B ank i Kredy t 47(3), 2016, 251-266

Examination of the directions of spillover effects

between the real estate and stock prices in Poland
using wavelet analysis

Marcin Kołtuniak*

Submitted: 14 March 2016. Accepted: 4 May 2016.

Abstract
The main purpose of this article is to discuss the existence of the spillover effects between the direct
real estate, indirect real estate (real estate company stocks) and stock market price indices in Poland and
to verify whether the directions of these relationships are stable in the time and frequency domains.
For this purpose we have used wavelet analysis methods, such as wavelet coherency and wavelet
phase difference analyses, which allow to check the indicated stability and detecting dates of any
potential structural changes. The results have not confirmed the hypothesis that the directions of these
relationships were constant in the examined period of 2004 Q4 – 2014 Q4, which indicates the existence
of structural changes. This means that diffusion scenarios of crises opposite to the 2007−2008 global
crisis scenarios, when local property pricing collapses resulted in the global financial crisis, would also
be possible. Consequently, the indicated results disprove the usefulness of the knowledge on the current
directions of these effects in terms of projecting and/or monitoring the long-term macroprudential or
long-term housing policies in Poland, and also give some new insights into investment policy issues.

Keywords: real estate market, real estate company stocks, spillover effects

JEL: E30, G10, G14, R30

* Warsaw School of Economics, Collegium of Business Administration; e-mail: mk48691@doktorant.sgh.waw.pl.

252 M. Koł t un iak

1. Introduction

The starting point of our analysis is the recent work of Miller et al. (2015) on the time stability of the
spillover effect direction between the housing market and the stock market in the USA in the period
1890−2012. The authors have found that on the US markets the credit-price effect, when the changes
in real estate returns cause changes in stock market returns, was recurrently replaced by the opposite
wealth effect (Kapopoulos, Siokis 2005), which was potentially caused by some structural changes
on the above-mentioned markets. These findings have been based on the wavelet analysis methods,
derived from the physics methodology, such as the wavelet coherency and the wavelet phase-difference.
Such methods are contemporarily used also by economists in order to simultaneously examine changes
in terms of co-movement strength, causality sign and causality direction between some economic
phenomena in the time and frequency domains (Mandler, Scharnagl 2014; Scharnagl 2011; Soares,
Aguiar-Conraria 2011a–c; 2014).
The relations between real estate and stock markets prices were previously examined mainly using
traditional methodology of econometrics, such as Granger causality tests (Miller et al. 2015), despite the
fact that this method does not provide an opportunity to check the time stability of the direction under
consideration and also requires an examination of only one investment horizon at once. Consequently,
such methods are strictly dominated by the wavelet analysis methods.
We have examined three research hypotheses:
1. There exists a stable direction of the domestic spillover effect between the levels of the rates
of return on the price indices of the direct real estate market and stock market in Poland, indicating
the stable existence of the wealth effect or the credit-price effect, when the pricing change impulses
are constantly originated on the stock market or on the real estate market, respectively.
2. There exists a stable direction of the domestic spillover effect between the levels of the rates of
return on the price indices of the direct real estate and indirect real estate (real estate company stocks)
markets in Poland.
3. There exists a stable direction of the domestic spillover effect between the levels of the rates of
return on the price indices of the indirect real estate (real estate company stocks) and stock markets
in Poland.
Knowledge on the stable direction of the first effect would be useful in terms of conducting
macroprudential policy and long-term housing policy in Poland. In particular, in the scope of
macroprudential policy it would allow to predict the consequences of potential economic shocks, such
as economic crises, due to the presence of the potential constant sequence of property markets and
financial markets price diffusion, including their impact on economic growth and financial system
stability (Łaszek 2006; Główka 2012). Knowledge on the stable direction of the second and third effects
would be potentially useful in terms of conducting investment policy, especially in the light of the
ongoing process of the development of the real estate company stocks (indirect property) market as
a separate asset class in Poland.
Such research has not yet been conducted using the same methodology in the case of national
economies other than the US in the scope of the first of the indicated spillover effects and has not
yet been performed at all in the scope of the last two effects. Furthermore, the article contributes
to the literature by delivering results of the examination of the time stability of the domination of
informational efficiency between the major domestic investment markets in Poland.
 Examination of the directions of spillover effects... 253

2. Literature review

The results of the previous studies on the long-term relationships between the indices under
consideration have referred almost solely to some developed economies. These studies have particularly
aimed to examine the direction of the causal relationships between the rates of return on price indices
of real estate and stock markets, especially studies on the existence of wealth or credit-price effects.
Literature on causality directions between the markets under consideration referred especially to the
results obtained using Granger causality tests, vector autoregressive models, vector error correction
models and threshold error correction models, delivering often mutually exclusive results (Myer, Webb
1993; Okunev, Wilson, Zurbruegg 2000; Lean 2012; Tsai, Lee, Chiang 2012; Hoesli, Oikarinen, Serrano
2015). Such relationships have not yet been examined in the case of Poland, due to the informational
inefficiency of the domestic property market, which results in the lack of commercial property indices.
On the basis of the literature review, it should be noted that the existence of such discrepancies
could be explained by the differences in frequencies of measurement, selection and methodology of
used market price indices, taking into account scopes in time and place dimensions and qualitative
time changes in terms of the composition of market portfolios. Such discrepancies could be triggered
by the relatively constant occurrence of the flows of capital into the specialized real estate investment
funds – REIT markets, which have caused their split as a separate asset class, as well as by the existence
of some structural changes between the investment markets under consideration.

