Being recognised as the third largest economy in South East Asia, Malaysia has gone through several stages of structural changes, witnessing recessions, crises, resilience, rapid growth, and development, ever since its independence from the United Kingdom in 1957. The country is now classified as an upper-middleincome country by the World Bank with per capita income of US $10760 as of 2014 year-end, and an average annual growth of 7% over and above Thailand and Indonesia. Nevertheless, the economy has been highly dependent on favourable external terms of trade to stimulate its growth. This study investigated the Malaysian economic behaviour triggered from its interdependency on its major trading partners, specifically the USA and China, over a 25-year period from 1990 to 2014. The strand of literature on economic forecasting suggested development of several sophisticated econometric techniques, including survey forecasts and reduced-form equations for an appropriate theory-based economic behaviour modelling and determination, as the mainstream practices applied in practical policy decision making by several policy institutions and banks around the world. However, there is not a unanimous clue as to whether these forecasting techniques are opportune or promising. It is further noted that imposing simple theory-based cointegrating restriction can lead to more reliable outputs when dealing with long-forecast horizons(Giacomini, 2014). Hence this study applied a simple modelling approach using an appropriate and robust econometric methodology to revisit the issue of Malaysian economic forecasting concerning its relationship with its trading partners. As a pre-estimation, this study implemented a test on structural break of the Malaysian GDP and IPI. A review of behaviour of the series revealed the fact that there is evidence of multiple significant changes or breakpoints being identified from a recent test developed by Bai and Perron (2003). This study further implemented an ARDL bound test of Pesaran, Shin, and Smith (2001) to reexamine the effect of gross domestic products (GDP) and industrial production index (IPI) as two fundamental variables representing Malaysian economic growth, earnings, and income. The aim was to establish a simple, parsimonious modeling technique with all the timeseries properties being fulfilled and verified. New findings were reported pertaining to high dependency of the Malaysian economy on its trading partners, with its behavior being determined within a year-prior-impact. Also identified was new evidence of a long-run relationship between Malaysian economic factors and those of its trading partners. This would suggest crucial implications corroborating to the prediction of the Malaysian economic behaviour within a year ahead of its occurrence from the information obtained on economic factors of its closely linked trading partners. The test procedure applied in this study incorporated a fully specified model with optimum lags - after performing a large number of iterations of regression equations to be verified from high R-Squared statistics and the absence of serial correlation. To the best of our knowledge, this is by far the most precise estimation method for identification of appropriate lag parameters of the explanatory variables.

In the remainder of this paper, in Section 2, a brief review of the literature on this topic is presented, before describing the hypothesis and model development in Section 3. Findings are presented and discussed in Section 4 and the conclusion is in Section 5. 2. Literature Review Numerous studies investigated evidence of macroeconomic and business cycle interdependency and/or co-movements of trading partners. This strand of literature suggested significant implications arising from cross-country correlation analyses using parsimonious principles of variable and model specifications. In other words, many scholars found that inclusion of additional explanatory variables in a model is found to have no marginal effect on the estimation output (Blonigen, Piger, & Sly, 2014). The very preliminary study of such type was conducted by Frankel and Rose (1998) to demonstrate a significant strong correlation among trading partners being determined solely from their real GDP behaviour. Similarly, Clark and Van Wincoop (2001) documented evidence of significant correlations between US regions. A rather contrary relationship was found to hold true among European nations, as in Bayoumi and Eichengreen (1993), and Fatás (1997). Such mixed findings are partly associated with different methodological applications. The so-called univariate and multivariate approaches for economic determination are applied very commonly. In spite of several sophisticated statistical and econometric applications to economic forecasting, the simple univariate and bivariate approaches are found to be often more reliable benchmarks, against which to assess the performance of those multivariate models (Stock, 2001). There are also several studies examining the degree of interdependence of financial markets and trade sectors of different countries. Yang, Cai, Zhang, and Hamori (2016) documented evidence of significant interdependence between exchange rate markets of European zone and a rather modest contagion between Japanese Yen and British Pound, especially during the crisis times. Similarly, a recent study by Shahzad, Kumar, Ali, and Ameer (2016) reported new findings on the interdependence of Greek stock market with that of European zone. Daniel, Surugiu, and Surugiu (2015) elaborated on the effect of international trade activities in transport and ICT sectors on development and growth of economies that are substantially interdependent on each other. The very first version of univariate modelling is known as the exponential weighted determined by moving average (EWMA) method, which uses a parameter the forecaster or estimated by nonlinear least squares using historical data (Stock & Watson, 2001). The model however is not very accurate or reliable, given its broad range of applications in practice. The auto-regressive moving average (ARMA) model however represents more accuracy and reliability in presence of serially uncorrelated residuals and imposition of lag polynomials of orders p and q for a univariate regression equation, as:

