
is an increasing function of L on {2, . . . , [(N + 1)/2]}. Similar
L,N
approach for the set {[(N + 1)/2]+1, . . . , N 1} implies that TH
is a decreasing function
L2 ,N
L1 ,N
of L on this interval. Note also that TH TH
in Equation (6) increases as the value
L,N
of L2 increases too proving that TH
is an increasing function on {2, . . . , [(N + 1)/2]}.
L2 ,N
Therefore, the maximum value of TH
is attained at the maximum value of L, which is
[(N + 1)/2].

L,N
Lmax ,N
Corollary 3.3 Let Lmax denote the value of L such that TH
TH
, for all L, and
the inequality to be strict for some values of L. Then,

Lmax =

N +1

2 ,

N
2

and

if N is odd;
N
+ 1, if N is even.
2

Corollary 3.3 shows that L = median{1, . . . , N } maximizes the sum of squares of the
Hankel matrix singular values with fixed values of N . Applying Corollary 3.3 and Equation
(5), we can show that
wjLmax ,N

N + 1 N + 1

j .
=
2
2

L,N
Equation (7) shows that h[(N +1)/2] has maximum weight at TH
.

(7)

48

R. Mahmoudvand and M. Zokaei

3.2 Eigenvalues of HH> and rank of H

Here, some inequalities between the ordered eigenvalues for different values of L are derived.
According to Cauchys interlacing theorem, it can be given the following theorem; see
Bhatia (1997).
Theorem 3.4 Let H be an L K Hankel matrix as defined in Equation (1). Then, we
have
m
L,N
Lm,N
L,N
j
j
j+m ,

j = 1, . . . , L m,

where m is a number belonging to the set {1, . . . , L 1}.

Proof Consider the partition

HH> =

>
HH>
(1) HH(3)

HH>
(2)

HH>
(4)

where

HH>
(1)

K
P

K
P

h2j

hj hj+1

j=1
j=1

P
K
K
P

hj+1 hj
h2j+1

j=1
j=1
=

..
..

.
.
K
K
P
P
hj+Lm1 hj
hj+Lm1 hj+1
j=1

j=1

...

K
P

hj hj+Lm1

...
hj+1 hj+Lm1
.
j=1

..
..
.

K
P 2

...
hj+Lm1
j=1
K
P

j=1

Using this partitioning form, we can say that the sub-matrix HH>
(1) is obtained from a
Hankel matrix corresponding to the sub-series HN m = (h1 , . . . , hN m ) and its eigenvalues
m
m
m
are Lm,N
Lm,N
Lm,N
0. Therefore, the proof is completed
1
2
Lm
using Cauchys interlacing theorem.

Now, we would like to find a relationship between Lm,N

and L,N
. Therefore, Theorem
j
j
3.4 should not use directly. Next, we consider four cases and show that we can find general
relationships for some classes of the Hankel matrix.
3.2.1 Case 1: L 1, rank of H = 1
In this case, it is obvious that we have one positive eigenvalue. Therefore, we can write
L,N
=
1

L
X
j=1

L,N
= tr(HH> ) =
j

L K+l1
X
X
l=1

h2j .

j=l

According to Theorem 3.2, eigenvalue L,N

increases with L till [(N + 1)/2] and then
1
decreases for L [(N + 1)/2] + 1. Therefore, we have Lm,N
L,N
and L [(N + 1)/2]
1
1
providing that the conditions of Case 1 are satisfied.

Chilean Journal of Statistics

49

3.2.2 Case 2: L = 2, rank of H = 2

In this case, HH> has at most two eigenvalues which are the solution of the quadratic
equation

N
1
X

h2j +

j=1

N
X

h2j +

j=2

N
1
X
j=1

h2j

N
X
j=2

N
1
X

h2j

2
hj hj+1 = 0.