3. Methodology

The traditional econometric models, as well as traditional spectral Fourier analysis, do not provide tools
to analyse economic phenomena in both the time and frequency domains simultaneously.
The econometric methods, including Granger causality direction tests, require the creation of
a time constant model (except for in the case of the estimation of the time-varying parameter VAR
models) in the form of one or more equations, using at once only one data time interval. In turn, the
Fourier transform does not require the creation of a model and provides an opportunity to analyse
simultaneously different data intervals. However, its use results in the loss of the time domain
information, which makes identification of the dates of transient changes impossible. Therefore, these
methods are not adequate to analyse data in terms of recurring structural changes, which may cause
temporal reversals of causality directions between economic phenomena. Furthermore, traditional
Fourier analysis refers only to stationary time series, whereas the price indices analysed in this article,
as well as a large number of macroeconomic time series, exhibit non-stationarity. As a result, in this
article we aim to analyse the stability of the relationships under consideration in the time and frequency
domains using the wavelet analysis tools, which are commonly used, inter alia, in signal processing and
geophysics applications. The toolbox contributed by Aguiar-Conraria and Soares – ASToolbox (Soares,
Aguiar-Conraria 2014) has been used to perform simultaneous analysis of the analysed time series in
both the time and the frequency domains.
The further part of the “Methodology” section describes the continuous wavelet transform (CWT),
which enables to extract self-similarities of data sets, and is based on Miller et al. (2015), Soares,
254 M. Koł t un iak

Aguiar-Conraria, Azevedo (2008), Soares, Aguiar-Conraria (2011b; 2014), Grinsted, Moore, Jevrejeva
(2004) and Reboredo and Rivera-Castro (2012). The discrete wavelet transforms, such as the maximum
overlap discrete wavelet transform that enable to perform noise reduction and to compress data sets
under consideration, are another approach to solving the problem, but require more computational
time than the CWT.
Contrary to the infinite duration sine and cosine wave functions used in the Fourier transform,
a periodic mother wavelet function is localized in the time and frequency domains finite duration wave
Ψ(t ) ∈ L2 (R )
function Ψ (t ) ∈ L 2
(R ) where Ψt(tdenotes
) L (Rtime,
∈ 2
) which Ψ(t ) ∈satisfies L (R ) the so called admissibility condition:
2
Ψ(t ) ∈ L2 (R )
Ψ(t ) ∈ Ψ
ΨL(t )(2R
2(t ) ∈ L
∈)L (R 2
2
(R) ) ∞ Ψ(t ) ∈ L2 (R )
∞ ∞ Ψ(t ) ∈ ∞L (R ) ∫ Ψ (t ) dt = 0
∫ Ψ∫∞(Ψ (t ) dt = 0 –∫∞ Ψ (t ) dt = ∞0
t ) dt = 0 ∫
∞ Ψ ( t ) dt = 0
–∞
∞
(1)
∞ ( t )∫dtΨ
–∞
Ψ∫Ψ ∫ Ψ (t ) dt = 0
∞– ∞ 2
–∞
(t ) ∈
∫ Ψ (=t()0t dt
L ( R )) dt==00
– ∞∫ Ψ (–t∞ Ψ (t ) – ∞
– ∞ ) dt = 0
Ψ ( t ) Ψ(t) transformed
Function Ψ(t) is subsequently –∞
∞ Ψ
using (t) a translation (or shift) (τ) and a scaling (or dilating)
Ψ∫ Ψ t t Ψ dt ( t) 0 location and a proper Ψ1 (t)width, t – τ respectively:
(s) constant parameters
Ψ (t ) =
1 int order
Ψ1
– τ to obtain
(())
tΨ – (τt )
(
1 – ∞Ψ(tt) – τ
)
Ψ Ψ (t ) =
Ψ(t)= time
a proper
( )
( ) 1 t τ
Ψ (t ) = ( ) Ψ
s

)((() )))
– s
Ψ(t) =s Ψs = Ψ t –τ 1 t –τ
1 τ
s s sΨ(t) =Ψ(t)Ψ
Ψ Ψ(t)1== 1 tssΨ
Ψ(t) = s Ψs s
s 1
(( t –
–Ψτ s
t – τs
s s 1

Ψ (t ) = ( ) s
Ψ
s (2)
s s >
s >1 1 t –τ
s >1
Larger windows are obtained when s > 1and
s <1
s >1 Ψ(t) = s > 1Ψ
s s>>s11
(
narrower )
s when s < 1. s > 1
In this study s <it1 has been assumed s < 1 that the s 1
> wavelet s < 1 function is given by the Morlet function, which
is often used in economic applications, and s <the 1 sangular
s < <11 Fourier frequency ωs10 < has1 been
t2
set to 6 in order
ss > < 1 1 – –
to facilitate the frequency – – period reversion analysis (Soares, Aguiar-Conraria Ψ(t) = π e 2014):
2
1 t iω0t
– 2 4
e 2
= π 4 e– 0 eiω t –
2 t
Ψ(t)
i1ω t 1 t2 1 t2
Ψ(t) = π 4 e 0 eΨ2 (t) = π 4 esiω0t e1 2 – (1t) = π––1t42 e iω0t e2–t 22
– –
1 t2
< Ψ 4 iω0t 1– 2 iω t t– (t) = π e e 2
– –

Ψ(t) =Ψ π (t)1e= π– e 4 eitω2 t 0 –e2 2 Ψ

4 iω0t
(3)
Ψ =4 iω0t 2 π Ψ t L R)
2
( t ) e
4 – 0
e ( ) ∈ (
Ψ(t) ∈ L∈(R2)
–
Ψ π
2

