Forecasting Tourist Arrivals in Greece and the Impact of Macroeconomic Shocks from the Countries of Tourists Origin

Transcription

1 Forecastig Tourist Arrivals i Greece ad the Impact of Macroecoomic Shocks from the Coutries of Tourists Origi Dimitrios Gouopoulos Uiversity of Surrey, UK Dimitris Petmezas Uiversity of Surrey, UK Daiel Satamaria Caterbury Christ Church Uiversity Abstract: This paper geerates short-term forecasts o tourist arrivals i Greece ad performs impulse respose aalysis to measure the impact of macroecoomic shocks from the origi coutry o future tourism demad. We fid the ARIMA (1, 1, 1) model outperforms expoetial smoothig models i forecastig the directio of oe year out of sample forecasts. However, this does ot traslate ito poit forecastig accuracy. Impulse respose aalysis o the impact of uemploymet ad tourists cost of livig shocks shows that the source of dowside risk to future tourism umbers is limited i scope, magitude, ad duratio. Shocks to cosumer cofidece from the origi coutries have o impact o future tourism demad. Our results offer importat isights ad implicatios for policymakers ad tourist operators. Keywords: Tourist Arrivals, Macroecoomic Shocks, ARIMA, Holt s Expoetial Model With Tred, Double Expoetial Smoothig, Impulse Respose Fuctio Dimitrios Gouopoulos, (Surrey Busiess School, Uiversity of Surrey, Guildford, Uited Kigdom, GU2 7XH; <d.gouopoulos@surrey.ac.uk>), is a Lecturer i Accoutig ad Fiace i the Faculty of Busiess Ecoomics ad Law at the Uiversity of Surrey; Dimitris Petmezas, is a Professor i Fiace i the Faculty of Busiess Ecoomics ad Law of the Uiversity of Surrey, UK; ad Daiel Satamaria, is a Seior Lecturer i Fiace i the Faculty of Busiess ad Maagemet, Caterbury Christ Church Uiversity, Caterbury, Ket, CT1 1QU, UK.

2 INTRODUCTION The tourism idustry is oe of the most crucial sectors for a thrivig ecoomy as it accouts for a large part of some coutries Growth Domestic Product (GDP) ad employmet figures. Tourism is characterized by large variatios i umbers o a yearly basis ad, as a result, predictig future arrivals is a very difficult task. Forecasts of tourist arrivals are essetial for plaig, policy makig ad budgetig purposes by tourism operators (Uysal ad O Leary (1986)). I respose to this, a growig body of literature has focused o tourism demad ad arrivals forecasts i several coutries (for istace, Law (2) for Taiwa ad Hog Kog, Burger et al. for South Africa, Chu (28) for ie major tourist destiatios i the Asia-Pacific regio, Dharmarate (1995) ad Dalrymple ad Greeidge (1999) for Barbados, Gozález ad Moral (1996) for Spai, Chu (24), Sog ad Witt (26), Chu (29) for Asia-Pacific coutries, Lim ad McAleer (21), Athaasopoulos ad Hydma (28) for Australia, Smeral ad Weber (2) ad Papatheodorou ad Sog (25) for iteratioal tourism treds ad She et al. (21) for the Uited Kigdom outboud tourism demad) uder the research framework that the tourism idustry is a key sector i the ecoomic developmet strategy of may developig coutries. A secod strad of literature that has emerged i recet years is the use of macroecoomic factors to explai tourism demad usig structural time series models. For istace, Metzge-Quemarez (199) used real GDP figures from the Uited States, amogst other factors; Var et al. (199) ad Icoz et al. (1998) employed Turkish Cosumer Price Idex (CPI) figures ad the Turkish Lira currecy exchage rate agaist the currecy uits from the tourists coutry of origi, respectively; Greeidge (21), used real GDP ad CPI of the coutry of origi as well as the price idex of tourism i Barbados ad fially, Sog et al. (21) employed GDP data of the coutry of origi ad CPI i Hog Kog relative to the coutry of origi adjusted by the exchage rate. This paper seeks to break ew groud by aalyzig, for first time i the literature, the impact of macroecoomic shocks from the coutry of origi o future short term tourism demad to Greece. We examie the effect of tourists cost of livig, uemploymet ad cosumer cofidece i the coutry of origi as the source of macroecoomic shocks. 1

3 Particularly, the two latter variables have ot bee cosidered i the prior related literature. Tourists cost of livig is used as a measure of price competitiveess of the destiatio ad, as such, has a major impact o tourism demad. Uemploymet ad the cosumer cofidece idicator serve as useful proxies for the state of the ecoomy i the origi coutry, which implies a impact to future demad for tourism. The ituitio behid uemploymet lies i two aveues of research which have focused o the wage curve hypothesis ad the psychological impact o the level of happiess. Both explaatios imply a egative impact o future tourism umbers i periods of high uemploymet. The cosumer cofidece idicator reflects the level of ecoomic ucertaity ad/or expectatios o future icome ad the level of precautioary savigs. The build up of precautioary savigs feeds ito fallig levels i tourism demad as cosumers postpoe or cacel vacatios. Additioally, o study to our kowledge has attempted to forecast future arrivals i Greece, which is oe of the most popular tourist destiatios worldwide. Accordig to the Natioal Statistical Service of Greece, i 22 the coutry welcomed 14.9 millio iteratioal tourists placig Greece the 12 th place most visited destiatio iteratioally. This yielded a icome of $9.74 billio, boostig Greece i the top te i the world. It is therefore of paramout importace for policy makers ad idustry that forecastig models are developed ad tested to provide a accurate ad reliable picture of future tourism arrivals i Greece. As a result, ulike previous studies, we use a array of forecastig models to geerate short term predictios o tourism arrivals i Greece. The idetificatio ad aalysis of the impact of macroecoomic shocks from the coutry of origi o tourism flows i Greece itroduces a added dimesio by recogizig the mai source of risk to future arrivals. I this paper, we idetify the potetial risk comig from uemploymet, tourists cost of livig, defied as the CPI of Greece relative to the CPI of the coutry of origi, ad the cosumer cofidece idicator. 1 Although relative CPI has bee used 1 Although we recogize the importace of the CPI of other major competitors i determiig tourism flows to Greece, the uderlyig assumptio i this paper is that travellig to domestic destiatios is a viable substitute. The ituitio behid this otio lies i the coutries of origi used i this study (U.K., Frace, Germay, Italy, the Netherlads, Japa ad the U.S.) which accout for 51 millio domestic trips per aum i the E.U. (Peeters et al. (27). Furthermore, accordig to the WTO (25), domestic tourism i the U.S. amouted to more tha a billio trips per year. As a cosequece, the CPI of the origi coutry is viewed as the CPI of potetial competitors that is icorporated ito the tourists cost of livig used i this paper. 2