(8)

j=1

Equation (8) has two real solutions so that we have two real eigenvalues. The first eigenvalue (larger one) is given by
NP
1

2,N
=
1

j=1

h2j +

N
P
j=2

!2
u
u
NP
1
2
t
2
2
2
h1 hN + 4
hj +
hj hj+1
j=1

(9)

Equation (9) shows that

2,N
1

1,N , PN 1 hj hj+1 h2 h2 ;
1 N
1
j=1

2
=
1,N , PN 1 h h
2
h1 h2N ;
j j+1
1
j=1

(10)

PN
2
where 1,N
=
1
j=1 hj , when L = 1. Practically, it seems that the first condition of
Equation (10) is usually satisfied for a wide classes of models. For example, it can be
seen that the condition is equivalent to monotonicity of the sequence {hj , j = 1, . . . , N }.
P 1
For a non-negative (or non-positive) monotone sequence, we have N
j=1 hj hj+1 h1 hN .
PN 1 2
1,N 1
2,N
. A greater class is obtained
Applying Equation (9), it follows 1 j=1 hj = 1
if we consider positive data, where all observations are bigger P
that the first one and h1
N 1
hN /(N 1). Under this condition, it is easy to show that
j=1 hj hj+1 h1 hN and
therefore 2,N
1,N
1
1 . In the next section, we see some examples of models that have not
these conditions but 2,N
1,N
1
1 .
It is worth mention that we can state a geometrical display of Equation (8) as

2 ||h1:N 1 ||2 + ||h2:N ||2 + ||h1:N 1 ||2 ||h2:N ||2 (sin (1,2 ))2 = 0,

(11)

where h1:N 1 and h2:N denote the first and second rows of H, ||.|| the Euclidean norm and
1,2 the angle between two rows of H. Notice that last expression in Equation (11) is the
magnitude of the cross product between two first rows of H. Since (sin (1,2 ))2 1, it is
1
easy to obtain the inequality 2,N
1,N
, which is a direct result of Theorem 3.4 from
1
1
characteristics given in Equation (11).
3.2.3 Case 3: L > 2, rank of H = 2
>

In this case, HH has two positive eigenvalues. To obtain the eigenvalues, first of all note
that

det I HH> = L + c1 L1 + + cL1 + cL ,

where the coefficients of cj can be obtained from following lemma.

(12)

50

R. Mahmoudvand and M. Zokaei

Lemma 3.5 (Horn and Johnson, 1985, Theorem 1.2.12) Let A be an n n real or complex
matrix with eigenvalues 1 , . . . , n . Then, for 1 k n,
(i) sk () = (1)k ck , and
(ii) sk () is the sum of all the k k principal minors of A.
Equation (12) shows that the eigenvalues of HH> in this case are the solution of
2

L K+l1
X
X
l=1

h2j +

l=1 i=1

j=l

h2j

L1
Ll
XX

K+l1
X
j=l

K+l1
X

h2j+i

j=l

k+l1
X
j=l

hj hj+i
= 0.

(13)

The first eigenvalue (larger one) is given by

L K+l1
X

X
p
1
2
L,N
=

,
h
+
L
j
1

2
l=1

(14)

j=l

where L is the discriminant of the quadratic expression given in Equation (13). According
to Equation (14), it is easy to see that, for L [(N + 1)/2],
L,N
j

L1,N
j

L+1
p
p N X

L1 L
h2l ,

j = 1, 2.

l=L

Similar to the previous case, Equation (13) may be reformulated in the language of multivariate geometry for the L-lagged vectors given by
2

L
X
j=1

||hj:K+j1 ||2 +

L1
X

L
X

||hi:K+i1 ||2 ||hj:K+j1 ||2 (sin (i,j ))2 = 0,

i=1 j=i+1

where notations are defined similarly as mentioned in Case 3.