Ψ(t) L (R ) Ψ(t) ∈ L2 (R ) ( t ) = e e
Ψ 1(t ) ∈ L t(2 R )
2
The continuous wavelet transform of Ψ a time
(t) ∈ series L–2(4(tR denoted
i)ω0t 2 2
– by x(t), using Ψ(t) ∈ L2 (R ) is delivered as:
Ψ(t) =Ψπ(t)2 ∈ Ψ e L e(R )
) ∈ L ( R ) 1 ∞ t–τ
( )
∫ x(t )Ψ s dt, τ, s ∈ R s ≠ 0
2
1 ∞ ∞ t–τ Ψ W (τ , s ) =
W (τ , s ) = 1∫ x(t )Ψ
W (τ , s ) =s – ∞ ∫Wx((tτ)Ψ ( () )
t –dt
, ss) = s dt∫, xτW
( t ) ∈
τ1, τ ,∞s ∈ R st ≠– τ0
(,ts)(2Ψ
L ( R )
∈ R ∞s ≠1 0, ∞ ∈ t – τ≠ s ∞ 1 ∞
( ) ( ) (( ))
(τ(–t,∞)s ) =L τ(,Rs )) ∫=s x1(1t )Ψ∞–∫∞ x(t )Ψdt
dt τt ,–s τ R s dt 0
–

0 t –(4) τ
s –∞ WΨ s ∈
1
W (τ , s )s1=– ∞ ∞ ∫–∫∞x(t s)tΨ
τ Ψ t t–, –ττsτ,Ws ∈(τR,, sτ) ,=ss ≠∈0R ∫ xs(≠
τ ( )≠
t )Ψ dt, τ , s ∈ R

( () )
W ( , s ) = s ∞ x ( t ) dt
– τ s dt, τ , s ∈ Rs s∞ ≠ 0 , , s ∈ R – s 0 s
W (τ , s ) = ∫ sxs(–t∞)Ψ WPS svariance
s(τ,, sτ), s= ∈WPS
dt R s ≠ 02
τ, s )
(series
In turn, the(τwavelet
, s ) = WPS power (τ, spectrum
s ) 2 (WPS), describing
2
s –∞
localized of a time is defined as:
WPS (τ, s ) = WPS (τ, s ) W (τ , s ) = 1 2 ∫ x(t )Ψ t – τ dt
WPS Ψ
( )
∞ 2
(t ) ∈ L (R )
2 , τ, s ∈ R s ≠ 0
WPS (τ, s ) = WPSWPS
WPS (τ, s ) = s(WPS
(τ, s ) τ,– s ) (=τ,WPS s ) (sτ, s )2 WPS (τ, s ) = WPS (τ, s )
2 2
∞
WPS(τ(,τ,s s) ) =WPS WPS(τ(2,τ,s s) )2 ∞ – 1 J (W (τ, s ))
τ, s ) ϕx (τ,∫sΨ
WPS =
– 1 J (Wx (τ, s ))
τ WPS ( τ , s ) = WPS ( ) =(ttan ) dt = 0 x (5)
ϕ (τ , s ) = tan J ( W ( , s ) ) ∞ R(Wx (τ, s ))
– 1 J (Wx (τ, s ))
ϕx (τ, s ) = tanR1(W (τ,x s ))
x – –

(τx,(sτ), s=))tan J ( W ( τ , Js2()W

) ( τ , s )) J (W (τ, s ))
R
ϕ(xW
x
ϕx (τ, (sτ)ϕ
WPS ,=sx ()tan τ(=τ, ,sWPS
– 1
) )=) tan (τ–x, s )J (W (τ, s )) ϕx (τ, s ) = tan
– 1 x – 1 x

The wavelet phase angle is obtained (in case R ( W s

xof a complex valued) (τ,(τs,)Ψ s) )) functions) as: R(Wx (τ, s ))
t
(Wxτ(,Rτs,(W
– 1x (J (RW)(W ( )
ϕϕ(x (τ, s, s) ) =tan
Rtan(JW 1 x

ϕx (τ, sx )τ= tan=–1

sx ))x(τ, s ))
R(W (τ,xys ()τ) , s ) = Wx (τ, s ) Wy (τ, s )
x
Wxy (τ, s ) = Wx (τ, s ) Wy (τ, s ) R(Wx (τ, sx )W ) 1 t –τ

Wxy (τ, s ) = W
τ
Wxy (τ, s ) = Wx (τ, s ) Wy ( , sϕ) (τ, s ) = tan
Wxyx ((ττ,, ssW
x
))xy=WW
– 1 J (W
(yτx(,τs(,τ)sR,=)s(W
x (τ, s ))
(τ, s )
Ψ (t ) = ( ) Ψ
W (τs, s ) = Wxs (τ, s ) Wy (τ(6)
x (y(ττ,,ss)) Wy (τ, s ) xy
)WW , s)
τ , s) ==WWx(τ, ,s s) WW τ W ( τ τ
, s ) W ( τ , s )
Wx (τ, s ) Wy (τ, s) WW (
xy(τ , s ) ( ) xy
(τ, ,s s)
( ) y