4 i past papers as a driver of tourism demad (Sog et al. (21) ad Abbas ad Ibrahim (211)), o study has cosidered the impact of uemploymet ad cosumer cofidece as macroecoomic iputs. Macroecoomic shocks from these factors are itroduced ito a Vector Autoregressive (VAR) system of equatios from which oe ca gauge the reactio ad time it takes to impact o future arrivals. Apart from ivestigatig the impact of radom macroecoomic shocks o future tourism arrivals, this paper also provides a methodological cotributio by utilizig impulse respose fuctio withi a VAR framework. This ivolves simulatig impulse resposes from the macroecoomic shock to provide iformatio o the size of the reactio ad the duratio of the effects o future arrivals. Cofidece bads are computed usig Mote Carlo Simulatio to determie the statistical reliability of the respose. Our results reveal a umber of iterestig observatios. First, prelimiary aalysis reveals that the ARIMA (1, 1, 1) model outperforms the double expoetial smoothig ad the Holt s expoetial smoothig model with tred as a short-term directioal forecastig tool. However, the success rate of the ARIMA model i capturig log term treds does ot traslate ito forecastig accuracy. Istead, based o the Mea Absolute Error (MAE), Mea Absolute Percetage Error (MAPE) ad Root Mea Square Error (RMSE) measures, the Holt s expoetial smoothig approach is the best performig model as poit forecastig tool. The ext set of results focused o how radom macroecoomic shocks, itroduced ito a system of equatios, could impact o future tourism demad i the short term. Iitial fidigs show that radom uemploymet shocks, as well as shocks to the tourists cost of livig, have profoud, yet time varyig, effects o short term tourism demad i Greece. I cotrast, shocks to cosumer cofidece from the coutry of origi have a beig impact o tourism demad. Closer ispectio of the results idicate that despite the lack of diversificatio i the sources of tourism demad to Greece, dowside risk i the two mai coutries of origi, the Uited Kigdom ad Germay, is limited. To sum up, this paper has two mai cotributios to the literature. First, we cosider tourist arrivals i Greece ad provide a prelimiary aalysis o iitial short term forecasts i future tourism demad. Give the importace of the tourism idustry i Greece ad the level of tourism demad, this addresses a major gap i the literature. The secod cotributio, ad oe 3

5 that forms the overridig objective of this paper, is that it explores the impact of macroecoomic shocks of various sources from the coutry of origi o future tourism demad. The rest of the paper is structured as follows. Sectio 1 reviews the related literature o the importace of the macroecoomic variables used i this study. Sectio 2 discusses the data used ad provides descriptive statistics. Sectio 3 presets the methodology. Sectio 4 aalyses the empirical results, followed by a discussio of the fidigs i Sectio 5. Sectio 6 discusses the implicatios of this study. Fially, Sectio 7 summarises ad cocludes the paper. 1. RELATED LITERATURE Although o studies have yet to documet the impact of macroecoomic shocks o future tourists arrivals, reviewig the wider literature o the impact of uemploymet, chages i the tourists cost of livig, ad cosumer cofidece could provide useful ifereces o potetial effects to iteratioal tourism flows Uemploymet I this paper, uemploymet is defied as a proxy used for the state of the ecoomy i the origi coutry which might have a potetial impact o future arrivals. The ituitio behid this partially lies i the literature o the wage curve hypothesis. This theorem is based o the relatioship betwee uemploymet i the local labour market ad the level of pay, where real wages are suggested to be egatively related to the uemploymet rate. Lower real wages imply fallig umbers goig abroad. Early studies have reported covicig evidece that the level of pay is lower i areas of high uemploymet across differet coutries (Blachflower ad Oswald ((19/9), (1994) ad (26)) based o the uemploymet elasticity of wages measure. 3 O the 2 We recogize the use of GDP used i previous studies as a determiat of tourism demad. However, we have restricted our database to iclude macroecoomic variables available mothly to maximize the umber of data observatios used whe geeratig short to medium term impulse resposes o future tourism demad. As a result, GDP figures, oly released o a quarterly basis, would lead to meaigless results i geeratig oe year ahead impulse respose paths that would esure cosistecy with short term forecasts aalyzed later i the paper. Moreover, although tourists cost of livig is a variable used i previous studies, a importat iovatio of this paper is the use of uemploymet ad cosumer cofidece as alterative proxies for the state of the ecoomy i the coutry of origi. 3 Further support is provided from studies that examie Europea coutries which documet a elasticity of approximately -.1 (Wager (1994), Estevao ad Nargis (22) ad Motuega, Garcia ad Feradez (23), ad Saz de Galdeao ad Turue (25)). Similarly, Deller ad Tsai (1998) reach the same coclusio for the Uited States. 4