3.2.4 Case 4: L > 2, rank of H > 2
Applying Equation (12), it can be obtained the characteristic equation whose solution
gives the eigenvalues of HH> . However, their functional forms are very sophisticated in
this case and, therefore, we consider several series to check the interesting relationship
between eigenvalues.
L,N
3.3 Contribution of eigenvalues in TH
P
P
L,N
is the characteristic of the best r-dimensional approxi/ L
The ratio rj=1 L,N
j=1 j
j
mation of the lagged vectors in the SSA technique. Furthermore, this ratio is an obvious
criterion for choosing the proper values of the parameters r and L in the SSA. Therefore,
the study for changing this ratio with respect to LPand r is important for the SSA techL,N
nique. First of all, note that, if let CjL,N = L,N
/ L
, then it is easy to see that,
j=1 j
j
for some j {1, . . . , L m} and all values of m belonging to {1, . . . , L 1}, we have

CjL,N CjLm,N .

(15)

Since inequality given in Equation (15) is satisfied for all values of m belonging to
{1, . . . , L 1}, it appears that C1L,N is decreasing on L {2, . . . , [(N + 1)/2]}. In the
next section, we see examples that show whether such a behavior is true for polynomial
models or not.

Chilean Journal of Statistics

4.

51

Examples and Application

In this section, we discuss some examples related to the theoretical results obtained in
Section 3. Also, we provide an application of these results.
4.1 Examples

160
150
140

Singular value

170

180

Example 4.1 Let ht = exp(0 + 1 t), for t = 1 . . . , N . It is easy to see that the corresponding Hankel matrix H has rank one. Figure 1 shows first singular value of H for this
model with 0 = 0.1, 1 = 0.2 and N = 20, which is convex with respect to L and attains
maximum value at L = 10, 11, i.e., the median of {1, . . . , 20}.

10

15

Figure 1. Plot of the first singular value of H for different values of L: example 4.1.

Now, we consider two different examples where their corresponding Hankel matrices have
rank two. The first one is a simple linear model and the second is a cosine model. As we
see for both models, roughly speaking, we can say that the results are somewhat similar
to Example 4.1.

12
6

160

10

Singular value

200
180

Singular value

220

14

240

16

Example 4.2 Let ht = 0 + 1 t, for t = 1, . . . , N . It is easy to show that rank of the

corresponding Hankel matrix H is two. Figure 2 shows the first and second singular values
of H for 0 = 1, 1 = 2 and N = 20. From this figure, we can say that both first and
second singular values of H increase for L [(N + 1)/2] and then decrease.

10

15

10

15

Figure 2. Plots of the first (left) and second (right) singular values of H for different values of L: example 4.2.

52

R. Mahmoudvand and M. Zokaei

15

Singular value

10

10

20
15

Singular value

20

25

25

Example 4.3 Let ht = cos (t/12), for t = 1, . . . , N . First and second singular values of
H are depicted in Figure 3 for series length 100. If we connive some small fluctuations in
the plots, we can say that behavior of singular values of H is similar to Example 4.2.

20

40

60

80

100

20

40

60

80

100

Figure 3. Plots of the first (left) and second (right) singular values of H for different values of L: example 4.3.

15
10

Singular value

250
200

Singularvalue

150

4500

50

3500

100

4000

Singular value

300

5000

350

Example 4.4 Let ht = 0 + 1 t + 2 t2 , for t = 1, . . . , N . It is easy to show that rank

of the corresponding Hankel matrix H is 3. Figure 4 shows the singular values of H for
0 = 1, 1 = 2, 2 = 3 and N = 20. From this figure, we note that all the singular values
of H increase for L [(N + 1)/2] and then decrease, which coincides with Theorem 3.2.

10
L

15

10
L

15

10

15

Figure 4. Plots of the three largest singular values of H for different values of L: example 4.4.

Example 4.5 Let ht = log(t), for t = 1, . . . , N . Then, it can be seen that rank of the
corresponding Hankel matrix H is four. Singular values of H are shown in Figure 5 for
N = 20. The results of this example are in concordance with Example 4.4.
Figure 6 shows two singular values for models ht = cos(t/12) (left) and ht = log(t)
(right), for N = 5, . . . , 100. Solid and dashed lines in Figure 6 denote the singular values
for L = 2 and L = 1, respectively. Both of these values confirm our expectation for
discrepancy between two singular values. Notice that the cosine model is not monotone,
but 2,N
1,N
1
1 .

