Returns spillovers between tourism ETFs

Returns spillovers between tourism ETFs

North American Journal of Economics and Finance 50 (2019) 100977 Contents lists available at ScienceDirect North American Journal of Economics and F...

5MB Sizes 0 Downloads 2 Views

North American Journal of Economics and Finance 50 (2019) 100977

Contents lists available at ScienceDirect

North American Journal of Economics and Finance journal homepage: www.elsevier.com/locate/najef

Returns spillovers between tourism ETFs Chang Shu-Lien , Lee Yun-Huan ⁎

T

Department of Finance, Ming Chuan University, Taipei, Taiwan, ROC

ARTICLE INFO

ABSTRACT

Keywords: Spillover ETF GARCH Tourism

This paper investigates the relationship between tourism-related markets and equity, consumer staples equities and consumer discretionary equities by means of spillovers and volatility transmission. Relying on the recently introduced Exchange Traded Founds (ETFs), this study is the first to analyze return spillovers derived from an E-GARCH model and to take into account frequency dynamics to understand changes in connectedness across periods of time. The results uncover numerous channels of return transmission across the selected ETF markets over the last 13 years and highlight the role of equity ETFs as the most influential market in the sample. Furthermore, the study provides insights into the characteristics of tourism-related markets using a hidden semi-Markov model. Finally, we prove that tourism-related markets (except casino and gaming) have gained importance as investment assets over the last few years.

1. Introduction Over the past several decades, international tourism has been steadily increasing, as has the importance of the tourism industry for the economies of many countries. The relationship between tourism market development and economic growth has been a focus of study in recent years. A general consensus by previous studies indicated that tourism market development not only increases foreign exchange income but also creates employment opportunities, stimulates the growth of the tourism industry and, by virtue of this, triggers overall economic growth (Gunduz & Hatemi-J, 2005; Kim et al., 2006; Lean & Tang, 2010; Lee & Hung, 2010). As such, tourism development has become an important target for most governments. According to the estimates of the World Tourism Organization (WTO, 2016), international tourism revenues earned by destinations worldwide have surged from US$ 2 billion in 1950 to US$ 1220 billion in 2016. Furthermore, the World Tourism Travel Council (WTTC, 2017) indicates that travel and tourism’s direct contribution to GDP grew by 3.1% in 2016, which was faster than the global economy as a whole, which grew at 2.5%. The direct contribution of travel and tourism to employment grew by 1.8% in 2016, meaning that almost 2 million additional jobs were generated directly by this sector. That result also means almost 1 in 5 new jobs created in 2016 was linked to travel and tourism. When all components of the tourism industry are taken into account, the travel and tourism sector also outperformed several other major global economic sectors in 2016, and the sector is expected to grow at an average of 3.9% per year (the increase in its share of global economic activity across each of GDP, employment, exports and investment) over the next ten years. For these very reasons, thoroughly investigating all aspects of travel and tourism is extremely important for individuals, institutions, and the government. Most research focused on the long-run movements and the casual relationships between economic growth and tourism development in a multivariate model. Recently, however, tourism has received increased attention from investors due to the introduction of new exchange-traded funds (ETFs). Being relatively new asset classes (data for leisure and entertainment and media ETFs are available from June 23, 2005, and casino and gaming ETF are available from January 24, 2008, onward), the problem of



Corresponding author. E-mail addresses: [email protected] (S.-L. Chang), [email protected] (Y.-H. Lee).

https://doi.org/10.1016/j.najef.2019.04.020 Received 8 November 2018; Received in revised form 19 March 2019; Accepted 22 April 2019 Available online 24 April 2019 1062-9408/ © 2019 Elsevier Inc. All rights reserved.

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

interconnectedness between tourism markets is still somehow unexplored in academic literature. Therefore, this paper uses tourism ETFs to fill a gap by investigating the dynamic linkages between leisure and recreation, media and publishing, and global casinos and gaming industries, as well as other significant asset classes, such as total stock market, consumer non-cyclicals, and consumer cyclicals. More precisely, the study tests how total stock market, consumer non-cyclicals, and consumer cyclicals affect the behavior of tourism-related markets and vice versa. There are many reasons to believe that linkages between these target markets should exist. In principle, a well-developed stock market should increase saving and efficiently allocate capital to investments, which leads to an increase in the rate of economic growth (Demirgüç-Kunt and Levine, 1996; Levine & Zervos, 1996). Moreover, wealth creation was one of the dominant themes of economic development, and more than 60 percent of the wealth creation was due to the rising value of household stock holdings (Poterba, 2000). The increase in wealth and its effects on consumer spending are introduced in different studies. Ando and Modigliani (1963) established the aggregate consumption connecting per capita consumer expenditures to per capita labor income, capturing the price of current versus future consumption (i.e., consumption is usually measured as expenditures on nondurables and services). Bhatia (1972) and Peek (1983) included capital gains directly in the consumption function, and more recent work has focused on wealth-based models. Furthermore, the empirical studies have supported that tourism expansion can boost economic development (Balaguer & Cantavella-Jorda, 2002; Dritsakis, 2004; Fayissa, Nsiah, & Tadasse, 2008; Gunduz & Hatemi-J, 2005; Lee & Chang, 2008; Proença & Soukiazis, 2008). Therefore, a high correlation among economic development, stock markets, and tourism development can be expected. However, there are a few issues have still not been delivered by previous research. Reviewing the literature, there are only a few studies examining how individuals and households make use of funds for discretionary expenditures, but those studies have not included tourism expenditures. This neglect of tourism expenditures as related to the other spending is interesting, particularly when one considers that a larger proportion of funds available for spending is due to the increase in wealth in developed economies. Our study augments the existing literature in several ways. First, this paper examines the time-frequency dynamics of connectedness for total stock markets, consumer non-cyclicals, consumer cyclicals, and tourism-related markets (leisure and recreation, media and publishing, and global casinos and gaming markets) ETFs using the new variance decomposition methodology proposed by Barunik and Krehlik (2016). Therefore, we discuss new stylized facts about cyclical properties of transmission mechanisms in the tourism-related markets and examine the time frame of connectedness. Second, we provide valuable insights to investors who are interested in tourism-related markets by running an E-GARCH model in order to quantify the impacts of daily returns of total stock markets, consumer non-cyclicals and consumer cyclicals on the returns of leisure and recreation, media and publishing, and global casinos and gaming markets. Finally, we use the Hidden semi-Markov model (HSMM) to provide novel evidence of the return characteristics of tourism-related ETF markets. The remainder of this paper is organized as follows: Section 2 discusses the relevant academic literature, Sections 3 and 4 shed light on data and the methodology, while Sections 5 and 6 present and discuss the results. A conclusion of the paper is provided in Section 7. 2. Literature review Over the past few years, the tourism industry has garnered increased attention because of not only the high correlation between tourism development and economic development but also new investment vehicles (i.e., exchange-traded funds, ETFs). ETFs are tradeable securities that derive their value from a pre-defined basket of securities comprising constituents of an index. The trade volume of ETFs has recently increased because they provide investors with an inexpensive and efficient method by which to earn profits in a selected market segment. Nowadays, ETFs are proliferating in terms of both their numbers and the market value of total assets. Research examining the relationship between ETFs and equity markets can take several paths. One strand of ETF research investigates the proposition that trading in ETFs transmits volatility to equity markets. These findings are consistent with positive volume-volatility relationships and trading-based explanations of volatility (Krause, Ehsani, & Lien, 2014). Other studies emphasize returns on international ETFs and conclude that ETF returns closely track their respective country’s equity indices (Khorana, Nellis, & Trester, 1998; Tse & Martinez, 2007). The present paper differs from the above cited ones in that we explore not only the price and volatility transmission between tourism ETFs, but also spillovers of returns between tourism ETFs and other ETFs. Linkages between tourism development and economic growth can be found in empirical results related to bidirectional causality between tourism and economic growth (Dritsakis, 2004; Durbarry, 2004; Kim, Chen, & Jang, 2006) and to unidirectional causality with either tourism-led growth (Balaguer & Cantavella-Jorda, 2002; Ghali, 1976; Eugenio-Martı’n & Morales, 2004) or economicdriven tourism growth (Narayan, 2004; Oh, 2005). In general, the good thing is that if the relationship between tourism development and economic growth is bidirectional, both areas can benefit. As such, tourism-generated proceeds have come to represent a significant revenue source, increasing employment, household income, and government income in countries worldwide. Besides, the connectedness between economic growth and financial development has been established by Levine and Zervos (1993), Atje and Jovanovic (1993), Levine and Zervos (1998), Rousseau and Wachtel (2000), and Beck and Levine (2004). The evidence shows that stock-market development is strongly correlated with growth rates of real GDP per capita. More importantly, both stock-market liquidity and banking development can help predict future economic growth. The research has proven that stock markets and banking development provide numerous job opportunities and can boost economic development. Tourism development generates sales, output, earnings, and employment in both the private sector and the public sector. Many studies have presented estimates of the so-called multiplier impact of tourism expenditures and confirmed that tourism expenditures contribute to total final demand and value-added revenue induced from tourism-related industries. The rapid growth of tourism can 2