6 other had, Malley ad Moutos (1996) provided aother agle by aalyzig uemploymet as a measure of aggregate icome ucertaity. Usig quarterly data from the Uited States, they fid a iverse relatioship betwee the level of cosumptio ad uemploymet that is attributable to a icrease i precautioary savigs durig periods of high uemploymet. Further ituitio behid the use of uemploymet as a proxy lies i the growig body of work o the psychological effect of uemploymet o the level of happiess ad well beig. Oe coclusive fidig that was held relatively uchalleged is that the level of uemploymet reduces the level of happiess ad well beig sigificatly. For istace, Blachflower ((1996), (21)), amogst others, reaches this coclusio after ivestigatig twety-three differet coutries. Further support for this fidig is provided by Ah et al. (24) who examie this effect for all coutries i the Europea Commuity. They fid evidece that uemploymet reduces the level of satisfactio both i fiacial terms ad vocatioal activity. This fidig varies across coutries, with uemploymet i Demark ad the Netherlads havig the least sesitive impact o well beig Cost of Livig-Cosumer Price Idex (CPI) The oly variable cosidered i this study that has bee used i the iteratioal tourism literature is the tourists cost of livig (or relative CPI) defied as the CPI for the destiatio coutry relative to the origi coutry (Habibi ad Rahim (29), Arsad ad Johor (21), Sog et al. (21) ad Abbas ad Ibrahim (211) amog others). There is recogitio amogst academics ad the tourism idustry o the relevace of price competitiveess of the destiatio coutry (Dwyer et al. ((2a), (2b), (22)). Theoretically, a proxy for the tourists cost of livig should iclude travel cost to ad from the destiatio, i additio to the cost of accommodatio, tour services ad restaurats. However, due to a lack of data, most studies have omitted travel costs. Hece, the questio mark posed by previous studies is whether the CPI i the destiatio coutry is a reasoable proxy for tourism prices. Morley (1994) ivestigated the reliability of the CPI as a proxy for tourism prices for 1 major destiatios. Tourist expeditures estimated were foud to correlate with the CPI i the coutry of destiatio, a fidig that was foud to be robust to the removal of liear time treds 5

7 from the series. A similar coclusio was reported i a more recet study by Salma et al. (27) i relatio to formulatig ad testig the demad fuctio for tourism i Swede. I view of the importace of price competitiveess i the coutry of destiatio, shocks to relative CPI could have profoud implicatios for future tourism arrivals Cosumer Cofidece The fial macroecoomic variable cosidered i this study is the cosumer cofidece idicator as providig ifereces o the degree of ucertaity associated with future ecoomic coditios i the coutry of origi. Accordig to Ludvigso (24), high cosumer cofidece reflects reduced ucertaity o future ecoomic coditios, which traslates ito reduced precautioary savigs ad icreased preset cosumptio at the expese of future cosumptio. However, ecoomic theory surroudig cosumer cofidece exteds beyod precautioary savigs ad curret cosumptio. Theory also judges cosumer cofidece as a meas of capturig expectatio o future icome ad wealth ad, as such, it could impact o future cosumptio (Ludvigso (24)). I this paper, the cosumer cofidece idicator is used as a secod proxy for the state of the wider ecoomy. The ituitio behid this otio is provided by Matsusaka ad Sbordoe (1995), who report evidece that cosumer cofidece has a statistically sigificat impact o macroecoomic fluctuatios. Based o the above aalysis, we argue that high cosumer cofidece i the coutry of origi implies a fall i the level of precautioary savigs ad therefore may feed ito icreased tourism flows from that coutry; whether it has a immediate impact depeds o which theoretical explaatio holds. The implicatio of past studies cosiderig the role of cosumer cofidece was to examie whether the idicator merely serves as a proxy for a broader ecoomic cycle (Matsusaka ad Sbordoe (1995)). Doms ad Mori (24) added to this otio by reportig evidece that the cosumer cofidece idicator is more volatile i times durig ad after recessios, whe ews coverage is greater coicidig with a willigess to adjust expectatios. However, the vast majority of studies have ivestigated whether cosumer cofidece ca be used to forecast future household spedig ad cosumptio. The geeral cosesus formed is that it adds predictive power to short term forecasts (Carroll et al. (1994), Bram ad Ludvigso 6

8 (1998) ad Howrey (21)). Eppright et al. (1998) fid egative shocks have a adverse impact o cosumer cofidece that i tur traslates ito lower cosumptio. Take together, it is plausible to argue that uemploymet levels, havig a major effect o real wages, as well as people s level of happiess ad well beig, could affect the level of tourism activity. Additioally, the otio that tourists cost of livig as a reflectio of price competitiveess of the destiatio coutry, i additio to cosumer cofidece proxies for ecoomic ucertaity ad future icome, could have far reachig implicatios o future tourism arrivals to Greece. With o evidece available from previous studies, this will provide a more complete picture for policymakers o future arrivals especially i times of recessio whe poor ecoomic fudametals could feed ito short to medium term forecasts. As a matter of fact, this study is of particular iterest, give that Greece has bee i a recessio recetly. 2. DATA 2.1. Sample To coduct this study, our database cosists of mothly data o tourist umbers to Greece as well as uemploymet levels, Cosumer Price Idices (CPI) ad Cosumer Cofidece Idicators from the coutry of origi for the period Jauary 1977 to December 29. Macroecoomic variables for the Uited Kigdom, Uited States, Frace, Germay, Italy ad the Netherlads were dowloaded from Datastream. I additio to the coutries of origi, we also collected CPI data for Greece to compute relative CPI as a proxy for tourists cost of livig. Cosistet with previous studies, 4 the relative CPI (CP) is defied as: where CPI i, t ad CPI, are the cosumer price idexes of Greece (the destiatio coutry i) ad the coutries of origi j, respectively. j t CPI CPt CPI Tourism data was collected from a variety of differet sources. First, data regardig the arrivals i Greece ad coutries of tourists origi were obtaied from the Helleic Statistical Authority. Cross checks ad additioal iformatio were extracted from airlies, cruise i, t j, t 4 See Daiel ad Ramos (22), Gari-Muoz (26) ad Gari-Muoz ad Motero-Marti (27), amog others, i relatio to Portugal ad Spai respectively. 7