accuracy and reliability in presence of serially uncorrelated residuals and imposition of lag polynomials of orders p and q for a univariate regression equat ls of orders p80and q for a univariate regression The International Journalequation, of Banking andas: Finance, Vol. 12, No. 2, 2016: 77-97 or or or𝑦𝑦𝑡𝑡 = 𝜑𝜑1 𝑦𝑦𝑡𝑡−1 + 𝜑𝜑1 𝑦𝑦𝑡𝑡−2 + ⋯ + 𝜑𝜑𝑝𝑝 𝑦𝑦𝑡𝑡−𝑝𝑝 + 𝑢𝑢𝑡𝑡 + 𝜗𝜗1 𝑢𝑢𝑡𝑡−1 + 𝜗𝜗2 𝑢𝑢𝑡𝑡−2 + ⋯ + 𝜗𝜗

+ ⋯ + 𝜑𝜑𝑝𝑝 𝑦𝑦𝑡𝑡−𝑝𝑝 + 𝑢𝑢𝑡𝑡 + 𝜗𝜗1 𝑢𝑢𝑡𝑡−1 + 𝜗𝜗2 𝑢𝑢𝑡𝑡−2 + ⋯ + 𝜗𝜗𝑞𝑞 𝑢𝑢𝑡𝑡−𝑞𝑞 or where 𝑦𝑦𝑡𝑡 is determined by (i) its own values 𝑦𝑦𝑡𝑡−𝑖𝑖 in th where 𝑦𝑦𝑡𝑡 is determined by (i) its own values 𝑦𝑦𝑡𝑡−𝑖𝑖 in the preceding period determined own 𝑦𝑦𝑡𝑡−𝑖𝑖 inin the restricted to where by (i) its values thepreceding preceding where y𝑦𝑦t𝑡𝑡isisdetermined 𝑖𝑖 is value |𝜑𝜑| < 1period values 𝑦𝑦𝑡𝑡−𝑖𝑖 in the preceding period and 𝜑𝜑𝑖𝑖 is restricted to own be values stationary, provided ,period andand (ii) 𝜑𝜑the of the own values 𝑦𝑦𝑡𝑡−𝑖𝑖 in the preceding period restricted to beof reeceding 𝑦𝑦𝑡𝑡 is determined period and and by 𝜑𝜑𝑖𝑖 (i) is restricted totobebe 𝑖𝑖 is (ii) |𝜑𝜑| <and stationary, provided 1 , ,𝜑𝜑 and (ii) the the value value of the the immediate past er is its restricted stationary, provided and stationary, provided < 1 , and (ii)invertible the value process, of the immediate past error 𝑢𝑢𝑡𝑡−𝑗𝑗 is subject to the value of theimmediate immediate past error error 𝑢𝑢𝑡𝑡−𝑗𝑗 |𝜑𝜑| subject to past is subject to an an invertible process,provided provided |𝜗𝜗| < 1.. Hence, an importan |𝜑𝜑| < onary, 1 , isand (ii) the value of the immediate past error|𝜗𝜗| 𝑢𝑢𝑡𝑡−𝑗𝑗 is Hence, subjectantoimportant an subject to an mediateprovided past error 𝑢𝑢𝑡𝑡−𝑗𝑗 invertible process, provided < 1. Hence, an important condition for ARMA to be met is stationarity of time series condition for ARM |𝜗𝜗| < 1. Hence, invertible process, important for ARMA to in bemost met iseconomic stationa nce, an importantwhich condition forthe ARMA to most beprovided meteconomic is stationarity not the case of an time series deficiency 𝑦𝑦condition 𝑡𝑡 which isof ismet not case in time series. Another ARMA rtible process, provided <is1.stationarity Hence, an important ARMA to be ndition for ARMA to be|𝜗𝜗| which is not themet caseisinstationarity most economic time series. Anothe of timecondition series 𝑦𝑦𝑡𝑡for is biasedof estimation presence ofthelarge roots ofeconomic moving time average processes is case most Another deficiency ARMA time series deficiency 𝑦𝑦in𝑡𝑡 which estimation in series. presence of large roots of of moving n most economic time series. Another ofnot ARMA is in biased (Davis & Dunsmuir, 1996; Stock, 1994). A study by Meese and Geweke (1984) which is not the case in most economic time series. Another deficiency ARMA is me series 𝑦𝑦 me series. Another deficiency of ARMA is 𝑡𝑡 biased estimation in presence of large ofroots of moving average processes (D biased estimation presence of1996; large of moving (Davis & Dunsmuir, the special of ARMA usingroots purely autoregressive models Stock, 1994).Aaverage study byprocesses Meese with and Geweke (1984)found oots of moving found averagethat processes (Daviscases &in Dunsmuir, ed presence ofplarge roots of moving average (Davis Dunsmuir, 1996; by that the special cas ageestimation processesin(Davis & Dunsmuir, 1996; Stock, 1994).Aprocesses study by Meese & and Geweke (1984)found lag orders (AR(p)) are more accurate, when optimum lags are identified 1994).A by Meese and Geweke (1984)foundmodels that thewith special cases of ARMA are using pu autoregressive lag using orders p (AR(p)) more weke (1984)foundinformation that theStock, special cases (e.g., ofstudy ARMA using purely criteria AIC and BIC). It is therefore suggested that 1994).A and Geweke (1984)found that the models special cases of ARMA purely atk, the specialstudy casesbyofMeese ARMA using purely autoregressive with lag orders pusing (AR(p)) are more accurate, when optim OLS with distributedmodels lags would result in more consistent outputs when from what lags are identified with lag orders are more accurate, information criteria (e.g., AIC optimum and BIC). It is therefore s AR(p)) are more accurate,autoregressive when optimum lags are identified by p (AR(p)) is obtained when testing ARMA models. The question to an appropriate regressive withlags lagare orders p (AR(p)) more accurate, when(e.g., optimum identified by lagsuggested that using urate, whenmodels optimum identified by are information criteria AIC lags and are BIC). It is therefore information criteria AIC andlags BIC). Itwould is therefore suggested that using OLS with distributed specification leading towith a (e.g., fully-specified regression equation however is still result in more consistent outputs from what is It is therefore suggested that using OLS distributed mation criteria is therefore suggested that using OLSeconometric withoutputs distributed lags ested that using(e.g., OLS AIC with and distributed result more consistent from what is obtained when test unsolved; weBIC). tendItlags to solvewould this issue inusing latest techniques. would in ARMA moreforecasting consistent outputs from use what is appropriate obtained testing ARMA models. question tomodels an lag specification leading to a s from what isOther obtained when result testing models. The (notably univariate) approaches to bewhen estimated ld result in more outputs from is obtained when testing ARMA models. The to a fully-specified regre ained when testingconsistent ARMA models. The what question to an appropriate lag specification leading on a nonlinear basis, with relatively substantial estimation errors and technical question to an appropriate lag specification leading to a we fully-specified regression howeve still unsolved; tend