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

lead to growth in household incomes and government revenue, whether directly or indirectly, by means of multiplier effects, improved balances of payments, private decision making, and public policies. As a result, the development of tourism has usually been considered a positive contribution to economic growth (e.g., Khan, Phang, & Toh, 1995; Lee & Kwon, 1995). Tourism expansion also affects the demand for certain goods and services (Syriopoulos, 1995) specific to tourism insofar as tourism is akin to a non-traded good as opposed to a general-use good. Hazari (1993) pointed out that tourism affects most of the tertiary and nondurable goods consumption sectors. Beyond this fact, tourism markets have been seen as sources of investment assets owing to the distinct differences between the markets, with equity markets having unique characteristics. Chen (2010) showed that improved economic conditions caused by tourism expansion could raise corporate earnings and strengthen the financial performance of tourism-related firms. In other words, the expansion of tourism is expected to promote corporate performance. In addition, Chen, Kim, and Liao (2009) noted that, in Taiwan, the tourism industry has experienced significant growth and that foreign institutional holdings of tourism stocks have grown since the Taiwanese government changed its weekend policy in 2001. The increase in foreign institutional holdings of tourism companies’ stocks reflected not only the change in the weekend policy but also the development of Taiwan’s tourism industry. The literature review summarized above has revealed several gaps linkages among tourism, economic, and financial developments. First the present paper uses daily, weekly, and monthly time-frequency data as opposed to only one category of time-frequency data, which are common in other studies. Second, the previous research has failed to find a significant linkage between equity-market returns and the equity performance of tourism-related firms even though the financial performance of tourism-related firms is assumed to be closely related to both economic development and the expansion of tourism-related industries. The present study argues that equity-market returns (e.g., equity ETF) have a notably direct impact on tourism-related firms’ stock performance (e.g., tourism ETFs). Third, tourism-related industries generally belong to the consumer-cyclicals segment because of their highly cyclical nature. Typically, during periods of economic prosperity, tourism-related industries thrive as consumers indulge in activities and services that afford them some mental satisfaction, and vice versa. Hence, the changes in consumer spending on tourism activities and services may trigger changes in individuals’ and households’ non-discretionary and discretionary expenditures and could confound the link between equity-market returns and tourism-related markets’ returns. 3. Data description and preliminary data analysis Tourism-related ETFs hold baskets of equities for companies providing rest, relaxation, and enjoyment in the sports, leisure, and entertainment industries. Because of their highly cyclical nature, these industries are included in the consumer-cyclicals segment. With significant recovery underway from the subprime crisis in 2008 and the sovereign-debt crisis in 2009 and 2012, 2106 was a landmark year for tourism-related sectors because of governments’ and businesses’ pledges to develop tourism, which will benefit from major investments in virtual-reality and augmented-reality projects. Investors looking to capitalize on this growth would do well to consider ETFs that serve tourism. This paper employs daily ETF prices of tourism sectors, including leisure and recreation (dynamic leisure and entertainment ETF, PEJ), media and publishing (dynamic media ETF, PBS), casinos and gaming (vectors gaming ETF, BJK), and consumer non-cyclicals (consumer goods ETF, IYK), consumer cyclicals (consumer discretionary select sector SPDR fund, XLY), and U.S.-total market (total stock market ETF, VTI) (see Table 1), between June 23, 2005, and March 5, 2018. The earlier data for the casinos and gaming ETF is from January 24, 2008. Daily returns are defined as:

R (t ) = ln (Pt )

ln (Pt 1)

where ln (Pt ) is the natural logarithm of the closing price at date t and ln (Pt 1) is the natural logarithm of the closing price at date t The conditional distribution that produces a sequence of observations with length d can be specified as:

(Xt + 1:t + d ) =

1.

(Xt + 1:t + d St + 1:t + d) = i

where i represents one of the states (i.e., i = {1, 2, , m} ), Xt + 1:t + d denotes the sequence of observations from time t + 1 to t + d , and St + 1: t + d indicates the sequence of states starting at time t + 1 and ending at t + d . Therefore, state i beginning at t + 1 produces a sequence of d observations. Details on the distribution of return series are discussed in Section 4.4. Summary statistics of price returns in the six ETFs can be found in Table 2, where sample means, medians, maximums, minimums, standard deviations, skewness, kurtosis, and the Jarque-Bera (JB) statistic are reported. The JB statistics indicate a departure from normality and the existence of nonlinear components in the data generating process. Table 3 reports the unit root test results, where various tests reject the null hypothesis of a unit root at the 0.01 significance level. Table 1 Summary of ETFs. ETF Name

ETF Abbreviation

Issuer

Market Segment

Dynamic Leisure and Entertainment ETF Dynamic Media ETF Vectors Gaming ETF U.S. Consumer Goods ETF Consumer Discretionary Select Sector SPDR Fund Total Stock Market ETF

PEJ PBS BJK IYK XLY VTI

Invesco PowerShares Invesco PowerShares VanEck iShares SPDR Vanguard

U.S. Leisure & Recreation U.S. Media & Publishing Global Casinos/Gaming U.S. Consumer Non-cyclicals U.S. Consumer Cyclicals US Total Market Index

3

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Table 2 Summary statistics.

Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis Jarque-Bera P-value

PEJ

PBS

BJK

IYK

XLY

VTI

0.0004 0.0009 0.1258 −0.0793 0.0150 −0.1016 9.2084 4078 0.0000

0.0003 0.0009 0.1146 −0.1002 0.0148 −0.4327 9.9366 5165 0.0000

0.0001 0.0009 0.1981 −0.1594 0.0185 −0.1799 15.1365 15,583 0.0000

0.0003 0.0007 0.0856 −0.0719 0.0101 −0.2201 12.0592 8695 0.0000

0.0005 0.0009 0.0933 −0.1236 0.0141 −0.4361 10.7146 6371 0.0000

0.0003 0.0007 0.1207 −0.0982 0.0129 −0.2561 13.6572 12,033 0.0000

Table 3 Stationarity test with outlier test. Unit root test

Outlier Test

Variables

P-value (ADF)

P-value (PP)

P-value

Break Date

PEJ PBS BJK IYK XLY VTI

0.01 0.01 0.01 0.01 0.01 0.01

0.01 0.01 0.01 0.01 0.01 0.01

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

2008/11/23 2008/11/23 2008/10/12 2008/10/12 2008/10/27 2008/10/12

Note: Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) unit root tests are used at level form and first difference of each series. The ADF tstatistics for the series without the constant and the trend term are all statistically insignificant to reject the null hypothesis of the unit root. Regarding the PP test, the selected truncation for the Bartlett Kernel are based on the suggestion by Newey and West (1994). The optimum lag order is selected based on the BIC criterion. The innovation outlier test follows Perron (1989) assuming that the breaks occur gradually while following the same dynamic path as the innovations. The results for univariate unit root tests with structural breaks are based on the asymptotic one-sided pvalues of Vogelsang (1993).

It can therefore be concluded that the price return series are stationary. 4. Methodology 4.1. Regime-switch cointegration test In the process of cointegration test, in general, the study uses ADF, Za and Zt three residual based test statistics provided by Engle and Granger (1987) and Phillips (1987) to examine the existence of cointegration. However, the power of the standard co-integration tests deteriorates once significant time-varying relationships and structural breaks become evident in the data generating process, resulting in the bias of rejection of the null hypothesis of no co-integration. Based on the foregoing, Gregory and Hansen (1996) used regime-switch cointegration because the traditional cointegration tests may spuriously fail to reject the null hypothesis of non cointegration where there are structural changes. Based on the framework of Gregory and Hansen (1996), the model in this study we used is developed by Hatemi-J (2008) that considers the impact of two structural breaks on both the intercept and slopes (two regime shifts). The method used in this study can be specified as:

yt =

0

+

1 D1t

+

2 D2t

+

0 xt

+

1 D1t xt

+

2 D2t xt

(3)

+ ut

where D1t and D2t are dummy variables defines as:

D1t =

0, if t [n 1] 1, if t > [n 1]

and

D2t =

0, if t [n 2] 1, if t > [n 2]

The date of pairwise breaks is estimated with the unknown parameters 1 (0, 1) and 2 (0, 1) . To test the null hypothesis of no cointegration, the statistics is defined as TS = inf TS , where TS could be ADF, Za or Zt , and T =(0.15n, 0.85n) . The range of T is ( 1, 2) T

based on the suggestions by Gregory and Hansen (1996). Approximate asymptotic critical values for tests of cointegration with two regime shifts can be referred to Hatemi-J (2008). 4

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

4.2. E-GARCH GARCH is a statistical model that can be used to estimate the volatility caused by the impacts, however, the model cannot distinguish the positive/negative impacts influencing the level of the volatilities. Therefore, the study adopts E-GARCH model proposed by Nelson (1991) to capture the asymmetric volatility on PBS, PEJ, and BJK causing by VTI, IYK, and XLY. The previous researches showed the asymmetric leverage effects that means the negative information (events, news, and so on) tend to impact volatility more than the positive information. For instance, the study uses E-GARCH(1,1) and the mean equation can be modeled as:

rt =

+

0

1 rt 1

ln(h2t ) =

+

1 VTIt t 1

+

ht

+

2 IYKt t 1

+

ht

1

1

+

3 XLYt

+

(1)

t

+ ln(h2t 1)

(2)

where t is innovation term, rt is the daily returns of PBS, PEJ, and BJK, is the intercept, ht is conditional volatility. In the variance equation, /ht > 0 represents the positive information in the market. On the other hand, /ht < 0 describes the negative information in the market. In addition, when = 0 means that the impact of the information causing the symmetries effects in the market, otherwise, there is the leverage effects existed in the market ( < 0 ). 4.3. Frequency dynamics of connectedness Vector Autoregressive (VAR) Models (Sims, 1980) are used for multivariate time series to examine and forecast the macroeconomics, to establish the impulse response function measuring the comparative static analysis, and to forecast the error variance decomposition for evaluating the volatility of endogenous variables. However, the ordering of the variables is important as it may affect the results when the equations within the VAR are analyzed through a matrix called Cholesky decomposition (Kilian, 2011). Hence, Pesaran and Shin (1998) proposed the generalized impulse response function to improve the drawback of Cholesky decomposition. Based on Pesaran and Shin (1998) research, Diebold and Yilmaz (2012) developed the spillover index and obtained the directional spillovers. Moreover, Lau, Vigne, Wang, and Yarovaya (2017), Barunik and Krehlik (2016), and Křehlík and Baruník (2017) adopt the spectral representation to decompose the generalized impulse response function and to forecast the error variance for capturing frequency dynamics of connectedness (long-term, medium-term, or short-term) based on frequency domain. The study will adopt Křehlík and Baruník’s (2017) model to evaluate the tourism ETFs’ connectedness. Consider a stationary VAR model defined over a time domain {t = 1, 2, , T} with order p, the dependency structure can be represented as follows: p

xt =

i xt i

+

t,

(4)

i=1

where xt = (x1t , x2t, , xmt) is an m × 1 endogenous variables vector, { i , i = 1, , p} is a m × q coefficient matrix, t is white noise with covariance matrix . According to Pesaran and Shin (1998) assumption 2.2, Eq. (4) can be shown by Wald representation by using iterative method that xt can be represented as linear function of t i , xt = i = 0 i t i , where i is the m × m coefficient matrix. Using recurrence relation showed as follows: i

=

1 i 1

+

+

i p,

p

i = 1, 2,

(5)

,

with 0 = Im and i = 0 for i < 0. When t follows the hypothesis of multivariate normal distribution, Pesaran and Shin (1998) proposed the generalized impulse response functions (GIRF’s) describing the dynamic process when there is one standard error shock at time t and (t + h) by one endogenous variable. GIRF is defined as follows: j (h)

=

jj

1 2

h

e j,

(6)

where jj is the diagonal element j of covariance matrix , ej is the vector when element j is 1 and the others are 0. Based on the concept of GIRF, forecast error variance decomposition is described as follows: ij (h )

=

jj

1

h k=0

(ei

h e k=0 i

k

k

ej )2 k ei

, i, j = 1, …, m .

(7)

Eq. (7) can be used to measure the percentage of variable volatility caused by itself or specific shocks. After standardizing Eq. (7), the equation can be used to calculate spillover effect index (Diebold & Yilmaz, 2012). On the other hand, Stiassny (1996) and Křehlík and Baruník (2017) discussed unconditional connectedness relations in frequency domain using spectral representation to define frequency response function in order to decompose GIRF. The spectral behavior of series xt can be expressed as follows:

Sx ( ) =

E(xt xt h)e

ih

=

(e

ih

)

(e

ih

ih

is Fourier transform of the impulse response

),

(8)

h=0

where

is the frequency,

(e

ih

)=

h=0

he

5

. Eq. (8) is called frequency response

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

function. On the frequency , forecast error variance is: jj

( ( ))i, j =

1

h=0

h=0

( (e

( (e ih

)

) )i2, j

ih

(e

ih

.