9 compaies, travel idustry sources, big tourism operators, such as the Associatio of British Travel Agets (ABTA), the Iteratioal Air Trasport Associatio (IATA), the Greek Natioal Tourism Orgaisatio (GNTO), the Associatio of Greek Tourist Eterprises (SETE), the Helleic Associatio of Travel & Tourist Agecies, the Europea Travel Commissio (ETC), the Geeral Secretariat of Natioal Statistical Service of Greece (NSSG) Miistry of Ecoomy ad Fiace, Athes Iteratioal Airport (Eleftherios Veizelos), Mediterraea Cruise Ports (MedCruise), Piraeus Port Authority ad the Uited Kigdom Office for Natioal Statistics. Figure 1 illustrates the importace of the origi coutries chose i this paper as the source of tourism demad i Greece. These coutries are: Frace, Germay, the Uited Kigdom, Italy, the Netherlads, ad the Uited States. The coutries selected make up o average more tha 5% of the total arrivals i Greece throughout the etire sample period. [Please Isert Figure 1 About Here] 2.2. Descriptive Statistics Tourist Arrivals The total umber of tourist arrivals i Greece from are preseted i Table 1. The table cosiders tourist arrivals from the coutries of origi cosidered i the paper. The statistics geerally idicate a rapidly icreasig tred that bega i the 198 s ad reached a peak of 17.2 millio tourists i 26. It is worth metioig that Greek tourism uderwet much developmet durig this period i which icreases i tourism arrivals were registered for twelve out of the thirtee years mostly due to spatial polarizatio, the itesificatio of seasoality, ad the productio ad distributio of tourism cosumptio (Galai-Moutafi (24)). The two years that followed (28 29) clearly showed a declie due to the global fiacial crisis ad icreased competitio from ewer holiday hot spots, such as Moteegro, Croatia ad Turkey offerig similar attractios. Greece has a high percetage of repeat customers, but as a member of the Eurozoe, it is more expesive tha some of the up-ad-comig destiatios ad less appealig to those o a fixed icome (retirees for istace) or families seekig a budget holiday (Alegre et al. (21)). Table 1 also reveals that tourist arrivals from Europe comprise the majority of foreig 8

10 tourists i Greece. For istace, Germay ad the Uited Kigdom, two importat sources of tourism, reported a average aual growth rate of approximately six percet. Arrivals from the Uited Kigdom peaked at over three millio i 23. However, sice the, there has bee a steady drop of over 1, arrivals per year maily due to itese competitio from other destiatios, icludig domestic, offered by tour operators as well as the iteret (Miller, Rathouse, Scarles, Holmes, ad Tribe (21)). However, further aalysis of Table 1 reveals clear evidece that the source of tourism arrivals is udiversified ad heavily reliat o the Uited Kigdom, followed by Germay. Both coutries cotribute, o average. 16.5% ad 15.2% respectively, to total tourism demad betwee 1977 ad As a result, establishig the Uited Kigdom ad Germay as the mai source of tourism demad has implicatios i idetifyig the greatest source of risk whe impulse respose aalysis is performed later i the study. [Please Isert Table 1 About Here] Macroecoomic Factors Table 2 provides descriptive statistics o the macroecoomic variables for the coutry of origi from 1977 to 29. Relative CPI, from a low of about five percet i Italy to almost ie percet i Germay, idicates that the cost of livig i Greece has icreased sigificatly relative to all coutries of origi over the sample period. As a result, high iflatio i Greece relative to other coutries seems as a potetial source of risk to future arrivals. Uemploymet appears to be the most volatile macroecoomic factor cosidered eve though it has icreased over the period, from just over oe percet i Italy to aroud five percet i Germay, thus suggestig aother potetial source of risk to future tourism demad i the short term. O the other had, despite showig greater variability, as implied by the miimum ad maximum values, the cosumer cofidece idicator has remaied relatively stable over the sample period. 5 Further evidece of a lack of diversificatio i the source of tourism demad is provided whe we performed the same aalysis (ot reported for space purposes) to o-eu coutries ad regios. No-EU coutries such as Russia, Turkey ad Japa cotributed oly.9%,.8% ad.9%, respectively, over the same period. For regios, we looked at the E.U. block, Asia, Africa ad Oceaia zoes, oly to fid uequivocal evidece that the source of tourism demad is udiversified with 66.3% arrivals comig from E.U. block coutries, 4.2% from Asia,.9% from Africa ad 1.1% from the Oceaia zoe. 9

11 Uemploymet Treds Table 2 also provides ifereces o aual treds i the macroecoomic factors by takig the year o year percetage chage as of December of each year. Focusig o uemploymet treds i relatio to table 1, the Uited Kigdom, regarded as the most importat source of tourism, appears to be isesitive to icreases i aual uemploymet except durig Cotrary to the geeral coclusios of Malley ad Moutos (1996), sigificat icreases i uemploymet betwee ad were associated with icreases i tourists umbers. Accordig to ureported data from Laborsta orgaisatio, this may be attributable to the upward tred i employmet levels over the past three decades, as from approximately 24.8 millio people beig employed i 1977, the umber icreased to aroud 3.8 millio people by 25. However, the sigificat fall i uemploymet betwee 1993 ad 2 was associated with the largest icrease i tourist umbers i Greece over the same period. The icrease i uemploymet durig coicided with a 19.26% reductio of British tourists who visited Greece. Germay, the secod most importat source of tourism demad, experieced the highest uemploymet icreases durig , ad Despite this, table 1 shows that tourism from Germay icreased durig the first period ad little chaged i the secod period; this was followed by a fall of 1% i 23 relative to the previous year. Tourism flow appears to be isesitive to upward treds i uemploymet i Italy betwee 1978 ad Accordig to Table 1, the greatest icrease i tourism umbers coicided with a sustaiable period of decliig uemploymet from 1999 to 27. Similar patters occur i relatio to Frace ad the Netherlads whereby tourist arrivals from both coutries appear to be relatively isesitive to icreases i uemploymet Relative CPI Tred Tourists Cost of Livig Table 2 provides also iformatio o the tred i the tourists cost of livig defied as the CPI i Greece relative to the coutries of origi. Double digit icreases i the cost of livig i Greece was observed relative to the coutries of origi betwee 198 ad Iterestigly, the rate of icrease declied quite dramatically later i the sample to the extet that, i some cases, the cost of livig i the coutry of origi icreased relative to Greece (Uited Kigdom, Uited 1

