Economic Growth and Business Cycle: The Case of Thailand

Transcription

1 Economic Growh and Business Cycle: The Case of Thailand Economic Growh and Business Cycle: The Case of Thailand Paravee Maneejuk *, Pahaira Paspipakul and Songsak Sriboonchia Faculy of Economics, Chiang Mai Universiy, Chiang Mai, Thailand Absrac: This sudy is aimed a analyzing business cycle in Thailand using a Markov-swiching Vecor Auoregressive approach wih a special concern abou srucural changes. Our specific ineress are o examine he cyclical movemens among economic aciviies, and o deec he nonlinear economic growh hrough he conex of business cycle. Empirical resuls show ha all parameers are saisically significan and differ across economic sages. These significan resuls can, in urn, suppor he aspecs of comovemen of differen economic aciviies over he business cycle and also he exisence of disinc relaionships among hese variables across differen economic regimes. Addiionally, he economic growh as measured by growh rae of real GDP, is affeced by changes in he economic aciviies in which magniude of effecs are differ across regimes. This resul reasonably suppors he persisence of nonlinear economic growh pah as being due o he flucuaion in economic aciviies associaed wih business cycle. Keywords: Business cycle; Economic growh; Comovemen; Markov-swiching model; Srucural change 1. INTRODUCTION There is no doub abou he imporance of economic growh for naions. As in any counry s economic policy, he focus is ofen on economic growh and on a way o make he economy grows susainably. The erm economic growh is usually quanified by he percen rae of increase in real gross domesic produc (real GDP), which is a measure of he values of goods and services produced wihin a counry irrespecive of who owns he facors of producion, so ha he economic growh refers o he increase in naional oupu. However, no economy can enjoy any growh a he same consan level hrough ime (Lucas, 2000). The economic growh as measured by he growh rae of real GDP moves eiher upward or downward around is long-erm growh rend due o he flucuaions ha economy experiences (Burn and Michell, 1946). To a large exen, he economy moves following a cerain paern. Tha is, as a counry develops, 263 Inernaional Journal of Economic Research

2 Paravee Maneejuk, Pahaira Paspipakul and Songsak Sriboonchia here exiss a period of expansion where he growh rae is above-average growh, coninues unil i reaches a maximum growh rae or peak. Bu afer he peak, he expansion will urn ino a period of recession and evenually reaches a minimum growh rae or rough. However, he rough will no exis eernally because he economy will somehow recover. As such, he rough is hen followed by anoher expansion, peak, conracion and rough again. This cyclical movemen is considered o be a consequence of flucuaions in he economy associaed wih business cycle, which is a basic for explaining he economic behavior. Acually, he business cycle is associaed wih many economic aciviies or economic organizaions regarding a counry s economic growh. One of he mos influenial works on business cycle, Burn and Michell (1946) indicaed ha he expansion phase occurs a abou he same ime in many economic aciviies, such as producion, consumpion, invesmen, and employmen, as illusraed by he increases in index numbers of economic aciviy, such as indexes of producion, price, naional aggregae of income, and employmen. These expansions are hen followed by similarly general recessions where hose indexes are decreased. However he conracions are followed laer by revivals, which in urns merge ino new expansion phase of he nex cycle. Similarly, he 1995 Nobel Prize winner in Economics, Rober Lucas emphasizes he coordinaion of aciviy among various economic secors in which heir oupus should move ogeher. This aspec of cyclical movemens of differen economic aciviies has become a key fac on business cycle measuremen ha researchers should ake ino accoun. Moivaed by his reasoning, his sudy aims o examine his nonlineariy in economic behavior associaed wih he business cycle. However, we do focus paricularly on he business cycle in developing counry. This is because, he economic flucuaions associaed wih he business cycle is based on imporan assumpion ha individuals mus have a ruly economic freedom. Wha do we mean by economic freedom? This can be viewed as a freedom o produce, rade, and consume any goods and services a he marke prices wihou he governmen inervenion. The governmen can play imporan role as a sabilizing force bu only when he economy severely conracs (Cooley and Presco, 1995). Therefore he major force ha drives he economy and essenially brings abou he counry s economic paern is considered o come from a public secor. Unlike a ruly free counry like he Unied Saes, he economic growh pah and business cycle in developing counries, paricularly Thailand are somehing else. The abiliy of he Thai people o underake economic acions or so-called he economic freedom is relaively less. Alhough, he Thai governmen does no ake conrol of all economic aciviies as happened in Nazi Germany or Sovie Russia, bu someimes he freedom of enerprise is limied by governmenal conrols, such as price conrols and wage conrols, which in urn make he business cycle fade ou. Moreover, he poliical insabiliy in Thailand ha occurs over he pas few decades could ruin he economy, undermine he invesmen, and hold oher economic aciviies far below he poenial levels. This raises a quesion wheher he business cycle in Thailand sill follows ha cerain paern suggesed by he radiional heory of business cycle. Are he economic sages as in he common business cycle sill persisence? Or hey fade ou? And imporanly, is here he consisen paern of comovemen among economic aciviies over he business cycle? To address hese quesions, we will employ an effecive ool for examining he business cycle characerisics ha will be described horoughly in Secion 2. Laer, all included variables as well as a model specificaion will be explained hrough Secion 3. In Secion 4, we will discuss abou esimaed Inernaional Journal of Economic Research 264