))i, i

(9)

According to Eq. (9), we can construct the connectedness table as Diebold and Yilmaz (2012) showed describing the impact part of the spectrum of xi due to shocks in xj. However, Křehlík and Baruník (2017) think the results of specific frequency on the connectedness table is discontinuous. Hence, Křehlík and Baruník (2017) presented cumulative connectedness table constructed on arbitrary frequency band d = (a, b) and expressed as follows:

( d ) i, j =

b

( ( ))i, j d

a

(10)

where

( ( ))i, j

( ( ))i, j =

m j=1

( ( ))i, j

is the Eq. (9) standardization. Accordingly, we can calculate overall connectedness, Cd , describing the connectedness intensity on spectral band (d = (a, b)) . Cd is defined as follows: m i = 1, i j

Cd =

i, j

( d ) i, j

( d ) i, j

=1

m i=1 i, j

( d ) i, i ( d )i, j

(11)

The larger implies strong connections within the spectral band (d = (a; b)) while the aggregate connectedness amongst the variables could be low. Moreover, we can measure the contribution of ETFi to ETFj , which can be defined as within from connectedness on the spectral band d:

Cd

k

Cid · =

( d )i, j j = 1, i j

and we can measure the variance from market ETFi to another ETFj as the within to connectedness on the spectral band d: k

Cid

( d )j , i

· j = 1, i j

In addition, we are interested in the difference between variance received and variance given from an ETFi which named the net connectedness, and the difference between ETFs which called pairwise connectedness. The net connectedness is defined as:

Cid, net = Cid

·

Cid

·

and the pairwise connectedness can be specified as:

Cid, j = ( d )j, i

( d )i, j .

If we sum up over disjointed intervals that give a range of frequencies, the conditional connectedness measure is defined as d

C= d

d

C , where C = C d (d ) , and

(d ) =

k

i, j = 1

( d )i, j /

k

i, j = 1

( )i, j = 1/ k

k

i, j = 1

( d )i, j . The

(d ) is the spectral weight, that is the con-

tribution of frequency band d to the whole VAR system. The parameter setting will be described in the next section. 4.4. Hidden-semi-Markov model In the study, we use Hidden Semi-Markov Model (HSMM) to analyze return characteristics of the tourism-related ETFs. A hidden semi-Markov model (HSMM) is proposed by Ferguson (1980) as an extension of the A hidden Markov model (HMM) by allowing the underlying process to be a semi-Markov chain with a variable duration or sojourn time for each state. The important difference between HMM and HSMM is that one observation per state is assumed in HMM while in HSMM each state can emit a sequence of observations. In addition, the duration d of a given state is explicitly defined for the HSMM because of the duration is dependent on both the state transition probability, the previous state transition probability, and its duration. The study defines the HSMM based on Bulla and Bulla (2006) research in the next paragraph. HSMM includes Observed process {Xt } and Unobserved state process {St } , {Xt } is related to {St } by the conditional distributions. Unlike HMM, the diagonal of translation probability matrix is 0 which means states are non-absorbing. The definition of translation probability matrix can be defined as:

pij = P (St + 1 = j|St + 1 where

j

i , St = i),

pij = 1 and pii = 0 , i , j

{1, 2,

(12)

, J } . The sojourn time of each state will follow probability distribution dj (u) , dj (u) is 6

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Table 4 Hatemi-J (2008) cointegration test results. ETFs

Test Statistic

CV 1%

CV 5%

Break one

Break Two

Conclusion

PEJ-IYK PEJ-XLY PEJ-VTI PBS-IYK PBS-XLY PBS-VTI BJK-IYK BJK-XLY BJL-VTI

−63.449 −60.396 −57.081 −59.206 −56.716 −53.947 −60.015 −60.160 −56.112

−7.903 −7.903 −7.903 −7.903 −7.903 −7.903 −7.903 −7.903 −7.903

−8.353 −8.353 −8.353 −8.353 −8.353 −8.353 −8.353 −8.353 −8.353

2015/1/16 2014/5/28 2012/6/6 2014/8/5 2014/7/15 2011/11/25 2014/10/16 2015/2/12 2012/7/24

2015/2/2 2014/5/30 2015/2/2 2015/1/7 2015/1/7 2015/1/9 2014/12/19 2015/2/12 2014/12/19

Cointegrated Cointegrated Cointegrated Cointegrated Cointegrated Cointegrated Cointegrated Cointegrated Cointegrated

The critical values (CVs) are obtained from Hatemi-J (2008).

defined as:

dj (u) = P (St + u + 1

j, St + u

v

= j, v = 1,

,u

1|St + 1 = j, St

(13)

j)

The study follows Bulla and Bulla (2006) and Lau et al. (2017) and apply the right-censored type HSMM to analyze the data that means the data will not translate immediately after the state of one observation. In the process of analyzing the data, the study sets conditional distribution is normal distribution, sojourn time distribution is negative binomial distribution, and adopts ExpectationMaximization algorithm to estimate the model parameters (Bulla & Bulla, 2006). Finally, we use Viterbi algorithm to decode the state sequence of ETFs. 5. Empirical results 5.1. Cointegration Table 4 reports the relationship between the three tourism ETFs (PEJ, PBS, and BJK) and other asset classes (IYK, XLY, and VTI). The results show that all markets are well integrated, while break dates can be identified around the years 2011, 2012, 2014, and 2015. The results on the relationship between tourism and equity markets are mixed in the literature, including either bidirectional causality between tourism and economic growth or unidirectional causality (i.e., the tourism-led economy growth and economicdriven tourism growth). Overall, our approach confirmed the presence of cointegration between all nine ETFs pairs: PEJ-IYK, PEJXLY, PEJ-VTI, PBS-IYK, PBS-XLY, PBS-VTI, BJK-IYK, BJK-XLY, BJK-VTI. 5.2. E-GARCH spillover effects Tables 5 to 7 present the parameters of the univariate E-GARCH(1,1) model for each tourism ETF market. Each table shows the estimated coefficients, standard errors, z-statistics and p-values for the conditional mean equation as in Eq. (6). All three tourism ETFs exhibit a significant own mean spillover from their first lagged returns. In this study, the mean spillovers are positive (except mean spillover from IYK is negative), results that are in line with the literature on other financial assets in the field. Table 5 presents the empirical results of return spillover from other asset classes to PEJ: there is a positive and significant relationship between the mean return in the leisure and entertainment market and the return in the consumer cyclical market. In particular, per unit increase in daily returns in XLY leads to an increase of 0.6786 unit of daily PEJ returns. The response of PEJ price to per unit increase in VTI is 0.4059 unit, while it is −0.0556 unit for IYK. Table 5 Inference on dynamic leisure and entertainment ETF (PEJ) (E-GARCH).

Panel A: mean equation 0

1

2 (IYK)

3 (XLY)

1 (VTI)

Panel B: variance equation 1 1

1

GED parameter

Coefficient

S.E.

t value

P-value

0.0001 −0.0934 −0.0556 0.6786 0.4059

0.0001 0.0276 0.0171 0.0425 0.0441

0.6300 −3.3869 −3.2546 15.9767 9.1977

0.5287 0.0007 0.0011 0.0000 0.0000

−0.0877 −0.0206 0.9915

0.0021 0.0119 0.0001

−41.1785 −1.7274 15978.4639

0.0000 0.0841 0.0000

0.1308 1.5386

0.0005 0.0572

7

260.8654 26.9080

0.0000 0.0000

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Table 6 Inference on dynamic media ETF (PBS) (E-GARCH).