where J andW (τ, s )to the

Rx refer y (τ, s)
Wimaginary andWxythe (τreal ,xys ) =parts, x
s ) Wy (τ, ys )
Wx (τ,respectively.
s >W1 (τ, s ) Wy (τ, s)
Wx (τ, s ) Wx W (τy, (sτW ), xs)(τ, W s )y (τ, s)Wy (τ, s) x
Wxy (τW , sx )(=τ,W (τ , s )W W τ(,τ ,
s s
XWP ) τ , s W τ , s
W (τ, s s)x ) ( ) ( ) = ( )
XWPxy (τ, s) = Wxy (τ, s) W y(τy , s) xy
s <1 xy
XWPxy (τ, s) = Wxy (τ, s) Wx (τ, xs ) Wy (τ, ys)
XWPxy (τ, sXWP )= W xyxy(XWP τ(τ, ,s)s)= (W τ,xys()τ=, s)Wxy (τ, s) XWPxy (τ, s) = Wxy (τ, s) 2
Wx 2(τ, 2sXWP τ S (st 2 1Wxy (τ, s))
–
) xy(W τ , ( , s )
y s) = W (τ, 2s)
τ
–1
S s W , s xy R ) (τ, s ) =
xy (τ, s) = Wxy (τ, sxy
1
( ( ) )
S (s xy1Wxy (τXWP , s))XWP xy
– – –
R xy2 (τ2, s ) = xy (τ,2 s ) = Wxy (τ–,1s2) Ψ(2 t) =Sπ(s4–12eW0 (eτ,2s ) 2 ) S (s ––11 W (τ, s ) 2 )2
iω t
R xy (τ, s ) = –1 (τS, (ss)) Wxy (τ, –s1)) S (s Wxyy (τ, s))
S (s W–x1 (τ2, s ) ) S2(s 2W–1y (τ, Ss )(s) W
2 –1 –1

(xys(τ,Ws )y(τ(=2τ,,ss)) )xyW (τ, s) S (s–1 Wxy R(τxy2, (sτ),2)2s2 ) =

2 x
S (s RWx ((ττ,, ss )) =) R SXWP
xyR xy (τ, = s–2)1 = xy – 2 S (s–1 –1W (τ, s))2
xy –1 2 –1 2
2
 s ∞
Ψ(ts) Ψ
= ( ) ( )( )
= (πt )4=se 0 e Ψ s 1
–
iω t s2 – ∞
– t –τ s
s 1 ∞– iω t –t t2– τ
2

1 Ψ(tWPS
2
WPS (τ, s ) = WPS (sτ, s>)Examination
W (τ , s ) Ψ
) ∈ L (of 2
= (t) = ∫π x(te)Ψ e
(Rs)) =directions
s –∞ 2
4 0
( )
s
dt, τ , s ∈ R s ≠ 0

s > 1τ, the WPS (τ,ofs )spillover effects... 255

Ψ(t) ∈ L (R )
2

s <1 t –(τ, s )
2
1 1 (τ,∞sx)(t=)Ψ
ϕx (τ, s ) = tan
In order
– 1 J (Wx (τ, W

to analyseRthe relationship
<WPS
s ))(τ , ss) =
ϕx (τbetween
, s )s= –tan
WPS
( )
∫ –1 J (Wx s(τ, s ))∞ τ, s ∈ R s ≠ 0
∞ two time1series x(t) and
dt ,
τ in both the time and
t – y(t)
(Wx (τ, s ))1
frequency domains, we introduce a4 cross
Ψ = ( t ) π
–
e e
–
t2
iω0t wavelet
2 –1
W (τ ,Rs ()W
transform
–
t 2
=x (τ, s )∫) x(t )Ψ
J (W
(XWT)
s (τ
–∞
, s )
as:
)
( )
s
dt, τ , s ∈ R s ≠ 0
Ψ( t ϕ ) =x (τπ, s4)e=iωtan
0t
e 21 2
– x

WPS (τ, s ) = WPS (τ, s ) R ( W x (τ, s

))
Wxy (τ, s ) = Wx (τ, s ) Wy (τ, s2 ) W (τ, s ) = W (τ, s ) W (τ, s )
(7)
Ψ(t) ∈ L (R )xy 2 WPS x
τ , s y
WPS τ , s
2
Ψ ( t ) ∈ L ( R ) ( ) = ( )

where Wx (τ, s ) and Wy (τ, s) areϕCWTs (–τ1 , sJ)(WW x ( ,(τ τ

s,))s ) W (τ, s )
x (τ1W =τ∞W
, sx)(of an
s ) x(t)tfunction
,tan
W
–=y∞τ(xτ, s) and ya y(t) function, respectively.
( ) ( )
xy

Ws (–τ∫∞, s ) = ϕx (sτ∫ x,xs()t =)Ψtan–1 dt,x τ , s ∈ R s ≠ 0

W (τ , s ) = x(t )Ψ1R(W (τdt , s, ))τ ,ts–∈τJR(W s(τ≠, s0))
The cross wavelet power (XWP), describing s –∞ s R(Wxbetween (τ, s )) two time series x(t) and
XWPxy (τ, s) = Wxy (τ, s) Wx (τ, sa)localized Wy (τcovariance
, s)
y(t), is introduced as: XWPxy (τ, s) = Wxy (τ, s)
Wxy (τ, s ) = Wx (τ2, s ) Wy (τ, s )
WPS (τ, s ) = WPS 2 (τ, s ) 2
S (s
–1
W xy ( τWPS
, s)) XWP(τ, s ) (=τW
xy ,WPS sτ(,)xyss=–()1τW
sxy) (=τ,(SW ,Wxs)((ττ, ,ss)))W2y (τ, s ) (8)
R xy2 (τ, s ) = R 2 (τ, s ) = 2
xy

S (s 1 Wx (τ, s )W)xS((τs,xys1 )WJy ((W

τ, WsS)y (()τ,–1sW
2
– 2 – ) 2
(τ, s ) )–1S (s –1 Wy (τ,2 s ) )
2
x (τ, s )) x
x (τ, s ) = tan
The squared wavelet ϕcoherency
–1
gives Wx–1(τJ, s(on
W (τS,(sW
) x time yW
))and s()τ, s))
(τxy, frequency
ϕx (Rτ,R2information
s()Wτ τ
=,xs(tan
) ,
= s )) localized comovements
(correlations) between a pair of analysed time series and1 is
xy R (W ( τ, s ))
x defined as: 1
2
x (τ, s ) ) S (s Wy (τ, s ) )
2
XWPxy (τ, s) = Wxy (Sτ(, ss) W
– –