12 States, Italy ad the Netherlads). Despite this, tourism demad to Greece icreased dramatically over the same period. Germay, the secod most importat source of tourism, icreased tourist arrivals i Greece by 16%, followed by 99% from the Netherlads, 8% from Italy, 57% from Frace, 56% from the Uited Kigdom ad, fially, 25% from the Uited States. O the other had, the decliig rate of icrease i the relative CPI from 1993 has bee associated with marked shifts i the tourism arrivals. For istace, tourism arrivals from the Uited States icreased by 13% due to a major surge i umbers i 25. Tourism arrivals from Italy icreased by 81% followed by 75% from Frace ad 57% the Netherlads. Tourism arrivals from the largest cotributor, the Uited Kigdom, oly icreased by 36% over the same period Cosumer Cofidece Idicator Oe of the key characteristics of the cosumer cofidece idicator is the stability ad lack of tred. However, closer ispectio of the results, i relatio to Table 1, reveals some iterestig fidigs. The largest year o year icreases i tourism arrivals betwee 1984 ad 1985, followed by 1998 ad 2, coicided with little or o icrease i the cosumer cofidece idicator. Evidece of decliig cosumer cofidece, traslatig ito decliig tourism arrivals, is restricted to Italy ad the Netherlads at the height of the fiacial crisis i 28, as show i table 2. O the other had, the same period was related with similar falls i cosumer cofidece i Frace ad Germay, ad tourism arrivals icreased by 2% ad 9%, respectively. [Please Isert Table 2 About Here] 3. METHODOLOGY I this paper we use the impulse respose fuctio to measure the impact of macroecoomic shocks o future tourism demad. The implicatio of geeratig impulse resposes is to idetify the source of risk to future tourism umbers due to the arrival of macroecoomic shocks. There is a body of work i the tourism literature which uses macroecoomic iputs ito structural time series models to explai future tourism demad. For istace, Metzge-Quemarez (199) used real Growth Domestic Product figures from the Uited 11

13 States, amogst other factors; Var et al. (199) ad Icoz et al. (1998) employed Turkish CPI figures ad the Turkish Lira currecy exchage rate agaist the currecy uits from the tourist s coutry of origi, respectively; Greeidge (21), used real Growth Domestic Product ad CPI of the coutry of origi as well as the price idex of tourism i Barbados ad fially, Sog et al. (21) employed Growth Domestic Product data of the coutry of origi ad CPI i Hog Kog relative to the coutry of origi adjusted by the exchage rate. However, ulike previous studies, we do ot use macroecoomic iputs to explai the demad fuctio of tourism arrivals. Istead, we geerate macroecoomic shocks from these variables through the impulse respose fuctio by utilizig the Vector Autoregressive (VAR) Model, first itroduced by Sims (198). Sice its itroductio, it has bee widely used i the ecoomics literature i the use of macroecoomic variables. For istace, withi the uemploymet literature, the VAR has bee used extesively to geerate forecasts of the atural rate (Groeewold ad Hagger (2) ad Kig ad Morley (27)) ad turig poits i the rate of uemploymet (Edlud ad Karlsso, (22)), just to list a few. However, i the cotext of this paper, the VAR approach is oly used to set up a system by which oe would itroduce radom macroecoomic shocks ad aalyze its impact o future tourism demad. To determie the statistical reliability of the respose, Mote Carlo Simulatio is used to costruct cofidece bads aroud the impulse respose. This is of paramout importace to policy makers ad idustry, as it provides useful ifereces o the sesitivity of future tourism arrivals to macroecoomic shocks ad the potetial source of risk from the coutry of origi The Impulse Respose Fuctio The impulse respose fuctio is a valuable tool that ca be used to isolate the impact of a macroecoomic shock from the coutry of origi o future tourists arrivals to Greece, assumig other variables are held costat. For the purpose of this study, cosider the simple VAR model cosistig of tourism demad i destiatio i deoted as Y i, t ad macroecoomic iputs X, from the coutry of origi j at time t: j t Y i, t X j, t 1 2 Y 11 i, t 21 X j, t 12 X 12 j, t Y u 22 i, t i, t u j, t (1)

14 i which, j, t X f UN, CP, CON j ij j where UN j, CP ij ad CON j represet uemploymet, tourists cost of livig ad cosumer cofidece idicators from each coutry of origi, respectively. The term deotes the vector of coefficiets that represet the effect of umber of lags i the system. X j t, ad i t Y, o i t Y, ad vice versa, ad is the The model of equatio (1) is a VAR() specificatio give that the variables i the system have a lag of. A chage i the iovatio u i, t will immediately chage all future values of Y ad X, sice lagged Y appears i both equatios. Assumig that the iovatios u i, t ad u j, t are ucorrelated, the iterpretatio of the impulse respose is straightforward. The u i, t is the iovatio for Y ad u j, t is the iovatio for X. The impulse respose fuctios for u j, t measure the impact of a radom macroecoomic shock o future tourism demad. The iovatios u i, t ad u j, t are, however, usually correlated, so that they have a commo compoet that caot be associated with a specific variable. A commo, but arbitrary, method of dealig with this issue is to attribute the full impact of ay commo compoet to the variable that comes first i the VAR system. I this case, the commo compoet of u i, t ad u j, t is u i, t give that the iovatio u i, t precedes u j, t. Hece, u i, t becomes the Y ad X iovatios, which are trasformed to remove the commo compoet. We trasform the iovatios by orthogoalisig the errors usig the Cholesky factorisatio. This is a popular method of trasformig the covariace matrix of the resultig iovatios i the VAR residuals ito a vector of orthogoal iovatios defied as e t. E ei, te j, t where i j (2) To trasform the error terms, a N N lower matrix defied as V is chose ad the orthogoalised iovatios e t are obtaied to satisfy the followig equatio: e uv where, the iovatio u t has a idetity covariace matrix such that: 1 (3) T Eee (4) 13