Panel A: mean equation 0

1

2 (IYK)

3 (XLY)

1 (VTI)

Panel B: variance equation 1 1

1

GED parameter

Coefficient

S.E.

t value

P-value

0.0000 −0.0476 −0.1442 0.5503 0.6246

0.0001 0.0241 0.0178 0.0224 0.0221

−0.3925 −1.9781 −8.1214 24.5601 28.3226

0.6947 0.0479 0.0000 0.0000 0.0000

−0.0784 −0.0203 0.9926

0.0042 0.0114 0.0004

−18.8620 −1.7849 2435.4551

0.0000 0.0743 0.0000

0.1051 1.4341

0.0115 0.0635

9.1787 22.5711

0.0000 0.0000

Coefficient

S.E.

t value

P-value

0.0002 −0.0290 −0.0364 0.1154 0.9956

0.0002 0.0238 0.0832 0.0485 0.0623

0.7925 −1.2178 −0.4381 2.3789 15.9926

0.4281 0.2233 0.6613 0.0174 0.0000

−0.0389 −0.0023 0.9959

0.0027 0.0104 0.0002

−14.2874 −0.2203 4169.2330

0.0000 0.8257 0.0000

Table 7 Inference on vectors gaming ETF (BJK) (E-GARCH).

Panel A: mean equation 0

1

2 (IYK)

3 (XLY)

1 (VTI)

Panel B: variance equation 1 1

1

GED parameter

0.1045 1.4603

0.0223 0.0191

4.6803 76.3457

0.0000 0.0000

Table 6 presents the empirical results for PBS: the study finds evidence for a positive and significant relationship between the mean return in the total stock market and returns in the consumer cyclicals market. More specifically, per unit increase in daily returns in VTI leads to a 0.6246 unit increase in daily returns in PBS. The response of PBS to per unit increase in XLY is 0.5503 unit, which it is −0.1442 unit for IYK. Table 7 presents the results for BJK: a positive and significant relationship between the mean return in the total stock market and the consumer cyclicals market is observed. The effect of the total stock market on the casinos and gaming market is that per unit change in VTI returns leads to an increase in 0.9956 unit in the price of BJK. The response of BJK to per unit increase in the price of XLY is 0.1154 unit. Furthermore, the study can detect the presence of an asymmetric leverage effect, indicated by the term k = 0. Generally, the total stock market plays a principal role (the consumer-cyclicals market plays an important secondary role) in investors’ decision to put a tourism ETF, but not the IYK, in their investment portfolio. In addition, VTI returns are highly correlated with tourism ETFs; in other words, per-unit changes in VTI returns are strongly associated with tourism-ETF returns, especially for the vectors gaming ETF. 5.3. Frequency dynamics of connectedness Table 8 displays the decomposition of time-frequency dynamics of connections. The largest portion of connections is created from different time periods, including daily, weekly, and monthly cycles. The largest portion of connections is created from the higher frequency of one day up to one month (top panel of Table 7), with a value of 60.35. The connectedness of weekly and monthly cycles is 11.20 and 4.01, respectively. Considering the time dynamics of frequency connections, an interesting observation is that higherfrequency connectedness (monthly cycle) has been driven mostly by information up to low-frequency connectedness (daily cycle). The result is consistent with the studies of (Gonzalo & Ng, 2001; Blanchard & Quah, 1989; Quah, 1992), in which the authors built a preliminary notion of disentangling frequencies in connectedness. A shock with a strong long-run effect will have high power at low frequencies, and in case the shock transmits to other variables, this effect points to long-run connectedness. For example, in the case of stock markets, low-frequency spillovers may be attributed to permanent changes in expectations about future dividends (Balke & Wohar, 2002). Table 8 also displays the net-spillover indices for individual ETF markets. The daily contribution of PEJ is 12.80%. Furthermore, PEJ contributes 11.72, 14.03, and 12.57 of the spillover indices to IYK, XLY, and VTI, respectively, as a close substitute. In contrast, BJK only transmits 7.52% to other ETF markets. The total stock market ETF (VTI) receives 13.31%, the most spillovers from other 8

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Table 8 Total connectedness of tourism-related markets over different time intervals. PEJ

PBS

BJK

IYK

XLY

VTI

From ABS

From WTH

Net

Daily cycle PEJ PBS BJK IYK XLY VTI To_ABS To_WTH

18.92 13.34 9.95 11.72 14.03 12.57 10.27 12.80

13.96 18.27 10.12 12.97 14.54 14.38 10.99 13.71

7.28 7.18 25.87 7.42 6.61 7.72 6.04 7.52

11.17 11.66 9.40 19.86 13.25 14.57 10.01 12.48

14.51 14.30 9.56 14.41 18.15 14.82 11.27 14.05

13.37 14.43 11.08 16.51 15.27 18.14 11.78 14.68

10.05 10.15 8.35 10.50 10.62 10.67 60.35

12.53 12.66 10.41 13.09 13.24 13.31

0.22 0.84 −2.32 −0.50 0.07 1.10

Five-day cycle PEJ PBS BJK IYK XLY VTI To_ABS To_WTH

3.47 2.61 2.65 1.91 2.35 2.12 1.94 13.31

2.60 3.30 2.51 1.92 2.30 2.19 1.92 13.19

1.44 1.44 4.72 1.11 1.15 1.30 1.07 7.36

2.27 2.34 2.37 3.03 2.22 2.35 1.93 13.22

2.95 2.89 2.67 2.32 2.96 2.45 2.21 15.20

2.57 2.74 2.73 2.34 2.40 2.71 2.13 14.61

1.97 2.00 2.15 1.60 1.74 1.74 11.20

13.55 13.75 14.78 10.98 11.92 11.92

Monthly cycle PEJ PBS BJK IYK XLY VTI To_ABS To_WTH

1.24 0.94 0.96 0.68 0.84 0.76 0.70 13.33

0.93 1.18 0.91 0.68 0.82 0.78 0.69 13.17

0.52 0.52 1.69 0.39 0.41 0.46 0.38 7.35

0.82 0.84 0.86 1.08 0.79 0.84 0.69 13.26

1.06 1.04 0.97 0.83 1.06 0.88 0.80 15.24

0.92 0.98 0.99 0.83 0.86 0.97 0.76 14.61

0.71 0.72 0.78 0.57 0.62 0.62 4.01

13.56 13.79 14.95 10.89 11.88 11.88

Sum of spillovers from all markets j to market i number of marketsi + j Spillovers from markets i to all markets j To ABS: number of markets i + j Sum of spillovers from all markets j to markets i From WTH: number of all markets j Spillovers from all markets i to all markets j To WTH: 100 number of all markets j

75.23 −0.03 −0.08 −1.08 0.33 0.48 0.39

76.90 −0.01 −0.03 −0.40 0.12 0.18 0.14

76.95

From ABS:

100

The other columns contain net pairwise (i, j)-th spillovers indices.