3 Economic Growh and Business Cycle: The Case of Thailand characerisics of he business cycle in Thailand as well as he impac of included variables on Thailand s economic growh ha may be inconsan across differen sages of he business cycle. 2. METHODOLOGY To analyze he business cycle, we consider an applicaion of Hamilon s model ha is a Markov-swiching vecor auoregressive (MS-VAR) model generalized by Krolzig (1997). For he saisical measuremen of economic flucuaions, he Markov-swiching model has become popular since Hamilon (1989) proposed he Markov-swiching auoregressive (MS-AR) model and applied o he business cycle measuremen in he US. Since hen, here have been a lo of exensions in which one of he influenial works is an applicaion of Krolzig (1997). He generalized he idea of MS-AR model o he MS-VAR model since he emphasized he co-movemen of imporan macroeconomic variables or he economic aciviies, and hen applied his echnique o measure he European business cycle. This ool has spread various sudies on economic flucuaions associaed wih business cycle. We also realize is poenial and employ his MS-VAR model o analyze he business cycle for our work Markov-swiching Vecor Auoregressive (MS-VAR) Model The MS-VAR model is well-known ha i provides a synhesis of he dynamic facor srucures and nonlinear approach which is useful for modeling flucuaions of he economy associaed wih business cycle. This model refers o an unobserved regime, denoed by S, driven by he Markov process, and he regime here can imply a phase of he business cycle. The MS-VAR model can be viewed as he VAR model wih regime changes. However, following he originaor of his model, Krozig (1997), ogeher wih Perlin (2015), he srucure of he MS-VAR model is as follows: p i S S 1, S i1 Y I A Y E. (1) From Eq.(1), Y is a ( K 1) vecor of endogenous variables or our ime series variables- ha follows he auoregressive form. i I S is a ( K 1) vecor of regime-dependen inerceps and A S is a ( K K) vecor of regime-dependen coefficiens of pas lags of Y, for lag i 1,.., p. E, S ~ NID(0, ) S in which S is referred o as he non-negaive variance of he residuals in each regime. In he general specificaion of MS-VAR model, all he parameers are condiional on he regime ( S ), meaning hey can vary across he regimes. Suppose we have M regimes in he MS-VAR model; herefore, Y I A Y... A Y I A Y... A Y 1 p 1 S 1, S 1 1 S 1 p S 11, 1 p M S 1, S 2 1 S M p S MM, (2) The regime variable S in he MS-VAR model, is generally governed by he firs order Markov process, which is defined by he ransiion probabiliy as follows: 265 Inernaional Journal of Economic Research