0 ≤ R xy (τ, s) ≤ 1 0 ≤ R xy (τ, sXWP ) ≤ 1 (τ, s) W (τ, s)

xy = xy
Wxy (τ, s ) = Wx (τ, s ) Wy (τ, ss) –1W (τ, s)) 2
1 J (Wxy ( τ, s2)) Wxy (τ , s ) = WxS((τ ,(s1)xyxyW(yτ(, τs,))s )
ϕx (τ, s ) = tan
R xy (τϕ ∈(τ–,1 sπ)J,≤W
, sx ()ϕτ=x0,(sτ≤), =sR)xytan π – τ,2 s) ∈ – π2, π
(9)
R (Wxy (τ, s )) S (s 1RW2 x((ττR, (sW
–
)
2
) (Sτ,(s s–1
)) W S((ϕτs,x s1(W
) (τ, s))
xy)
W (τ, s) xy , s ) =J (W –(1τ, s))
xy y
Wx (τ, s ) 2
τ, s(τ), s ) = W τ x (τ, s )ϕ ) (Sτ(
2
, s Wy–(τπ, ,sπ) )
–1
Wx (yϕ tany ( 1 ) S (
xys W ,ss) ∈
s )ϕ (τ, s) = ϕ (τ, s ) –Rϕ(W(xyτ,(sτ), s ))
x x
where ϕSxyis(aτ,smoothing ) – ϕ y (τ,and:
s) = ϕ x (τ, soperator
xy (τ, s) ≤ x
XWPxy (τ,≤s)R=xy Wxy (τ, s1)
0 y

XWPxy (τ, s)0= ≤WRxy (τ(τ, ,s)s) ≤ 1

ϕ xy (τ, s)J=(W ϕxyx ((ττ,,ss)))– ϕ y (τ (10)
xy
, s)
ϕx (τ, s ) = tan –1 1 2
))–)1 ϕJx(W (τ, s2) ∈ – π, π
S (s RW(W xy (τ(,τs,)s (τ, s))
where 0 indicates low correlation and
R xy (τ, s ) = 2
2 1 indicates high S s
xy correlation.
( W
ϕ2x (τ, s )–1= tan xy 2 sxy))
1 ( τ , ϕ x (τ, s) ∈ – π, π
R –(1τ, s ) =
S (xys Wx (τ, s ) )–1S (s Wy (2τ, sR)(W –)1xy (τ, s )) 2
S (s Wx (τ, s ) ) S (s Wy (τ, s ) )
The wavelet phase difference, ϕ xy (τ, s) = ϕ x (τ, sof
the outline ) –whichϕ y (τ, sis) depicted in the first diagram, provides
information on time changes of causality direction (phase lead of x(t) over y(t)), as well as information
(τ, s) ≤ 1 ϕ xy (τ, s) = ϕ x (τ, s ) – ϕ y (τ, s )
on time changes of the sign 0 ≤ofR xycomovements
0 ≤ R xy (τbetween , s) ≤ 1 two analysed time series and is defined as:
J (Wxy (τ, s))
ϕx (τ, s ) = tan 1 J (Wxyϕ x((ττ,,ss)) ∈ – π, π (11)
ϕx (τR
, s(W (τ, s ))
) =xy tan 1
ϕ x (τ, s) ∈ – π, π
R (Wxy (τ, s ))
or:

ϕ xy (τ, s) = ϕ x (τ, s ) – ϕ y (τ, s )
(12)
ϕ xy (τ, s) = ϕ x (τ, s ) – ϕ y (τ, s )

According to Aguiar-Conraria and Soares phase difference test results are significant only
when wavelet coherency test results are statistically significant in both cases of considered time and
considered frequency band (Soares, Aguiar-Conraria 2014).
256 M. Koł t un iak

4. Data

Our study focuses on the price movements on the three kinds of markets: the direct and indirect
property markets, as well as the stock market in Poland in the period starting on 31 December 2004
and finishing on 31 December 2014.
We consider the three time series of the standardised logarithmic rates of return on the price
indices, previously deflated using the consumer price index estimated by the Polish Central Statistical
Office as of 31 December 2004, encompassing the period 2004 Q4− 2014 Q4. The sequence of the data
transformations has been performed in line with the methodology used by Miller et al. (2015) to allow
for international comparison and to correct for any potential heteroscedasticity. However, the wavelet
analysis also enables to perform analogical research on price indices due to its ability to examine
stationary as well as only locally stationary time series (Bruzda 2013; Miller et al. 2015).
The list of the considered market price indices, which are presented in the Figure 1, is as follows:
1. Price index of real estate companies (the Polish indirect property market price index), retraced
using own estimations in accordance with the Warsaw Stock Exchange indices rules to replicate the
WIG-developers index, which was not estimated for the period before the 2007 Q2.
2. The investment property fair value index (the latent Polish direct property market price index,
covering especially the commercial real estate markets, based on the sector level appraisal data
available since 2005 when the International Accounting Standard no. 40 Investment property came into
force, enabling the use of an appraisal-based valuations approach instead of a historical cost approach),
according to own estimations based on the cyclical revaluations of the aggregated investment portfolios
held from the beginning of successive periods, disclosed in the quarterly financial statements (done in
accordance with the IAS no. 40) of the enterprises included in the WIG-developers portfolio as of the
31 December 2014, mirroring to some extent the stock market price indices estimation rules, including
not taking into account the disposal gains (extensively explained in Kołtuniak 2016).
3. The WIG index (the Polish overall stock market price index), according to the Warsaw Stock
Exchange data.
According to the results published in Kołtuniak (2016), indicated price indices were cointegrated
in the period under consideration, which means that the long-term equilibrium reversion tendencies
between them were present, which legitimatise the check on their causality directions.
Despite the fact that quarterly data have been used due to the lack of higher frequency data
connected with the real estate markets, the wavelet analyses, which are often used to analyse high
frequency data sets, have been performed with the aim of detecting structural changes that potentially
influenced the directions of spillover effects under consideration.