markets, making it the largest net recipient of price spillovers. To summarize, PEJ, PBS, XLY, and VTI are net contributors to the spillover index, while BJK and IYK are net recipients based on the daily cycle data. Concerning tourism ETFs, the results of our study can help investors integrate shares of PEJ and PBS into investment portfolios because both of these ETFs contribute to high spillover indices. However, BJK does not transmit too much spillover effects to other individual ETF markets. Figs. 1 to 3 illustrate the time-frequency PEJ, PBS, and BJK, and other markets ETFs (i.e. IYK, XLY, and VTI). The decomposition of the bidirectional connectedness of frequency bands in a daily cycle are denoted by the black line, in a weekly cycle are denoted by the red line, and in a monthly cycle are denoted by the blue line. Several interesting observations can be made. PBS-XLY and PBS-VTI maintain the highest spillover index, as expected, since the connectedness is increasing in mid-2016. The smallest connectedness is between BJK and XLY. Regarding PEJ, PEJ-XLY and PEJ-VTI exhibit the highest spillover index, while the connectedness has been decreasing since mid-2014 and January 2017; the smallest connectedness is found for BJK. Regarding BJK, BJK-XLY is found to have the highest spillover-index before the market suffered the subprime mortgage crisis in 2008. In the current study, we found that leisure-and-recreation markets and media-and-publishing markets recovered faster than consumer cyclicals and total stock markets, a finding that should prove useful to investors. GDP growth was again strongly associated with tourism and market performance across multiple countries. This resilience of tourism markets can well serve countries severely impacted by a financial crisis. 5.4. Hidden semi-Markov model results Table 9 displays the estimated parameters of the fitted two-state HSMM for PEJ, PBS, and BJK. Considering daily returns of all three tourism ETFs, the variances of state 2 are close to those of state 1; that result means that both state 1 and state 2 have the same level of volatility. The means of state 1 for all three ETFs are negative, z-statistics indicate that the negative means of state 1 are not significantly different from zero. Therefore, there is statistical evidence to associate the volatility with mean of returns. The zstatistics show that the means of state 1 for PEJ, PBS, and BJK are not significantly different from zero for their state of volatility. The average sojourn time of either state 1 or state 2 for PEJ is longer than that for PBS and BJK (except state 2 for PBS), suggesting that the volatility clustering effect in PEJ is more persistent. The average sojourn time for BJK is shortest among the three ETFs. Investors can buy and hold PEJ stocks because of their characteristic of long-term stay in low-volatility state. In the meantime, BJK is more volatile than the other two tourism-related ETFs, and this is the case to such an extent that the returns for BJK are relatively 9

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Fig. 1. Dynamic frequency connectedness of PEJ with other markets.

Fig. 2. Dynamic frequency connectedness of PBS with other markets.

10

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Fig. 3. Dynamic frequency connectedness of BJK with other markets. Table 9 Estimation results of two-state HSMM. PEJ

Mean Variance Z Sojourn time r Sojourn time p No. of days Average Sojourn time

PBS

BJK

State 1

State 2

State 1

State 2

State 1

State 2

−0.002 0.001 −2.743 0.031 0.009 1695 452.25

0.001 0.000 3.423 0.029 0.003 842 182

−0.002 0.001 −2.554 0.023 0.008 1809 330.6

0.001 0.000 2.996 0.029 0.005 728 221

−0.003 0.001 −3.396 0.044 0.008 1653 282.5

0.001 0.000 2.214 0.497 0.061 884 168.4

unstable. 5.5. Decoding results Figs. 4 to 6 show the global decoding results of fitted two-state HSMMs for the three tourism-related ETFs. The white backgrounds stand for the low-volatility state (i.e., State 1) while the red background represents the high-volatility state (i.e., State 2). The study can show that PEJ, PBS, and BJK share a large overlap of the periods of low- and high-volatility states. There are some remarkable periods: 1. Before the middle of 2009, leisure and recreation, media and publishing, and casinos and gaming markets were in a high-volatility state due to the subprime mortgage crisis. The leisure and recreation market first went into a low-volatility state from high volatility, with a short period of low volatility before the European sovereign credit crisis. The high volatilities of the leisure and recreation market occurred continuously in few months in each year except the period during 2017–2018. 2. In the end of 2008 (because of the U.S. subprime mortgage crisis) and the end of 2011 (because of the European sovereign credit crisis), all three tourism-related ETFs were in the high-volatility state. 3. During 2013, all three tourism-related ETFs were in the low-volatility state. 4. After the middle of 2016, the casinos and gaming market went into a low-volatility state, while the media and publishing market went into a high-volatility state, and the leisure and recreation market went into a high-volatility state in 2017. 11

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Fig. 4. Global decoding of fitted two-state HSMM for PEJ.

Fig. 5. Global decoding of fitted two-state HSMM for PBS.

6. Discussion of results Our findings indicate that VTI plays an important role in tourism-related ETFs. Previous literature has found evidence of the strong relationship between the total stock market and tourism-related markets. A well-developed stock market is strongly correlated with growth rates of real GDP per capita (Goldsmith, 1969; Shaw, 1973; McKinnon, 2010; Rousseau & Wachtel, 2000; Beck & Levine, 2004). The evidence shows that investors can confidently integrate tourism-related ETFs into investment portfolios, with particular enthusiasm for the return effect of the total stock market ETF on the vectors gaming ETF (per-unit increases lead to 0.9965 unit changes). In addition, PBS has positive-return effects on tourism-related ETFs while IYK has negative-return effects on PEJ and PBS. These findings comport with those of previous research: capital gains are directly correlated with the consumption function (Bhatia, 1972; Peek, 1983). Existing empirical estimates of how wealth is correlated with consumption span a substantial range. Meyer (1994) showed that a $1 increase in equity values raises consumption in the next quarter by 2 cents, and the long-run impact of a $1 increase 12

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Fig. 6. Global decoding of fitted two-state HSMM for BJK.

in stock market wealth is a consumption increase of 4.2 cents. Ludvigson and Steindel (1998) also found that the effect of total wealth on consumption is 0.040 for their full sample during the 1953–1997 period. Moreover, Bone (1991) emphasized the importance of discretionary income, in particular, “since it is probably more closely tied to purchase behavior than is total income.” In our study, PBS plays an important role in PEJ and BJK. Owing to the growth of the Internet as a global megatrend, new media in much of the world has nearly taken over our lives, with individuals and communities seeking to find sustainable life satisfaction (Mercer, 1994). Previous literature has uncovered evidence of the relationship between, on the one hand, media-and-publishing markets and, on the other, leisure-and-recreation markets. The results from Leung and Lee (2005) and McCormick and McGuire (1996) show a significant relationship between quality of life and person-centered or place-centered leisure activities. The relationship means that the more frequently a person engages in new-media social activities, the more satisfied the person will be with the psychological benefits derived from leisure (Lloyd & Auld, 2001). Yi (2006) pointed out that the gaming industry, especially in South Korea, had become one of the core businesses in the new-media sector, and gaming software had become a representative of the new-media content of the 21st century. The rapid development of IT and related technologies has become the foundation for the growth of online gaming. Despite experiencing dramatic growth over the past decade, the online gaming industry is now in a phase of stagnation because of increasing competition, and a few major online gaming companies are already experiencing eroding growth and profits (Whang, 2006). Our current study’s findings indicate a persistent and strong relationship between media-and-publishing markets and two tourism-related markets (specifically, the leisure-and-recreation markets and the casinos-and-gaming markets). These findings, which indicate that media is a net contributor to price spillovers, are consistent with previous findings by pointing toward a positive and significant relationship between PBS and the returns of the two aforementioned tourism-related ETFs. Furthermore, the findings show that tourism-related ETFs can be used to monitor business cycles. For instance, tourism-related ETFs become more volatile before the onset of economic turmoil, such as the subprime-mortgage crisis and the European sovereign-credit crisis. During these crises, all tourism-related ETFs were highly volatile. During the economic recoveries, the tourism-related ETFs exhibited low volatility. In bull markets, investors can buy and hold dynamic leisure-and-entertainment ETFs and dynamic media ETFs because of the longest average sojourn time in state 1, but need not worry about the vectors gaming ETF, because it has the shortest sojourn time (i.e., in state 1 and state 2) in both bull and bear markets. Overall, our findings are in line with previous findings: by conducting a cointegration analysis of tourism-related ETFs, we observed that a relationship exists between total stock market and tourism industries, and that, in addition, IYKs were significantly negatively associated with tourism-related ETFs. Investors can focus on the total stock market, media-and-publishing markets, leisure-and-recreation markets, and consumer-cyclicals markets because the ETFs of those markets are net contributors to the spillover index; by contrast, the ETFs of consumer-goods markets and casinos-and-gaming markets are net recipients of the spillover index. 7. Conclusion This paper provided new empirical evidence on return spillovers between tourism-related markets from 2005 to 2018. The results improve the understanding of the impact that consumer goods ETF, consumer discretionary select sector SPDR fund, and total stock market ETF have on the dynamics of the connectedness of tourism ETFs (except consumer goods ETF in the long-time interval). This study further highlighted the relationship between tourism development and economic growth and how the tourism-related ETFs 13