53

1.6
1.2

1.4

Singular value

20
18
14

1.0

16

Singular value

1.8

22

2.0

24

Chilean Journal of Statistics

10

15

10

15

0.007
0.006
0.003

0.06

0.004

0.005

Singular value

0.10
0.08

Singular value

0.12

0.008

10

12

14

16

10

12

14

16

50

Figure 5. Plots of the four largest singular values of H with respect to different values of L: example 4.5.

30
20

Singular value

40

L=1
L=2

10

Singular value

L=1
L=2

20

40

60

Length of Series

80

20

40

60

80

Length of Series

Figure 6. Plots of the first singular value for values of L and N in cosine (left) and logarithm (right) models.

Example 4.6 Let ht = 0 + 1 t + 2 t2 , for t = 1, . . . , N (a polynomial model). Figure 7

shows the ratio CjL,N for 0 = 1, 1 = 2, 2 = 3, N = 20 and j = 1, 2, 3. From this figure,
we note that C1L,N decreases for the values of L less than [(N + 1)/2] and then increases
on the set L {[(N + 1)/2] + 1, . . . , N 1}. Whereas C2L,N and C3L,N increase on the set
{1, . . . , [(N + 1)/2]} and decrease on {[(N + 1)/2] + 1, . . . , N }.

R. Mahmoudvand and M. Zokaei

10

15

1.2e05

10

0.0e+00

0.995

0.001

4.0e06

8.0e06

Ratio

0.003

Ratio

0.002

0.997
0.996

Ratio

0.998

0.004

0.999

0.005

54

15

10

15

Figure 7. Plots of CjL,N with respect to L for N = 20 and j = 1 (left), j = 2 (center), j = 3 (right): example 4.6.

Next, we examine the cases where the degree of the polynomial is greater than two.
Furthermore, different coefficients are considered. The results are similar to Example 4.6
and thus we do not report them here. As a general result, we can say that inequality given
in Equation (15) is satisfied for j = 1 in the polynomial models. Now, we consider the
P
L,N
ratio C1:r
= rj=1 CjL,N . Since C1L,N is bigger than CjL,N , for j > 1, and the discrepancy
between them usually is so much (see the polynomial model of Example 4.6), we expect
L,N
that the ratio C1:r
has a behavior such as C1L,N . In the following example, the behavior
of this ratio is depicted for the polynomial model with degree four.

10

15

0.999980

0.9975

0.999985

Ratio

Ratio

0.999990

0.9985
0.9980

Ratio

0.9990

0.999995

0.9995

1.000000

Example 4.7 Let ht = 0 + 1 t + 2 t2 + 3 t3 + 4 t4 , for t = 1, . . . , N . Figure 8 shows

L,N
the ratio C1:r
for 0 = 1, 1 = 2, 2 = 3, 3 = 4, 4 = 5 and N = 20. From this figure,
L,N
we note that C1:r decreases on L {2, . . . , [(N + 1)/2]}, for r 1, and then increases on
L > [(N + 1)/2], as expected.

10

15

10

15

L,N
Figure 8. Plot of C1:r
with respect to L for N = 20 and r = 1 (left), r = 2 (center), r = 3 (right): example 4.7.

Chilean Journal of Statistics

55

4.2 Choosing the SSA parameters

Several rules have been proposed in the literature for choosing the SSA parameters; see,
e.g., Golyandina et al. (2001) and Hassani et al. (2011). However, the list is by no means
exhaustive. Certainly, the choice of parameters depends on the data collected and on the
analysis we have performed. Anyway one important note is that singular values give most
effective information for choosing parameters in the SSA. In previous subsections, several
criteria and theorems were considered to investigate the behavior of singular values of the
Hankel matrix. Considering theoretical results about the structure of the Hankel matrix,
trajectory matrix and relationship with their dimensions, enable us to state that the choice
of L close to one-half of the time series length is a suitable choice for decomposition stage in
most cases. The previous empirical and theoretical results also confirm the results obtained
L,N
, it can be seen that
by us here. However, by using definition of the criteria TH
L,N
L1,N
=
TH
TH

K
X

h2j .