toare solve this issue equation using latest econ ion leading to a problems, fully-specified regression equation however is over although developed time. this Examples smooth tion to an appropriate lag specification leading to aunsolved; fully-specified regression is econometric techniques. y-specified regression equation howeverextensively is still we tend to solve equation issue however using latest transitiontechniques. auto-regressions by Teräsvirta (1994), and artificial neural unsolved; we (notably tend(STAR) to solve this issue using latest econometric techniques. Other univari forecasting approaches use models to be(notably estimated on using latest econometric still Other univariate) unsolved; we tend to solve this issue latestforecasting econometric techniques. Other (notably univariate) networks (NN) byusing McCulloch and Pitts (1943), use which are to widely appliedonto a nonlinear basis, w etric techniques. Other (notably univariate) approaches models be estimated forecasting approaches usesubstantial models of to these be estimation estimated on a nonlinear basis, with relatively substan time series. The fundamentals two methods are with errors andidentical technical problems, although ext be estimated on economic a nonlinear basis, with relatively casting models substantial to be estimatedestimation on a nonlinear basis,technical with relatively substantial onlinearapproaches basis, with relatively errors inputs and problems, although extensively developed the use linear autoregressive approach, in which are lagged values to estimate estimation errors and technical problems, although extensively developed (STAR) over time. Examples transition auto-regressions by Teräsvirta ms, although extensively developed over time. Examples are usessmooth future values as outputs. However, STAR a nonlinear function of past mation errors and technical problems, although extensively developed over time. Examples are data(1994), and artificial vely developed over time. Examples are smooth transition auto-regressions (STAR) by Teräsvirta that switches between lag polynomial (STAR) regimes,byand NN uses an and index model smooth transition Teräsvirta artificial networks (NNt McCulloch and(1994), Pitts (1943), which neural are widely applied R) by Teräsvirta (1994), and artificial neuralauto-regressions networks (NN) by oth transition auto-regressions (STAR) by Teräsvirta (1994), and artificial neural networks (NN) by formulation with nonlinear transformation. Further explanation about theseto economic time se 94), and artificial neural networks (NN) by McCulloch and Pitts (1943), which are widely applied McCulloch and Pitts (1943), which widely to economic time series. Thelinear fundamental approaches can be found Granger and Terasvirta (1993), andareSwanson and these applied two methods identical with the autoreg widely applied to economic time series. Theinfundamentals ofare Culloch Pitts (1943), which are widely economic Thewith fundamentals conomicand time series. The fundamentals of applied thesetotwo methodstime are series. identical the linear of autoregressive approach, in White (1997), respectively. these two inmethods are identical with the values linear to autoregressive approach, which inputs are ST lag estimateisfuture values asin outputs. However, he linear autoregressive The approach, which are lagged literature on inputs economic time-series forecasting alsolagged extended eive two methodsinarewhich identical with linear autoregressive approach, whichasinputs areHowever, approach, inputs arethe lagged values to estimate futureinvalues outputs. STAR uses a nonlinear to include multivariate approaches likely to However, improve STAR forecasting precision.function of past data to estimate future values outputs. usespolynomial a nonlinear switches between lag regimes, and NN use ts. However, STAR uses values a nonlinear function of past dataasthat There are four broad categories to multivariate forecasting, namely es to estimate future values as outputs. However, STAR uses a nonlinear function of past data that uses a nonlinear function of past data that switches between lag polynomial regimes, andstructural NN uses an index model fo switches between lag polynomial regimes,transformation. and NN uses an nonlinear indexexplanation model formulation withapproac nonli models, small linearnonlinear time-series models, small timeFurther about these es, and NN useseconometric an index model formulation with ches between polynomial regimes, NNtransformation. uses an index model nonlinear index model lag formulation with nonlinear explanation about these approaches can be found i series models, and and forecasts based on Further leadingformulation economicwith indicators. The Further explanation about these approaches can found(1997), in and Terasv (1993), and Swanson andbenumber White respectively. ut these approaches can transformation. beeconometric found in Granger and Terasvirta structural models generally rely on estimation of a large of Granger formation. Further explanation about these approaches can be found inWhite Granger andrespectively. Terasvirta can be found in Granger and Terasvirta (1993), and Swanson and (1997), simultaneous known as a system large numbers of factors being (1993),equations and Swanson and White (1997), with respectively. espectively. identified on the basis of economic theory. This approach is found to have poor 3), and Swanson and White (1997), respectively. out-of-sample performance. The small linear time-series models however are found to be more precise. Early versions of such models are known as vector auto-regressions (VARs) proposed by Sims (1980), as follows:

Early versions of such models are known as vector auto-regressions (VARs) propose

Early versions of small such models areSims known as vector (VARs) proposed by Early(1980), versions ofauto-regressions such models areto known as vector au s are known as vector auto-regressions (VARs) proposed by out-of-sample performance. The linear time-series models however are found be precise. rsions of such models are known as vector auto-regressions (VARs) proposed bymore Sims (198 as follows:

ws:

as such follows: asauto-regressions follows: Early versions of models are known as vector by Sims (1980), Alternative Approach to Determination of Malaysian Economic Behaviour: 77-97 (VARs) proposed 81 𝑌𝑌𝑡𝑡 = as𝜇𝜇follows: 𝑡𝑡 + 𝐴𝐴(𝐿𝐿)𝑌𝑌𝑡𝑡−1 + 𝜖𝜖𝑡𝑡 𝑛𝑛𝑛𝑛××11vector time series, × is a serially serially uncorrelated represents vector time series, 𝜖𝜖𝜖𝜖𝑡𝑡 𝑡𝑡isis 1 11isvector a aserially represents vector time series, isa aa𝑛𝑛 𝑛𝑛 is where𝑌𝑌 𝑛𝑛×× timeuncorrelated series, 𝜖𝜖𝑡𝑡 is dis a 𝑛𝑛 ctor time series, 𝜖𝜖𝑡𝑡 is a where𝑌𝑌 𝑛𝑛 where ×where𝑌𝑌 1 is𝑡𝑡 Yarepresents serially uncorrelated disturbance term, 𝑡𝑡 represents t𝑡𝑡 uncorrelated disturbance term, A(L) is a p-th order lag polynomial matrix, and