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

become a new investment asset. The results are not only important from a theoretical perspective but also highly significant for a broad range of investors and practitioners. The findings can be summarized as follows. First, the application of a regime-switch cointegration test taking into account structural breaks revealed that all market pairs (dynamic leisure and entertainment ETF (PEJ)-consumer goods ETF (IYK), dynamic leisure and entertainment ETF (PEJ)-consumer discretionary select sector SPDR fund (XLY), dynamic leisure and entertainment ETF (PEJ)-total stock market ETF (VTI), dynamic media ETF (PBS)-consumer goods ETF (IYK), dynamic media ETF (PBS)-consumer discretionary select sector SPDR fund (XLY), dynamic media ETF (PBS)-total stock market ETF (VTI), vectors gaming ETF (BJK)-consumer goods ETF (IYK), vectors gaming ETF (BJK)-consumer discretionary select sector SPDR fund (XLY), vectors gaming ETF (BJK)- total stock market ETF (VTI)) are cointegrated. Second, E-GARCH model results display a positive and significant relationship between the mean of returns in the total stock market ETF and the mean of returns on dynamic leisure and entertainment ETF, dynamic media ETF, and vectors gaming ETF – confirming the role of total stock market ETF as a main source of return spillovers on the tourism-related ETFs market. These findings are in line with previous evidence that bidirectional causality between tourism and economic growth (Dritsakis, 2004; Ongan & Demiroz, 2005; Kim et al., 2006) and unidirectional causality with either the tourism-led growth (Marin, 1992; Balaguer & Cantavella-Jorda, 2002; Dritsakis, 2004) or economic-driven tourism growth hypotheses (Khan et al., 1995; Lee & Kwon, 1995). The results reveal that total stock market ETF have a higher influence on tourism-related ETFs. Indeed, we show that consumer goods ETF has significantly positive short-, medium-, and long-cycle impacts on tourism-related ETFs, while consumer discretionary select sector SPDR fund have significantly negative short- and medium-cycle impacts on tourism-related ETFs. Third, this paper reported new evidence on dynamic time frequency connectedness across different tourism fields, consumer noncyclicals, consumer cyclicals, and total stock markets, using the approach suggested by Barunik, Kocenda, and Vacha (2013). The results show that higher frequency (monthly cycle) connectedness has been driven mostly by information up to lower frequency connectedness (daily cycle). The values of net-spillover indices for individual ETF markets indicate that total stock market ETF (VTI) and consumer discretionary select sector SPDR fund (XLY) are net contributors of spillovers, while the vectors gaming ETF is a net recipient in different terms of periods. Furthermore, dynamic leisure and entertainment ETF (PEJ) and dynamic media ETF (PBS) are net contributors while consumer goods ETF (IYK) is a net recipient in daily cycles; however, dynamic leisure and entertainment ETF (PEJ) and dynamic media ETF (PBS) are changed to be the net recipients, while consumer goods ETF (IYK) is a net contributor in medium- and long-term cycles. Finally, estimation results of a two-state HSMM identified that the presence of volatility clustering effect in dynamic media ETF (PBS) is more pronounced than that for both dynamic leisure and entertainment ETF (PEJ) and vectors gaming ETF (BJK). Decoding the results highlighted four periods during which the high- and lower-volatility states of the tourism-related ETF markets experienced significant changes. These findings provide useful insights into the nature of volatility of leisure and recreation, media and publishing, and global casinos and gaming markets, and are useful for the creation of new successful investment portfolio strategies. References Ando, A., & Modigliani, F. (1963). The “life cycle” hypothesis of saving: Aggregate implications and tests. The American Economic Review, 53(1), 55–84. Atje, R., & Jovanovic, B. (1993). Stock markets and development. European Economic Review, 37(2–3), 632–640. Balaguer, J., & Cantavella-Jorda, M. (2002). Tourism as a long-run economic growth factor: The Spanish case. Applied Economics, 34(7), 877–884. Balke, N. S., & Wohar, M. E. (2002). Low-frequency movements in stock prices: A state-space decomposition. Review of Economics and Statistics, 84(4), 649–667. Barunik, J., & Krehlik, T. (2016). Measuring the frequency dynamics of financial and macroeconomic connectedness (No. 54). FinMaP-Working Paper. Barunik, J., Kocenda, E., & Vacha, L. (2013). Asymmetric connectedness of stocks: How does bad and good volatility spill over the U.S. stock market?. (pp. 1–28). Available on SSRN. Beck, T., & Levine, R. (2004). Stock markets, banks, and growth: Panel evidence. Journal of Banking & Finance, 28(3), 423–442. Bhatia, K. B. (1972). Capital gains and the aggregate consumption function. The American Economic Review, 62(5), 866–879. Blanchard, O. J., & Quah, D. (1989). The dynamic effects of aggregate demand and supply disturbances. The American Economic Review, 79(4), 655–673. Bone, P. F. (1991). Identifying mature segments. Journal of Consumer Marketing, 8(4), 19. Bulla, J., & Bulla, I. (2006). Stylized facts of financial time series and hidden semi-Markov models. Computational Statistics & Data Analysis, 51(4), 2192–2209. Chen, M. H. (2010). The economy, tourism growth and corporate performance in the Taiwanese hotel industry. Tourism Management, 31(5), 665–675. Chen, M. H., Kim, W. G., & Liao, C. N. (2009). The impact of government weekend policy changes and foreign institutional holdings on weekly effect of tourism stock performance. Journal of Hospitality & Tourism Research, 33(2), 139–160. Demirgüç-Kunt, A., & Levine, R. (1996). Stock markets, corporate finance, and economic growth: An overview. The World Bank Economic Review, 10(2), 223–239. Diebold, F. X., & Yilmaz, K. (2012). Better to give than to receive: Predictive directional measurement of volatility spillovers. International Journal of Forecasting, 28(1), 57–66. Dritsakis, N. (2004). Tourism as a long-run economic growth factor: An empirical investigation for Greece using causality analysis. Tourism Economics, 10(3), 305–316. Durbarry, R. (2004). Tourism and economic growth: The case of Mauritius. Tourism Economics, 10, 389–401. Engle, R. F., & Granger, C. W. (1987). Co-integration and error correction: Representation, estimation, and testing. Econometrica: Journal of the Econometric Society, 251–276. Eugenio-Martı’n, J. L., & Morales, N. M. (2004). Tourism and economic growth in Latin American countries: A panel data approach. Social Science Research Network Electronic Paper. Fayissa, B., Nsiah, C., & Tadasse, B. (2008). Impact of tourism on economic growth and development in Africa. Tourism Economics, 14(4), 807–818. Ferguson, J. D. (1980). Variable duration models for speech. Proceedings of the symposium on the applications of hidden markov models to text and speech (pp. 143–179). New Jersey: Princeton. Ghali, A. (1976). Tourism and economic growth: An empirical study. Economic Development and Cultural Change, 24, 527–538. Goldsmith, R. W. (1969). Financial structure and development. (No. HG174 G57). Gonzalo, J., & Ng, S. (2001). A systematic framework for analyzing the dynamic effects of permanent and transitory shocks. Journal of Economic Dynamics and Control, 25(10), 1527–1546. Gregory, A. W., & Hansen, B. E. (1996). Practitioners corner: Tests for cointegration in models with regime and trend shifts. Oxford bulletin of Economics and Statistics, 58(3), 555–560.