5. Results

Table 1 depicts the results of the augmented Dickey-Fuller tests on the non-stationarity of the
considered time series. These results indicate that all of the considered time series are integrated of
order 1, which legitimatizes the use of the Granger causality tests. The results of the examination
of the considered spillover effects using Granger causality direction tests are depicted in Table 2.
 Examination of the directions of spillover effects... 257

These results indicate that in the period between 2005 and 2014 pricing change impulses on the stock
market in Poland originated on the real estate market.
The existence of the credit-price effect has been confirmed only in the case of the direct property
and indirect property markets causality and has not been confirmed in the case of the direct property
and stock markets causality. The existence of the wealth effect has been denied in the case of both the
direct and indirect property markets.
Figures 2−7 depict the results of the wavelet analyses performed in order to verify the research
hypotheses, using both the squared wavelet coherency and the wavelet phase-difference methods.
The thick black lines in Figures 2, 4 and 6 cut the regions in which exist the edge effects connected
with supplementing short data sets with zeros. The black thin lines in Figures 2, 4 and 6 designate the
5% level of significance, estimated using the Monte Carlo simulations. These lines indicate also time
periods in which phase difference functions presented in Figures 3, 5 and 7 were statistically significant
within the frequency bands, which were previously designated in line with the indicated significance
lines (Soares, Aguiar-Conraria 2014).
In the case of the scales of Figures 2, 4, and 6, the bright and dark colours designate low and high
squared coherency (correlation) between the analysed time series, respectively. The indicated figures
do not allow to distinguish between negative and positive correlations.
Figures 2 and 3 verify the first research hypothesis on the stable existence of the wealth or
the credit-price effects in Poland in the period 2004 Q4 – 2014 Q4. The returns on the direct real
estate and on the stock market price indices did not display significant comovements in the period
before and during the world crises on the real estate and financial markets (2005−2008). In the period
2009−2014, the comovements became significant for the frequency band between half a year and 1.5 years.
Consequently, for this period the phase differences in the frequency bands 0.5−1 and 1−1.5 years have
been checked. Figure 3 shows that in the indicated period, as well as in the entire sample period,
the credit-price effect was replaced by the wealth-effect and vice versa multiple times. Consequently,
the relationship direction is not stable in time. As a result, the first research hypothesis has been
rejected. This result is in line with the findings for the US markets published by Miller et al. (2015).
Figures 4 and 5 verify the second research hypothesis on the stable existence of the sectorial
spillover effect direction (Rua, Nunes 2009) between the logarithmic rates of return on the price
indices of the direct and indirect property markets in Poland in the period 2004 Q4 – 2014 Q4.
In the periods 2004 Q4 − 2007 Q2 and 2008 Q3 − 2013 Q1 indicated rates of return displayed statistically
significant high comovements (correlations) in the frequency band between half and 1.5 years. During
the real estate and the global financial crises these series displayed high correlations in the frequency
band near 2 years. After 2013 Q2, no statistically significant comovements have been displayed.
Consequently, for the period till 2013 Q2, the phase differences in the frequency bands 0.5−1 and
1−2 years have been checked. Figure 5 shows that the changes in the level of the indirect property
market price index returns led the changes in the level of the direct property market index returns,
only in the periods 2004 Q4 – 2007 Q1 and 2012 Q2 – 2014 Q4, which is in line with the hypothesis
on the efficient capital markets. In the vast majority of the entire period under consideration the
changes in the level of the direct property market price index returns led the changes in the level of
the indirect property market price index returns, which indicates the existence of some inefficiencies
in the Polish real estate company stocks market. Consequently, the second research hypothesis has
also been rejected.
258 M. Koł t un iak

Figures 6 and 7 verify the last research hypothesis on the stable existence of the direction of the
spillover effect between the logarithmic rates of return on the price indices of the indirect property
market and entire capital market. In the period 2005 – 2006 Q2 the indicated rates of return displayed
statistically significant high comovements (correlations) in the relatively high frequency band between
0.25 and 0.75 years and in the period 2008−2012 in the frequency band between 0.5 and 2 years, as well
as in the period 2007−2012 in the frequency band 2.5−6 years. After 2013 Q2 no statistically significant
comovements have been displayed. Consequently, for the period till 2013 Q2 the phase differences in
the frequency bands 0.25−1.5 and 2−5 years have been checked. Figure 7 shows that the changes in the
level of the indirect property market price index returns led the changes in the level of the entire stock
market price index returns in the period 2004 Q4 − 2011 Q4 and then the changes in the level of the
entire stock market price index returns led the changes in the level of the real estate company stock
market price index returns. Consequently, the last research hypothesis has also been rejected.