15 ad VV T (5) Upo makig the trasformatio of the orthogoalised iovatio ad replacig the u t with e V t, the VAR model, expressed as a movig average represetatio, ca be rewritte as follows: By defiig B A V, equatio (7) becomes: Y A Ve (6) t t Y B e (7) t t where, B represets the impulse respose of the market i the future to a shock of oe stadard deviatio i time t. Hece, the elemets of B are the impact multipliers. Assumig that tourism demad Y is statioary, the impulse respose should ted towards zero as icreases. 4. EMPIRICAL RESULTS 4.1. Prelimiary Aalysis Forecastig Tourism Arrivals Table 3 presets the performace of oe year out of sample forecasts for tourism demad Y i Greece usig the ARIMA (1, 1, 1), double expoetial smoothig, ad the Holt s expoetial smoothig model with tred. 6 The statistics relate to the performace of all models sice the estimatio period expads whe the model forecasts are updated as ew data becomes available. This icludes statistics o the success rate at which the model captures the directioal forecasts ad measures of forecastig accuracy usig the Mea Absolute Error (MAE), Mea Absolute Percetage Error (MAPE) ad Root Mea Square Error (RMSE). 7 Ituitively, this 6 Before implemetig the forecastig models, we tested aual total arrivals series for a uit root usig the Phillips- Perro (1988) test. The results are ot reported i the paper, but are available upo request. We fid the ull hypothesis of a uit root accepted i the log levels Y but rejected after takig the first differeces Y. Therefore, we ucover statioarity i the trasformed series Y. 7 The Mea Absolute Error (MAE) measures the degree to which forecasts ad the outcomes are close together. The Mea Absolute Percetage Error (MAPE) is similar except that it presets a measure of forecastig accuracy relative to the evetual outcome i terms of a percetage, whereas, the Root Mea Square Error (RMSE) is a measure of precisio based o the residuals aggregated over the back-test period. All three measures of accuracy are frequetly used i the literature (for istace, Preez ad Witt (23), Sog et al (23) ad Chu (29) to list a few). 14

16 would eable oe to idetify the forecastability of tourism demad to Greece i light of the potetial risks to future umbers. [Please Isert Table 3 About Here] I geeral, the ARIMA (1, 1, 1) model outperforms other expoetial smoothig models as a directioal forecastig tool over a expadig estimatio period. The performace is most impressive whe it forecasts a icrease i tourists umbers i oe year s time. The success rate rages from 79% to 68%. Despite this, the directioal forecastig performace of the ARIMA model does ot traslate ito superior accurate forecasts. Based o the Mea Absolute Error (MAE), Mea Absolute Percetage Error (MAPE) ad Root Mea Square Error (RMSE) measures, the most cosistet performig model as a poit forecastig tool is the Holt s expoetial smoothig model with tred. O the other had, the worst performig model based o these criteria is the ARIMA (1, 1, 1) model, a fidig that is robust as the estimatio period expads. These fidigs are cosistet with the results of Smeral ad Wuger (25) who reported that the aïve model outperformed the ARIMA model. O the other had, our results cotrasts with the early fidigs of Chu (1998), followed by Preez ad Witt (23), i which the superior performace of the ARIMA model i relatio to other approaches was highlighted. Fidig differeces i model performace usig differet approaches is also ot surprisig. For istace, Clemets ad Hedry (1998) argue that the performace of ecoometric models is determied by the methodology used to geerate forecasts The Impact of Macroecoomic Shocks o Future Tourism Arrivals The VAR Model I this sectio, we address the impact of macroecoomic shocks o future tourism demad for Greece. Firstly, we utilize the VAR () model o mothly data of tourists arrivals to Greece, uemploymet, tourists cost of livig, ad cosumer cofidece from the coutries of origi from 1977 to 29. Give that VAR models are modelled o a statioary time series, the first step requires the implemetatio of uit root tests o each series. Istead of usig the 15

17 Augmeted Dickey-Fuller (ADF) test, we employ the Phillips-Perro (1988) approach o the level series first followed by the trasformed series. The ituitio behid the use of the Phillips-Perro (1988) approach is that it has more power tha the Augmeted Dickey-Fuller test. Oe issue that arises with the Augmeted Dickey- Fuller is the selectio of the umber of lags that could lead to a bias towards rejectio of the ull hypothesis of a uit root i the evet of selectig too few lags. Coversely, bias towards acceptig the ull hypothesis teds to arise i the evet of selectig too may lags. This problem has bee overcome by this approach, as it applies a o-parametric correctio to deal with ay serial depedecies i the dataset. Table 4 presets the Phillips-Perro (1988) test results for each series. As expected, rejectio of the ull hypothesis of a uit root is reported whe the series is differeced. 8 [Please Isert Table 4 About Here] With the detectio of statioarity i the trasformed series ( Y ad X ), the followig VAR model, used as the system of equatios from equatio (1), is estimated: ARR UN t t t CON a t a CP a where, a 4 i i1,..., j1,..., k k 1,..., b ARR bjarr b ARR l l1,..., ti t j tk b ARR tl i i1,..., j j1,..., k k 1,..., c UN c UN l l1,..., ti c UN t j tk c UN tl i i1,..., j j1,..., k k 1,..., d CP l l1,..., ti d CP d CP t j tk d CP tl i i1,..., e CON j j1,..., k k 1,..., l l1,..., ti e CON e CON e CON ARR is the chage i total tourism arrivals to Greece ad a, b, c, d, e are coefficiets to be estimated withi the VAR system. Before estimatig equatio (8), we performed the Akaike (1974). Iformatio Criterio test to determie the optimal umber of lags () used i each u t j tk tl i, t u u u k, t l, t j, t (8) 8 Give the fidig of o-statioarity i the log series ad statioarity i the trasformed series, we tested whether there exist a log ru equilibrium relatioship betwee tourism flows to Greece ad the macroecoomic factors from the coutry of origi usig the Johase (1988) ad Johase ad Juselius (199) maximum likelihood approaches. The test results report overwhelmig evidece of multiple co-itegratig relatioships betwee tourism flows ad macroecoomic variables as implied by theory discussed earlier. Owig to the volume of results geerated, our fidigs are ot preseted, but are available upo request. 16