14

North American Journal of Economics and Finance 50 (2019) 100977

S.-L. Chang and Y.-H. Lee

Gunduz, L., & Hatemi-J, A. (2005). Is the tourism-led growth hypothesis valid for Turkey? Applied Economics Letters, 12(8), 499–504. Hatemi-J, A. (2008). Tests for cointegration with two unknown regime shifts with an application to financial market integration. Empirical Economics, 35(3), 497–505. Hazari, B. R. (1993). An analysis of tourists’ consumption of non-traded goods and services on the welfare of the domestic consumers. International Review of Economics & Finance, 2(1), 43–58. Khan, H., Phang, S. Y., & Toh, R. S. (1995). The multiplier effect: Singapore’s hospitality industry. Cornell Hotel and Restaurant Administration Quarterly, 36(1), 64–69. Khorana, A., Nellis, E., & Trester, J. J. (1998). The emergence of country index funds: An examination of world equity benchmark shares (WEBS). The Journal of Portfolio Management, 24(4), 78–84. Kilian, L. (2011) Structural Vector Autoregressions. CEPR Discussion Papers 8515, C.E.P.R. Discussion Papers. Kim, H. J., Chen, M. H., & Jang, S. C. (2006). Tourism expansion and economic development: The case of Taiwan. Tourism Management, 27(5), 925–933. Krause, T., Ehsani, S., & Lien, D. (2014). Exchange-traded funds, liquidity and volatility. Applied Financial Economics, 24(24), 1617–1630. Křehlík, T., & Baruník, J. (2017). Cyclical properties of supply-side and demand-side shocks in oil-based commodity markets. Energy Economics, 65, 208–218. Lau, M. C. K., Vigne, S. A., Wang, S., & Yarovaya, L. (2017). Return spillovers between white precious metal ETFs: The role of oil, gold, and global equity. International Review of Financial Analysis, 52, 316–332. Lean, H. H., & Tang, C. F. (2010). Is the tourism-led growth hypothesis stable for Malaysia? A note. International Journal of Tourism Research, 12(4), 375–378. Lee, C. C., & Chang, C. P. (2008). Tourism development and economic growth: A closer look at panels. Tourism Management, 29(1), 180–192. Lee, C. G., & Hung, W. T. (2010). Tourism, health and income in Singapore. International Journal of Tourism Research, 12(4), 355–359. Lee, C. K., & Kwon, K. S. (1995). Importance of secondary impact of foreign tourism receipts on the South Korean Economy. Journal of Travel Research, 34(2), 50–54. Leung, L., & Lee, P. S. (2005). Multiple determinants of life quality: The roles of Internet activities, use of new media, social support, and leisure activities. Telematics and Informatics, 22(3), 161–180. Levine, R., & Zervos, S. (1996). Stock market development and long-run growth. The World Bank Economic Review, 10(2), 323–339. Levine, R., & Zervos, S. (1998). Stock markets, banks, and growth. American Economic Review, 88(3), 537–558. Levine, R., & Zervos, S. J. (1993). What we have learned about policy and growth from cross-country regressions? The American Economic Review, 83(2), 426–430. Lloyd, K. M., & Auld, C. J. (2001). The role of leisure in determining quality of life: Issues of content and measurement. Social Indicators Research, 57, 43–71. Ludvigson, S. C., & Steindel, C. (1998). How important is the stock market effect on consumption?, Vol. 9821. New York: Federal Reserve Bank of New York. Marin, D. (1992). Is the export-led hypothesis valid for industrialized countries? Review of Economics and Statistics, 74, 678–688. McCormick, B. P., & McGuire, F. (1996). Leisure in community life of older rural residents. Leisure Sciences, 18(1), 77–93. McKinnon, R. I. (2010). Money and capital in economic development. Brookings Institution Press. Mercer, C. (1994). Assessing liveability: From statistical indicators to policy benchmarks. Urban and Regional Quality of Life Indicators, 3–12. Meyer, L. H. (1994). The WUMM model book. St. Louis: LH Meyer and Associates. Narayan, P. K. (2004). Economic impact of tourism on Fiji’s economy: Empirical evidence from the computable general equilibrium model. Tourism Economics, 10, 419–433. Nelson, D. B. (1991). Conditional heteroskedasticity in asset returns: A new approach. Econometrica: Journal of the Econometric Society, 347–370. Newey, W. K., & West, K. D. (1994). Automatic lag selection in covariance matrix estimation. The Review of Economic Studies, 61(4), 631–653. Oh, C. O. (2005). The contribution of tourism development to economic growth in the Korean economy. Tourism Management, 26, 39–44. Ongan, S., & Demiroz, D. M. (2005). The contribution of tourism to the long-run Turkish economic growth. Ekonomicky Casopis, 53(9), 880–894. Peek, J. (1983). Capital gains and personal saving behavior. Journal of Money, Credit and Banking, 15(1), 1–23. Perron, P. (1989). The great crash, the oil price shock, and the unit root hypothesis. Econometrica: Journal of the Econometric Society, 1361–1401. Pesaran, H. H., & Shin, Y. (1998). Generalized impulse response analysis in linear multivariate models. Economics Letters, 58(1), 17–29. Phillips, P. C. (1987). Time series regression with a unit root. Econometrica: Journal of the Econometric Society, 277–301. Poterba, J. M. (2000). Stock market wealth and consumption. Journal of Economic Perspectives, 14(2), 99–118. Proença, S., & Soukiazis, E. (2008). Tourism as an economic growth factor: A case study for Southern European countries. Tourism Economics, 14(4), 791–806. Quah, D. (1992). The relative importance of permanent and transitory components: Identification and some theoretical bounds. Econometrica: Journal of the Econometric Society, 107–118. Rousseau, P. L., & Wachtel, P. (2000). Equity markets and growth: Cross-country evidence on timing and outcomes, 1980–1995. Journal of Banking & Finance, 24(12), 1933–1957. Shaw, E. (1973). Financial deepening in economic growth. M, 1–278 19(73.21). Sims, C. A. (1980). Macroeconomics and reality. Econometrica: Journal of the Econometric Society, 1–48. Stiassny, A. (1996). A spectral decomposition for structural VAR models. Empirical Economics, 21(4), 535–555. Syriopoulos, T. C. (1995). A dynamic model of demand for Mediterranean tourism. International Review of Applied Economics, 9(3), 318–336. Tse, Y., & Martinez, V. (2007). Price discovery and informational efficiency of international iShares funds. Global Finance Journal, 18(1), 1–15. Vogelsang, T. J. (1993). A note on the asymptotic distributions of unit root tests in the additive outlier model with breaks (with P. Perron). Revista de Econometria, 13(2), 181–202. Whang, H. T. (2006). Gloomy domestic game firms before EC. Saegae Ilbo, p. 12. WTO (2016). Yearbook of tourism statistics. Madrid: World Tourism Organization. WTTC (2017). Annual reports, progress and priorities. The World Travel and Tourism Council. Yi, I. H. (2006). Korea, the everlasting empire in the online world. JoongAng Ilbo.

15