6. Conclusions and policy implications

Rejection of the first research hypothesis on the stability of the direction of the causal relationship
between the real estate and stocks prices in Poland indicates that knowledge on the current direction
of this causality would not be permanently viable, which invalidates its use for making assumptions
in the scope of projecting or monitoring the long-term macroprudential policy and the long-term
housing policy in Poland, as well as in terms of investment policy and in the scope of predicting overall
economic growth rates or directions of the diffusion of crises.
Rejection of the first research hypothesis is in line with the recently published article using
analogical methodology by Miller et al. (2015) in the case of the yearly rates of return on the developed
US housing market and stock markets price indices in 1890−2012.
Contrary to the research by Miller et al., the presented study also covered the comovements and
causalities between the direct and indirect property markets, as well as between the indirect property
market and stock market in Poland.
Rejection of the second research hypothesis indicates that there is no stability in terms of the
informational efficiency domination of the one over the other property markets in Poland, which
means that in the long term there is no possibility to anticipate price changes on these markets by
constantly using one of the real estate price indices as a predictor. Rejection of the third research
hypothesis brings analogical conclusions in terms of the indirect property market and entire stock
market in Poland.
The instabilities under consideration may have been caused by any structural changes on the direct
and indirect property markets, as well as in the shape of the real estate financing system in Poland.
Phase-difference analyses between all of the pairs of time series under consideration have confirmed
the constant presence of positive signs of the comovements between them, which is in line with the
economic theory and influences investment decision processes. Squared wavelet coherency analyses
have revealed that the statistically significant comovements between the indicated pairs proceeded
mainly in the mid and long term. The results of the phase-difference tests have also confirmed that
the direction of any of the considered economic spillover effects would depend on a frequency or on
a period assumed during the estimation of examined rates of return.
 Examination of the directions of spillover effects... 259

References

Bruzda J. (2013), Wavelet analysis in economic appliacations, Wydawnictwo Naukowe Uniwersytetu

Mikołaja Kopernika.
Główka G. (2012), System finansowania nieruchomości mieszkaniowych w Polsce. Doświadczenia
i kierunki zmian, Oficyna Wydawnicza SGH.
Grinsted A., Moore J.C., Jevrejeva S. (2004), Application of the cross wavelet transform and coherence
to geophysical time series, Nonlinear Processes in Geophysics, 11, 561−566.
Hoesli M., Oikarinen E., Serrano C. (2015), Do public real estate returns really lead private returns?,
The Journal of Portfolio Management, 6, 105−117.
Kapopoulos P., Siokis F. (2005), Stock and real estate prices in Greece: wealth versus ‘credit-price’ effect,
Applied Economics Letters, 12, 125−128.
Kołtuniak M. (2016), The long term cointegration between indices of the real estate, the property
enterprises stocks and the capital markets in Poland, Ekonomika i Organizacja Przedsiębiorstwa 3, 17–31.
Lean H. (2012), Wealth effect or credit-price effect?, Procedia Economics and Finance, 1, 259−268.
Łaszek J. (2006), Sektor nieruchomości z perspektywy banku centralnego, Finansowanie Nieruchomości,
6, 17−23.
Mandler M., Scharnagl M. (2014), Money growth, loan growth and consumer price inflation in the euro
area: a wavelet analysis, Deutsche Bundesbank, Discussion Paper, 33.
Miller S.M., Chang T., Li X.L., Balcilar M., Gupta R. (2015), The co-movement and causality between
the U.S. housing and stock markets in the time and frequency domains, International Review
of Economics & Finance, 38, 220−233.
Myer F.C.N., Webb J.R. (1993), Return properties of equity REITs, common stocks, and commercial
real estate: a comparison, The Journal of Real Estate Research, 1, 87−106.
Okunev J., Wilson P., Zurbruegg R. (2000), The causal relationship between real estate and stock
markets, Journal of Real Estate Finance and Economics, 3, 251−261.
Reboredo J.C., Rivera-Castro M.A. (2014), Wavelet-based evidence of the impact of oil prices on stock
returns, International Review of Economics & Finance, 1, 145−176.
Rua A., Nunes L.C. (2009), International comovements of stock market returns: a wavelet analysis,
Journal of Empirical Finance, 4, 632−639.
Scharnagl M. (2011), Wavelet analysis of loans in Germany, Deutsche Bundesbank, http://www.
suomenpankki.fi/en/tutkimus/konferenssit/konferenssit_tyopajat/Documents/FDR2011/FDR2011_
Scharnagl_paper.pdf.
Soares M.J., Aguiar-Conraria L., Azevedo N. (2008), Using wavelets to decompose the time-frequency
effects of monetary policy, Physica A, 12, 2863−2878.
Soares M.J., Aguiar-Conraria L. (2011a), Business cycle synchronization and the euro: a wavelet analysis,
Journal of Macroeconomics, 3, 477−489.
Soares M.J., Aguiar-Conraria L. (2011b), Oil and the macroeconomy. Using wavelets to analyze old
issues, Empirical Economics, 3, 645−655.
Soares M.J., Aguiar-Conraria L. (2011c), The continuous wavelet transform: a primer, Working Paper
Series, NIPE 16/2011, University of Minho.
Soares M.J., Aguiar-Conraria L. (2014), The continuous wavelet transform: moving beyond uni and
bivariate analysis, Journal of Economic Surveys, 2, 344−375.
Tsai I.C., Lee C.F., Chiang M. (2012), The wealth effect in the US housing and stock markets, Journal
of Real Estate Finance and Economics, 4, 1005−1020.
260 M. Koł t un iak

Appendix

Table 1
Results of the augmented Dickey-Fuller tests in the sample period 2004 Q4 – 2014 Q4

First differences of natural logarithms

Time series in levels
of the time series

test with test with test with test with a test with test with
a constant a constant a constant constant a constant a constant
Variable
and and and and
lags lags
a linear a linear a linear a linear
number number
trend and trend and
quadratic quadratic
trends trends
test test test test test test
statistic statistic statistic statistic statistic statistic
(p-value) (p-value) (p-value) (p-value) (p-value) (p-value)