18 model system. We fid that lag 5 is the optimal umber of lags chose for each system. 9 Table 5 presets the VAR estimatios for the total umber of arrivals to Greece. Oly coefficiets that are statistically sigificat at the five per cet level are reported give the volume of output. I brief, the results suggest that chages i tourism demad are sesitive to chages i uemploymet (UN) i the origi coutry, ad tourists cost of livig relative to the coutry of origi (CP), but least sesitive to chages i the cosumer cofidece idicator (CON). The Uited Kigdom, beig oe of the importat sources of tourism flows, is most sesitive to chages i the tourists cost of livig. O the other had, Germay, with the secod highest umber of arrivals, is subject to chages i uemploymet ad tourists cost of livig. Tourism flows appear to be most sesitive to chages i macroecoomic factors i the Netherlads. [Please Isert Table 5 About Here] Impulse Respose Results A issue with the VAR system estimatios of Table 5 are the difficulties associated i iterpretig the coefficiets owig to complicatios arisig from correlatio feedbacks i additio to fluctuatios of estimatios at differet lags. Followig the estimatio of the VAR, the ext step is to cosider the system s respose to shocks origiatig from macroecoomic surprises ad the extet to which these shocks cotiue to have a impact o future tourism arrivals. To be cosistet with the oe year forecasts i tourism arrivals aalyzed earlier, we focus o impulse resposes for the ext twelve moths. I udertakig such a exercise, we could idetify the potetial source of risk to future tourists arrivals. To this effect, impulse resposes take ito accout the variatios i the velocity to which the effects of macroecoomic shocks are trasmitted, as well as the duratio ad rate of decay. I order to determie the robustess ad reliability of the respose, we compute cofidece bads usig Mote Carlo Simulatio that is simulated 5 times as a robustess test of the impulse respose. Large cofidece itervals aroud the impulse respose call ito questio the credibility of the measuremet iformatio, ad as such, the robustess of the 9 The results from the Akaike (1974) Iformatio Criterio test are ot preseted i this paper for brevity; however, they are available upo request. 17

19 respose. I Figure 2 we geerate time paths of impulse resposes o future arrivals i each destiatio to a oe stadard deviatio uemploymet shock i the coutry of origi. To esure cosistecy with the aual forecasts reported earlier, we geerate impulse resposes i future tourists arrivals twelve moths ahead. The results provide a clear picture o the impact of macroecoomic shocks o future tourism demad to Greece. Cosistet with the coclusios implied by previous studies, a oe stadard deviatio uemploymet shock origiatig from Frace ad the Netherlads has a immediate egative impact o future tourists arrivals although its duratio is very temporary. However, more iterestigly future tourists arrivals appear to react positively after a delay of four to six moths followig the uemploymet shock from the coutry of origi (Germay, Frace ad the Netherlads). Tourists arrivals are forecasted to retur to pre-shock levels aroud te moths after the shock. I the case of Germay, the impulse respose does become egative te moths after the uemploymet shock to cotiue beyod the twelve moth horizo period. The impulse respose results show that tourists cost of livig have a cosistet impact o future short term tourists arrivals with varyig degrees of velocity ad rate of decay. Closer ispectio of the results reveals that the respose is asymmetric ad differs i magitude across coutries of origi. For istace, future tourists arrivals respod to a shock i relative CPI i a positive maer subject to a lag of three to four moths. However, i some cases the iitial positive respose of future arrivals does become egative o average aroud five moths after the itroductio of the shock oly for arrivals to retur to their pre-shock levels two moths later. The positioig of the cofidece bads suggests that the impulse respose is reliable ad robust. The results o uemploymet ad tourist s cost of livig are i marked cotrast to the cosumer cofidece idicator. Shocks to cosumer cofidece i the origi coutry appear to have o impact o future tourism demad, although the positioig of the bads raises questios o the reliability of the respose. [Please Isert Figure 2 About Here] 18

20 5. FURTHER DISCUSSION OF THE RESULTS The combiatio of oe year ahead forecast ad measuremet of the duratio, at which future tourism arrivals retur to pre-shock levels i the twelve moth horizo, provide importat iformatio to policy makers for short term plaig ad budgetig purposes. We fid all forecastig models beig useful short term directioal forecastig tools as log as the models forecast a year o year icrease i arrivals. I combiig the impulse respose of a uemploymet shock ito the VAR system, the delayed positive respose i future tourist umbers is cosistet to the success rate i forecastig a icrease i tourists arrivals i Table 3. For Germay, beig the secod most importat source of tourism demad, the uemploymet shock appears to have the greatest positive respose. The fidig of a delayed positive respose could be attributed to tourists deemig Greece as a relatively cheap destiatio that people travel at a time whe ecoomic fudametals are deterioratig i the origi coutry. The immediate egative respose to a uemploymet shock, whilst cosistet with the coclusios iferred by the wage curve hypothesis ad the psychological aspects of uemploymet (Blachflower ((1996), (21)) ad Ah, Garcia ad Jimeo (24)), is a major source of dowside risk to forecastig future projectios of tourism arrivals. Despite this, whe combied with the delayed positive respose, dowside risk from a uemploymet shock appears to be limited. Aother source of risk to future tourism demad is the delayed egative respose recorded followig the itroductio of a shock to the tourists cost of livig. This source of risk teds to materialize six moths after the itroductio of the shock i Frace, Italy ad the Netherlads. However, the risk from this source appears to be limited due to the positive respose from shocks origiatig from the Uited Kigdom, ad Germay that are ot forecasted to be reversed at least i the short ru. The impulse respose aalysis reports evidece that shocks to cosumer cofidece have o bearig o future tourists arrivals. This fidig has major implicatios i the light of previous studies which have tested the theory surroudig the importace of the cosumer cofidece idicator. Theory implies that falls i cosumer cofidece from the viewpoit of icreased ucertaity of future ecoomic coditios or pessimism regardig future icome leads to a icrease i precautioary savigs ad hece, egative reactio i future tourism arrivals. 19