4 Paravee Maneejuk, Pahaira Paspipakul and Songsak Sriboonchia M p Pr( S j S i) and 1 p 1 i, j 1,..., M, (3) ij 1 where p ij is probabiliy of he ransiion from regime i o regime j. We can collec all he ransiion probabiliies for M regimes in he ransiion marix P j ij P p p p p p p p p p M M M1 M 2 MM. (4) Here we assume ha he Markov chain is saionary so he MS-VAR process is also saionary, hereby ergodic (Cavicchioli, 2015). Moreover, for he esimaion echnique implemened for MS-VAR model, we consider he maximum likelihood esimaor (MLE), which is designed o esimae he parameers of he model where he observed ime series depends on he unobserved or hidden sochasic variable S (Perlin, 2015). We consequenly need he likelihood funcion o be maximized. We assume a normal disribuion for he errors, so he log likelihood can be wrien as: i in which IS, AS, approximaed by K 1 ( ) log e(2 ) loge 2 2 1, (5) (Y ) (Y ) p 1 p i i IS AS Y 1 IS AS Y 1 2 i1 i1. Following Krolzig (1997), he expeced log likelihood funcion for Eq. (5) is M T L( ) ( S ) ( ) S Y, (6) S 1 1 where ( Y ) are he filered probabiliies, which in urn is calculaed by Hamilon s filer (Hamilon, S 1994). This specific filer of Hamilon can be deermined using he following algorihm. 1. Giving an iniial guess for ransiion marix P as in Eq. (4) 2. Updaing he ransiion probabiliies wih he pas informaion including he parameers in he model,, and P for calculaing he likelihood funcion in each regime a ime as in Eq. (6). S The probabiliy of each regime is updaed by he following formula: ( Y ) Pr( S m ) S S M m1 f (Y S m, )Pr( S m ) m S 1 S 1 f (Y S m, ) Pr( S m ) m S 1 S 1. (7) Inernaional Journal of Economic Research 266

5 Economic Growh and Business Cycle: The Case of Thailand Noe ha we assume M regimes for his MS-VAR model in which m 1,.., M. The erm fm(y S m, S ) is referred as he likelihood funcion of regime m and Pr( S ) 1 m S is referredred 1 as he filered probabiliies a ime Ieraing Sep 1 and Sep 2 for = 1,..., T. However, for empirical applicaions, i migh be more helpful o use a model where only some parameers are condiional on he regime of he Markov chain, while he oher parameers are regimeinvarian. Krolzig (1997) emphasized his and invened a common noaion o provide simpliciy in expressing hose models. Following Krolzig (1997), we consider paricularly inerceps (I), auoregressive parameers (A), and heeroskedasiciy (H) o vary across differen regimes. And hence, we have go eigh alernaive specificaions of he MS-VAR model as shown in Table 1. In a case ha all he parameers are condiional on he regime, Krolzig considered his as a general specificaion of he MS-VAR model. However, o selec he bes one among hese compeing models, he Akaike informaion crierion (AIC) is used, in which he model wih lowes value of AIC is preferred. Table 1 Types of MS-VAR Models I varying I invarian A H invarian MSI-VAR VAR (linear) invarian H varying MSIH-VAR MSH-VAR A H invarian MSIA-VAR MSA-VAR varying H varying MSIAH-VAR MSAH-VAR Source: Krolzig (1997) Imporanly, we have o make clear ha his sudy considers, in paricular, MSI-VAR models (wih inercep) raher han MSM-VAR models (wih mean) because he MSI-VAR models adjus more smoohly afer he regime shif. In addiion, he esimaion procedure of he MSM-VAR models may face o a problem of nonlinear opimizaion if he endogenous variable does no depend on he unobserved variable (S ) (Krolzig, 2003) Likelihood Raio Tes This par is conduced since we concern abou a number of sages really occur in or fi o he Real-world, paricularly Thailand s, business cycle. As such, his sudy employs a likelihood raio (LR) es o examine number of regimes for he MS-VAR model. The LR es is a saisical es used for comparing he goodness of fi of wo models: null and alernaive models. Roughly, he LR saisic is simply wice he difference in he log likelihoods, which can be illusraed as follows: LR 2( L ( ) L ( )). (8) 1 S 0 From Eq.(8), L ( ) 0 is he log likelihood for he null model and L 1 ( ) 267 Inernaional Journal of Economic Research S is he log likelihood for he alernaive model. In case, he null hypohesis is he model wih no regime swiching, or equivalenly a