(16)

j=L

Equation (16) is the rate of change in tr(HH> ) for each unit change in the window length.
This rate is large for small values of the window length and decreases to attain minimum
value at L = K, where it is equivalent to L = Lmax = median{1, . . . , N }. This motivate
us to choose smaller values than Lmax when the rate given in Equation (16) is small. To
support this motivation, Golyandina et al. (2001) said that series with a complex structure
and too large window length L can produce an undesirable decomposition of the series
components of interest, which may lead, in particular, to their mixing with other series
components. Sometimes, in these circumstances, even a small variation in the value of L
can reduce mixing and lead to a better separation of the components, i.e., it provides a
transition from weak to strong separability.
Another important parameter to be chosen is the number of needed singular values r for
grouping in the reconstruction step. Election of this parameter is similar to the procedure
for obtaining the cutoff value in principal component analysis. It is known that there is
not a general way to choose an optimal value of the cutoff number and it depends on the
data; for a complete description and review of this topic, see Jollife (2002). Perhaps the
most obvious criterion for choosing the cutoff value is to select a (cumulative) percentage
of the total variation, which one desires that the selected singular values contribute, say
a 80% or 90%. The required number of singular values is then the smallest value of r for
L,N
which this chosen percentage is exceeded. This criterion is equivalent to the ratio C1:r
previously defined.
5.

Conclusions

We have considered one of the main and most important issues in the singular spectrum
analysis, that is, the selection of parameters. As stated, singular values of the trajectory
matrix in the singular spectrum analysis play an important role. Specifically, election of the
parameters values of the window length (L) and the number of needed singular values for
reconstruction of series (r) depend on the behavior of the singular values of the trajectory
matrix. In this paper, we have studied the behavior of the singular values of a Hankel
matrix (H) with respect to its dimension. We have shown that, for a wide classes of time
series, the singular value of HH> (L,N
) increases with L in L {1, . . . , [(N + 1)/2]} and
j
decreases in L {[(N + 1)/2] + 1, . . . , N }. In addition, we have investigated the behavior
of the sum of square and the contribution of each singular value. The results based on
these criteria have shown that the choice of L close to one-half of the time series length is
a suitable choice for decomposition stage in most cases for the singular spectrum analysis.

56

R. Mahmoudvand and M. Zokaei

Acknowledgements
Authors would like to thank anonymous referees for their valuable comments that improved
the exposition of the paper.
References
Bhatia, R., 1997. Matrix Analysis. Springer, New York.
Elsner, J.B., Tsonis, A.A., 1996. Singular Spectrum Analysis: A New Tool in Time Series
Analysis. Plenum Press, New York.
Golyandina, N., Nekrutkin, V., Zhigljavsky, A., 2001. Analysis of Time Series Structure:
SSA and Related Techniques. Chapman & Hall/CRC, New York.
Hassani, H., 2007. Singular spectrum analysis: methodology and comparision. Journal of
Data Science, 5, 239257.
Hassani, H., Heravi, S., Zhigljavsky, A., 2009. Forecasting European industrial production
with singular spectrum analysis. International Journal of Forecasting, 25, 103118.
Hassani, H., Mahmoudvand, R., Zokaei, M., 2011. Separability and window length in singular spectrum analysis. Comptes Rendus Mathematique (in press).
Hassani, H., Thomakos, D., 2010. A Review on singular spectrum analysis for economic
and financial time series. Statistics and its Interface, 3, 377397.
Horn, R.A., Johnson, C.R., 1985. Matrix Analysis. Cambridge University Press, Cambridge.
Jolliffe, I.T., 2002. Principal Component Analysis. Springer, New York.
Peller, V., 2003. Hankel Operators and Their Applications. Springer, New York.
Widom, H., 1966. Hankel matrix. Transactions of the American Mathematical Society,
121, 135.