21 Accordig to our results, this hypothesis is rejected. Furthermore, the implicatio of our impulse respose results is to questio the empirical fidig that the predictive power of cosumptio ad household expediture models is improved by the iformatio cotaied i the cosumer cofidece idicator (Carroll et al. (1994), Bram ad Ludvigso (1998), ad Howrey (21)). Such a coclusio is reached by the iferece that a high readig o the cosumer cofidece idex should traslate ito icreased expediture ad a willigess to go o vacatio. This also appears to be rejected by our results. Take together, fidig limited dowside risk to future demad from shocks to tourists cost of livig ad uemploymet is idicative of customer loyalty irrespective of macroecoomic shocks i the tourists coutry of origi ad is cosistet with the results reported by Gari- Muoz (26) ad later, by Gari-Muoz ad Motero-Marti (27). Usig lagged cosumptio i the destiatio coutry as a depedet variable, both studies foud compellig evidece of high customer loyalty that is attributable to word of mouth effect i relatio to the Caary ad Balearic Islads. Overall, our aalysis provides a added dimesio, ot cosidered i prior studies that uforesee macroecoomic shocks from the origi coutry play a importat role i uderstadig potetial risks to future tourism arrivals. Our results highlight the eed for third parties to perform impulse respose aalysis o total tourists arrivals to Greece as the size of the respose, the rate of decay ad velocity ad the duratio of the respose, vary cosiderably. 6. IMPLICATIONS ON TOURIST ARRIVALS The above aalysis has a umber of implicatios. Table 1 highlights the lack of diversificatio i the sources of tourism i Greece sice Negative macroecoomic shocks to the Uited Kigdom ad Germay pose major risks to the Greek tourism idustry. Although the impulse respose results suggest that future tourism umbers may icrease followig macroecoomic shocks, there is dowside risk that could coicide with the peak times of the year. With the source of potetial risk cofied to coutries i the Europea Uio, the results suggest that the Greek authorities would be beefited by diversifyig the sources of tourism as 2

22 meas of reducig the risk of global macroecoomic shocks from the coutries of origi. 1 This ivolves attractig a greater umber of tourists from coutries outside the Europea Uio, such as Chia, Idia, Russia ad the Uited States. For istace, idicatios o arrivals from Chia are very positive as over the studied period there is a average aual growth rate of 19.22%. 11 As a result of the impact of macroecoomic shocks i the coutries of origi, a potetial source of risk to future tourists arrivals has bee idetified by our fidigs. Safeguardig the Greek tourism idustry requires short, medium ad log term plaig ad ivestmet o the brad itself, i additio to promotig Greece to coutries beyod the Europea Uio as a meas of diversifyig away global ecoomic risk. The results i this paper should provide importat isights that will be useful to short ad medium term plaig process. A similar exercise for loger term plaig will provide ifereces o where to target future additioal ivestmet i promotig the brad to ew coutries ad/or regios i additio to its traditioal sources of tourism. 7. CONCLUSION This paper opes a ew aveue of research i tourism forecastig by ivestigatig the impact of radom macroecoomic shocks o short term forecasts for tourist arrivals. This approach differs from the established literature of ivestigatig the demad fuctio for tourists. Our study addresses a major gap i the literature by forecastig tourism arrivals i Greece. Forecastig tourists arrivals comes i two levels; first, we utilize a array of forecastig models, firmly established i the literature, to geerate oe year out of sample forecasts. Secodly, we estimate a VAR system from which we itroduce radom macroecoomic shocks to gauge the reactio of future tourism arrivals i terms of the sig, magitude ad duratio. Macroecoomic factors used iclude uemploymet, tourists cost of livig, ad fially, the cosumer cofidece idicator with the latter two variables actig as proxies for the state of the ecoomy i the origi coutry. Udertakig such a exercise breaks ew groud by idetifyig the greatest source of the risk to future tourism demad, both i terms of the variable ad locatio. 1 This view is give added importace give additioal aalysis reported earlier i footote 4, where 66.3% of tourists over the 1977 to 29 period arrive from E.U. coutries. This cotrasts with 4.1% from Asia,.9% from Africa ad 1.1% from the Oceaia regio. 11 Source: Helleic Associatio of Travel ad Tourist Agecies (21). 21

23 To begi with, our forecastig results were mixed. Accordig to the prelimiary aalysis, the ARIMA (1, 1, 1) model outperforms other expoetial smoothig models as a directioal forecastig tool. This fidig is robust o a ever expadig estimatio period. However, the directioal forecastig performace of the ARIMA fails to traslate ito forecastig accuracy. Istead, the Holt s expoetial smoothig model with tred geerates the most accurate forecasts. I idetifyig the source of risk, we established that the source of tourists arrivals to Greece is udiversified ad heavily over-weighted towards the Uited Kigdom ad Germay. Despite this, the impulse respose aalysis yielded some iterestig results. Cosistet with the implicatios of previous studies o uemploymet, there is some evidece that shocks have a immediate temporary egative impact o future tourists arrivals that is reversed four to six moths after the shock. Furthermore, future tourists arrivals appear to react to shocks o the tourists cost of livig, regardless of the coutry of origi, subject to a lag of three to four moths. This cotrasts with the lack of respose i future tourism demad to a shock i cosumer cofidece idex i the origi coutry. Take together, despite the udiversified ature i the source of tourism flow to Greece, the impulse respose results suggest that dowside risk to future arrivals is limited, at least i the short term. The impulse respose results preseted i this study ope a ew dimesio i the type of macroecoomic factors used ad how these shocks impact o tourism demad i the future. As a result, our comprehesive evidece offers importat implicatios ad isights to policymakers ad tourist operators regardig future tourism demad. 22