6 Paravee Maneejuk, Pahaira Paspipakul and Songsak Sriboonchia linear VAR model, while he alernaive hypohesis is he one wih regime swiching. The rejecion is equivalen o he rejecion of linear VAR in favor of he MS-VAR model. The probabiliy disribuion of he es saisic is approximaely a chi-squared disribuion wih degrees of freedom equal o df df L1 ( ) L0 (. The p-value is compued by 2 P( ) df 1 ( ) L ) df LR L0 (. For insance, if we consider one ) regime agains wo regimes for he MS-VAR model, we hen have and T L0 ( ) ( ), (9) 1 2 T L( ) ( ) ( S Y ) (10) S DATA AND MODEL SPECIFICATION In his secion, we presen he daase and inroduce he considerable variables o idenify he business cycle. To measure he cycle, he basic variable is real GDP since i is he bes measure o rack a counry s economic performance, and wih a growh rae form he real GDP can indicae he growh in economic oupu. Alhough he economic movemen in regards o business cycle is based on he oupu, we also consider he coordinaion of aciviy among various economic secors as highlighed by Rober Lucas. In paricular, we consider he growhs of number of unemployed people (UNP), indusrial producion (IPI), reail sales (Reail), and gross disposable household income (Income) over he business cycle. The variables we use are of quarerly frequency, spanning from 1995:Q1 o 2016:Q4, wih he descripion is given in Table 2. These variables specified in he MS-VAR model can ake he form as y v y s y u q q S s GDP UNP v ( s ) GDP GDP v ( s ) UNP UNP GDP 1 q UNP 1 q IPI v ( s ) IPI IPI IPI 1 q Reail v ( s ) Reail Reail Income v ( s ) Income Income Reail 1 q Income 1 q. GDP GDP 1, s q, s UNP UNP 1, s q, s IPI IPI 1, s q, s Reail Reail 1, s q, s Income Income 1, s q, s u u u u GDP UNP u IPI Reail Income. The model comprises 5 variables ha are GDP, UNP, IPI, Reail, and Income modeled as MS-VAR in which s represens regime or sae and q is he lag erm. To decide he number of regimes, we follow he radiional heory of business cycle. Tha is, he cycle has basically 4 saes namely expansion, recession, depression, and recovery, herefore 1,..., 4 s. However, he exac number of regimes occur in his experimen have o be invesigaed by he LR es, so ha number of sages may no follow he heory; I Inernaional Journal of Economic Research 268

7 Economic Growh and Business Cycle: The Case of Thailand can be one, wo, or hree regimes over he cycle. Bu prior o he esing regime, he characerisics of each sage can briefly explained as follows. In he expansion phase, he demand for goods and services increase. There are he increases in sales and profi which in urn induce companies o inves in more producion faciliies and invenories. Moreover, his srong demand also pushes he need for more workers, which spurs an increase in employmen levels. Mopping up of resources in he economy leaves no room for expansion; inpus become expensive, signaling ha he economy is a is peak. In he recession phase, he economy slows down, and he level of sales and producion orders sar declining. Producion faciliies become underuilized, and companies respond by reducing he work rae. Workers who had been hired on casual basis are laid off, and his reduces heir disposable income. Idle capaciy of producion faciliies reduces he oupu, and mos companies are forced o reduce prices of producs in an aemp o increase demand. Table 2 Descripive Saisic Reail IPI Income GDP UNP Mean Median Maximum Minimum Sd. Dev Skewness Kurosis Jarque-Bera *** *** *** *** *** ADF-es *** *** *** *** *** Noe: *** denoe rejecions of he null hypohesis a 1% significance levels. In he depression phase, a proraced period of recession ushers in a depression. Demand for producs and services decrease, forcing companies o shu down some producion faciliies. Closing of producion means a company canno susain is work force, and i is forced o lay hem off. Unemploymen leaves he consumers wih very lile disposable income needed o buy necessiies. The gross domesic producion declines and sandard of living of he people also declines. In he recovery phase, his sage is characerized by an increase in consumers confidence of he marke. The bank lending raes are low, and companies can afford o finance projecs. There is an increase in produciviy due o he increased aggregae demand in he economy. Increase in producion allows companies o sar employing, which in urn, increases he income of consumers who can now afford o purchase goods. Profi margins of companies sars rising, and he gross domesic produc also sar o increase. Addiionally, o avoid he phenomenon of false regression caused by he regression analysis of nonsaionary ime series, he variables should be aken as Augmened Dickey Fuller (ADF) uni roo es 269 Inernaional Journal of Economic Research