Real estate -1.08* -2.02* -1 .66* 7 -4.50 -3.43 -4.08 8

company (0.72) (0.57) (0.90) (0.00) (0.05) (0.04)
stock index

Investment -2.46* -1.87* -2.99* 1 -5.27 -5.89 -5.31 0

property (0.13) (0.67) (0.30) (0.00) (0.00) (0.00)
fair value
index

WIG index -1.97* -1.88* -1.89* 0 -3.95 -3.93 -3.97 0

(0.30) (0.65) (0.84) (0.00) (0.02) (0.04)

Notes:
The results indicate that all of the considered time series have been integrated of order 1, which legitimatizes the use of the
Granger causality tests. The results were also confirmed by the KPSS test results.
Sign * indicates that the 5% significance level of confirmation of the null hypothesis on non-stationarity of the analysed
times series was exceeded.
Lags number chosen automatically using AIC.
Real estate company stock index and investment property fair value index according to the own estimations; WIG index
according to WSE data.
 Examination of the directions of spillover effects... 261

Table 2
Results of the estimation of the F-statistics of the Granger causality tests

H0 = time series A does not cause time series B

Real estate Investment WIG index
company stock property fair value
index index
0.94 2.27
Real estate company stock index ×
(0.46) (0.09)
Investment property fair value 3.83* × 1.44
index (0.01) (0.25)
0.88 0.44
WIG index ×
(0.49) (0.78)

Notes:
The pairs of the first differences of natural logarithms of the price indices listed in the “Data used” section, in the sample
period 2005 Q1 – 2014 Q4 (assuming 4 lags), using quarterly data intervals.
Time series are presented in the following way: series from the set of time series A in the vertical dimension, series from
a set of time series B (i.e. replicated series A) in the horizontal dimension.
Sign * indicates that the 5% significance level of confirmation of the null hypothesis of no causal relationship between
analysed times series was not exceeded, i.e. the Granger causality relationship between them was not rejected, however its
confirmation requires additional corroboration with the economic theory; p-values in brackets.

Diagram 1
Outline of a wavelet phase difference function

π
– Y leads X
π2

+ X leads Y
0 t

+ Y leads X
–π 2

– X leads Y
–π
262 M. Koł t un iak

Figure 1
Comparison of the fluctuations in the price indices of the direct real estate, indirect real estate and stock
markets in Poland in the period 2004 Q4 – 2014 Q4

6 000

5 000

4 000

3 000

2 000

1 000

0
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Direct real estate property market index*
Indirect real estate property market index** WIG index***

* Appraisal based fair value index of the aggregated portfolio of investment property of enterprises being in the scope
of the research according to the own estimations.
** Price index of the property enterprises stocks being in the scope of the research retraced according to the own estimations.
*** According to the WSE data.
Note: data normalized to the level of 1000 points as of 31 December 2004.

Figure 2
Wavelet squared coherency between the logarithmic rates of return on the price indices of the direct real estate
and the stock markets in Poland in the period 2004 Q4 – 2014 Q4

0.25

0.50

0.75
Period (years)

1
1.25 1
1.50
2 0.5
3
0
4
5
6 -0.5
7
8
9
2006 2007 2008 2009 2010 2011 2012 2013 2014
 Examination of the directions of spillover effects... 263

Figure 3
Wavelet phase-difference between the logarithmic rates of return on the price indices of the direct real estate
and the stock markets in Poland in the period 2004 Q4 – 2014 Q4

From 0.5 to 1 year frequency band

π

π2

–π 2

–π
2006 2007 2008 2009 2010 2011 2012 2013 2014
Phase-difference

From 1 to 1.5 years frequency band

π

π2

–π 2

–π
2006 2007 2008 2009 2010 2011 2012 2013 2014

Figure 4
Wavelet squared coherency between the logarithmic rates of return on the price indices of the direct and
the indirect real estate markets in Poland in the period 2004 Q4 – 2014 Q4

0.25

0.50

0.75
1
Period (years)

1.25
1
1.50
2
0.5
3
4 0
5
6 -0.5
7
8
9
2006 2007 2008 2009 2010 2011 2012 2013 2014
264 M. Koł t un iak

Figure 5
Wavelet phase-difference between the logarithmic rates of return on the price indices of the direct and
the indirect real estate markets in Poland in the period 2004 Q4 – 2014 Q4

From 0.5 to 1 year frequency band

π

π2

–π 2

–π
Phase-difference

2006 2007 2008 2009 2010 2011 2012 2013 2014

From 1 to 2 years frequency band

π

π2

–π 2

–π
2006 2007 2008 2009 2010 2011 2012 2013 2014

Figure 6
Wavelet squared coherency between the logarithmic rates of return on the price indices of the indirect real
estate market and the stock market in Poland

0.25

0.50

0.75
Period (years)

1
1.25 1
1.50
2 0.5

3
0
4
5
6 -0.5
7
8
9
2006 2007 2008 2009 2010 2011 2012 2013 2014
 Examination of the directions of spillover effects... 265

Figure 7
Wavelet phase-difference between the logarithmic rates of return on the price indices of the indirect real estate
and the stock markets in Poland in the period 2004 Q4 – 2014 Q4

From 0.25 to 1.5 year frequency band

π

π 2

–π 2

–π
Phase-difference

2006 2007 2008 2009 2010 2011 2012 2013 2014

From 2 to 5 years frequency band

π

π 2

–π 2

–π
2006 2007 2008 2009 2010 2011 2012 2013 2014