8 Paravee Maneejuk, Pahaira Paspipakul and Songsak Sriboonchia before esimaing he MS-VAR model. We do no provide he resul of he ADF es, bu i suggess ha he null hypohesis of he uni roo es can be rejeced and ha all series are saionary a he 1%, 5%, and 10% confidence levels. This means ha we can use hese variables as endogenous variables in he MS-VAR model for he purpose of esimaing he business cycle characerisics. 4. EMPIRICAL RESULT In his secion, we presen he esimaed resuls sequenially saring wih lag lengh selecion for VAR model followed by he likelihood raio es for deermining he number of sages or regimes, and selecing appropriae srucure for he MS-VAR model, respecively. Afer hese core ingrediens for he MS-VAR model wih he srucural change are defined, we are hen able o esimae Thailand s business cycle characerisics in which he esimaed parameers a differen sages and he probabiliies of saying in, paricularly smoohed and filered probabiliies, will be shown a he end of his secion Lag Lengh Selecion We begin his secion by lag lengh selecion for deermining he vecor auoregressive lag lengh. We employ he AIC and BIC o choose an appropriae lag erm for he VAR model. Table 3 shows he AIC and BIC values of each model a lags 1 o 5. We found ha he appropriae lag for he VAR model using his daa se, is lag 1 since i has he lowes values of AIC and BIC. Table 3 Lag Lengh Selecion Lag AIC BIC Source: Calculaion. VAR 4.2. Likelihood Raio Tes for Deermining he Number of Markov Regimes The likelihood raio (LR) es is conduced in his par o deermine he number of Markov regimes under his daa se, which reflec he number of sages in he business cycle. Following formula for he LR es saisic, as shown in Eq. (8), i is measured by wice difference in he log likelihood values beween null model and alernaive model. In his sudy, boh null and alernaive models are in he form of MS-VAR model bu wih differen number of regimes. For example, he firs roll of Table 4 is he resul of LR es in which he null model is he MS-VAR model wih 1 regime his paricular model is equivalen o he linear VAR model- and he alernaive model is he MS-VAR model wih 2 regimes. L(Ho) sands for log likelihood of he null model and L(Ha) is of he alernaive model. LR es saisic measured following ha formula is shown in he fourh column. Inernaional Journal of Economic Research 270

9 Economic Growh and Business Cycle: The Case of Thailand Table 4 Resuls of he LR Tes Model L(Ho) L(Ha) LR es P-value 1 VS VS VS Source: Calculaion. Noe: If we compue he LR saisic for he hird es hrough he common formula as in Eq.(8), we will ge a negaive value. Hence, his LR saisic is paricularly measured by LR 2( L0( s ) L1( s )) (Paspipakul e al., 2016). Table 4 shows he resuls of hree LR ess. Resul of he firs LR es indicaes ha here exiss a rejecion of he MS(1)-VAR model in favor of he MS(2)-VAR model. However, he second LR es also exiss o rejec he MS(1)-VAR model, bu, in favor of he MS(3)-VAR model. As he model wih 2 and 3 regimes are boh preferable o he linear VAR model, we hen perform he hird LR es o sele his issue. I is found ha we canno rejec he null model which is he MS(2)-VAR a 10% significance level, so he appropriae model for his daa se is considered o be he MS(2)-VAR model Selecing Appropriae Srucure for he MS-VAR Model This par is abou srucural selecion based on MS(2)-VAR(1) model since we concern abou srucural change o measure in deail he characerisics of Thailand s business cycle. In paricular, he srucural change refers o changes in inerceps (I), auoregressive parameers (A), and heeroskedasiciy (H); herefore, we have eigh differen ypes of models as shown in Table 1 and here Table 5 provides he empirical resul. Table 5 MS Srucural Specificaion based on AIC MS Specificaion I varying I invarian A invarian H invarian A invarian H varying A varying H invarian A varying H varying Source: Calculaion. Wih a given se of models, we hen employ he AIC o choose he bes model among all candidaes. Table 5 shows relaive qualiy of each model hrough he AIC values. I is found ha he MS(2)-VAR(1) model wih varying A, I, and H has he lowes value of AIC ha is equal o Hence, he appropriae model for explaining he characerisics of Thailand s business cycle under his daa se, is considered o be he MSIAH(2)-VAR(1) Esimaes of he MS-VAR Model wih he Srucural Change The esimaes of he MSIAH(2)-VAR(1) model is illusraed in Table 6. This empirical resuls indicae ha he nonlineariy in economic growh as well as he flucuaion of economic aciviies associaed 271 Inernaional Journal of Economic Research

10 Paravee Maneejuk, Pahaira Paspipakul and Songsak Sriboonchia wih he business cycle, appear in erms of changes in he whole process of economic growh. The economic aciviies, i.e. reail sales, indusrial producion, gross disposable household income, (un)employmen, as well as he economic growh, behave differenly across regimes. As shown in Table 6, he esimaed parameers of each variable are significanly differen across regimes, in which some economic aciviies change in he same direcion as he change of real GDP growh, while some of hem move inversely. In erms of he modelling srucural change, i is found ha he MSIAH-VAR model is preferred. This means changes in he whole process of economic growh can be explained by changes in he srucures i.e. inerceps, auoregressive parameers, and heeroskedasiciy across differen regimes. Table 6 Esimaes of MSIAH(2)-VAR(1) Reail IPI Income GDP UNP Regime-dependen Inerceps Regime Regime Regime-dependen Auoregressive Parameers a Lag 1 Regime 1 Reail IPI Income GDP UNP Regime 2 Reail IPI Income GDP UNP Log-likelihood AIC BIC Transiion Prob. p 1 p 2 Duraion Observaion Regime Regime Source: Calculaion. Noe: Surprisingly, all esimaed parameers are saisically significan a he 1% significance levels. Table 6 shows ha inerceps of some growh variables in regime 1, says reail sales, indusrial producion, and real GDP, are greaer han he ones in regime 2, while inercep of growh rae of gross disposable household income in regime 1 is less han he one in regime 2. Addiionally, inercep of growh rae of unemploymen in regime 1 is much greaer han he one in regime 2. Due o hese disinc resuls, we canno conclude confidenly ha which regime is expansion (or recession). However, he probabiliies shown in Figure 1 jus provide helpful informaion. Basically, he probabiliies show he chance of saying in some cerain regime. Figure 1 is illusraing he probabiliies of saying in regime 1 Inernaional Journal of Economic Research 272

11 Economic Growh and Business Cycle: The Case of Thailand in which here are wo periods of ime ha he cycle is dragged down. Le s look a he hisorical informaion relaed o his figure. I is well known ha Thailand and many counries in Eas Asia wen hrough he grea financial crisis in , and again in due o he global economic crisis. We emphasize ha hese wo crises cause Thailand s business cycle o go down o he recession during 1997 coninuing o 1999 and again during Hence, regime 1 is supposed o be he expansion since Figure 1 shows ha he probabiliies of saying in his regime is almos zero during ha period of ime. Consequenly, regime 2 is supposed o be he recession. To a large exen, Table 6 shows ha in each cycle he lengh of saying in he expansion is approximaely 18 quarers. The probabiliy of changing from expansion (regime1) o recession (regime2) is equal o 0.06 bu he chance of remaining in he expansion is equal o On he oher hand, i is abou 5 quarers for he lengh of saying in he recession. The probabiliy ha Thailand s business cycle will say in he recession is abou 0.75 while he res 0.35 is he probabiliy of changing from he recession o he expansion. The esimaed auoregressive parameers are condiional on he regime. For insance during he expansion phase, a 1% increase in he growh rae of indusrial producion (IPI) leads o a rise in he growh of reail sales (Reail), indusrial producion (IPI) iself, disposable household income (Income), and he real GDP (GDP) abou 0.034%, 0.249%, %, and 0.265%, respecively. In addiion, a rise in indusrial producion jus 1% can cause a decrease in he growh rae of unemploymen (UNP) by 0.029%. On he conrary, he increase in he growh rae of indusrial producion during he recession brings abou he negaive impacs mosly on oher secors. I causes a decrease in reail sales, indusrial producion iself, and he real GDP abou 0.731%, 0.414%, and 0.947%, respecively. Alhough i sill causes he reducion in unemploymen, bu he rae is double lower; ha is abou 0.029% in he expansion bu 0.014% in he recession. As also presened in Table 6, he changes in considerable economic aciviies resul in disinc impac on economic growh measured by he real GDP. In he expansion, increases in he growh raes of hose economic aciviies by 1% lead o an increase in economic growh, for insance, by 0.313% (by reail sales) and 0.265% (by indusrial producion). However, hese impacs become negaive during he recession phase. This indicaes ha he impacs of changes in he economic aciviies on he real GDP growh are unequal hrough regimes, which in urn suppor he persisence of nonlinear economic growh pah as being due o he flucuaion in economic aciviies associaed wih business cycle. Figure 1: Filered and Smoohed Probabiliies of Expansion Phase (Regime 1) 273 Inernaional Journal of Economic Research

12 Paravee Maneejuk, Pahaira Paspipakul and Songsak Sriboonchia 5. CONCLUSION This sudy is on he analysis of business cycle paricularly in Thailand using a Markov-swiching vecor auoregressive approach (MS-VAR) wih srucural changes. This sudy yields wo imporan purposes: o examine he simulaneous cyclical movemens among economic variables, and o deec he nonlinear economic growh hrough he conex of business cycle. This sudy builds he business cycle by using he real GDP, unemploymen, indusrial producion, reail sales, and gross disposable household income. As a sequence of modelling, The LR es is used for deermining he number of Markov regimes, which refers o he sages of business cycle, and he AIC is used for boh lag lengh and srucural selecions. Hence, his sudy came up wih he MSIAH(2)-VAR(1) model o reflec he characerisic of Thailand s business cycle. The overall resuls show ha he relaionships among economic aciviies during he recession phase end o be negaive, bu are more likely o be posiive during he expansion. These significan resuls can reasonably suppor he aspecs of comovemen of differen economic aciviies over he business cycle as suggesed by many radiional heories of business cycle, and also he exisence of disinc relaionships among hese variables across differen economic regimes. Moreover, he impacs of changes in he economic aciviies on he real GDP growh are unequal hrough regimes. This suppors he persisence of nonlinear economic growh pah as being due o he flucuaion in economic aciviies associaed wih business cycle. To be more specific, he smoohed probabiliies show ha Thailand s business cycle mosly akes place in he expansion, excep for he periods of grea depression during and he global economic crisis during REFERENCES Burns, A. F., & Michell, W. C. (1946), Measuring Business Cycle. New York: NaionalBureau of Economic Research. Rerieved Sepember 4, 2013, from hp://papers.nber.org/books/burn46-1. Cavicchioli, M. (2015), Likelihood Raio Tes and Informaion Crieria for Markov Swiching Var Models: An Applicaion o he Ialian Macroeconomy. Ialian Economic Journal, 1(3), Cooley, T. F., & Presco, E. C. (1995), Economic growh and business cycles. Froniers of business cycle research, 1. Hamilon, J. D. (1989), A new approach o he economic analysis of nonsaionary ime series and he business cycle. Economerica: Journal of he Economeric Sociey, Hamilon, J.D. (1994), Time series analysis. Princeon Universiy Press, Princeon. Krolzig, H-M. (1997), Markov swiching vecor auoregressions: modelling, saisical inference and applicaion o business cycle analysis. Berlin: Springer. Rerieved Sepember 4, 2013, from hp://down.cene.org.cn/upfile/8/ pdf Krolzig, H. M. (2003, July), Consrucing urning poin chronologies wih Markov-swiching vecor auoregressive models: he euro-zone business cycle. In Eurosa Colloquium on Modern Tools for Business Cycle Analysis. Lucas, R. E. (2000), Some macroeconomics for he 21s cenury. The Journal of Economic Perspecives, 14(1), Paspipakul, P., Yamaka, W., & Sriboonchia, S. (2016), Effec of Quaniaive Easing on ASEAN-5 Financial Markes. In Causal Inference in Economerics (pp ). Springer Inernaional Publishing. Perlin, M. (2015), MS_Regress-he MATLAB package for Markov regime swiching models. Inernaional Journal of Economic Research 274