𝐴𝐴(𝐿𝐿) a1p-th order lagpolynomial polynomial denotes apolynomial 𝑛𝑛a × of deterministi 𝐴𝐴(𝐿𝐿) is1aand lag matrix, 𝜇𝜇𝑡𝑡 den nomial matrix, and 𝜇𝜇𝑡𝑡 𝑡𝑡represents denotes of deterministic denotes 𝑛𝑛 1×vector 1 vector ofand determi 𝐴𝐴(𝐿𝐿) order lag matrix, and 𝑡𝑡order where𝑌𝑌 vector time series, 𝜖𝜖𝑡𝑡 terms. isterms. amatrix, 𝑛𝑛Using ×Using isp-th a𝜇𝜇𝜇𝜇 serially uncorrelated disturbance term, denotes aais𝑛𝑛ais×p-th vector of deterministic VAR the parameters 𝑡𝑡method, represents 𝑛𝑛 × 1 vector time series, 𝜖𝜖𝑡𝑡 is a 𝑛𝑛 × 1 is a serially uncorrelated disturbance ter can be estimated efficiently subject to presence of no parameter restrictions. It is

VAR method, theparameters parameters can be efficiently subject presence ofe VAR method, parameters can to beterms. estimated ers can be 𝐴𝐴(𝐿𝐿) estimated efficiently subject to presence ofseries parameter isimportant a p-th order lag polynomial matrix, of and 𝜇𝜇no a 𝑛𝑛 × 1the vector of deterministic Using VAR can bebeestimated estimated subject to presence 𝑡𝑡 denotes the choice tomethod, note that the the selection to included inefficiently Yt and a p-th order lagofpolynomial matrix, and 𝜇𝜇𝑡𝑡 denotes a 𝑛𝑛 ×the 1 vector of deterministic lag order p play a very crucial role to satisfy preliminary conditions to terms. Usi restrictions. It is important to the notechoice that the of series totobe included 𝑌𝑌𝑡𝑡 and the restrictions. Itsubject is important that theinselection note that theVAR selection of series to beItincluded in be 𝑌𝑌𝑡𝑡 and of selection lag method, the forecasting. parameters can estimated efficiently presence of restrictions. is important tothe noteunivariate that the selection oftheto series tonote be in 𝑌𝑌 ano appropriate Likewise methods, choice of included lag no parameter 𝑡𝑡 ethod, thethe parameters canpachieved betoaappropriate estimated efficiently presence of no parame orders can be using information criteria. There hastocrucial been arole wide order play very crucial role to satisfy the preliminary conditions appropriate forecas order p subject play a very totoarray satisfy the prelimina le to satisfy preliminary conditions forecasting. Likewise restrictions. It is important to note that the selection of series to be included in 𝑌𝑌𝑡𝑡 and the choice of lag order p play a veryincrucial roleeconomic to satisfyand the finance preliminary conditions to appropriate for of applications of VARs empirical literature. Examples by Bates and Granger (1969), Granger and Newbold (2014), Stock the univariate methods, choice ofThere lag orders can achieved using information crite the univariate methods, the of lagchoice orders can hoiceItofislag orders canstudies achieved using information criteria. has ns. important tobe note that the selection of series toconditions be included in 𝑌𝑌choice the of l order pare play a very crucial role to satisfythe the preliminary tobeappropriate forecasting. Likewise 𝑡𝑡 and the univariate methods, the choice of lag orders can and be achieved using information and Watson (2004), Clark and McCracken (2005), and Smith Wallis (2009).

been athere wide array ofofapplications ofbeen VARs in empirical economic and finance literature. aachieved wide array of applications of VARsThere in empiric tions of VARs empirical economic and finance literature. Examples are Nevertheless, still exist some limitations the VAR approach pertaining theinunivariate methods, the choice lag orders can bein using information criteria. has lay a very crucial role theresponses preliminary conditions to appropriate forecasting. Likew beentoa satisfy wide array of applications of from VARs empirical economic and finance litera to distortions in impulse resulting theineffects of omitted variables studies by errors, Bates Granger (1969), Granger and Newbold StockGranger and Watson (20 studies by and Granger (1969), and are New (1969), Granger Newbold (2014), Stock and and Watson (2004), Clark andBates beenand aand wide array of applications VARs in empirical economic finance literature. Examples measurement orofmisspecifications being embedded in (2014), the residuals, byofBates Granger Grangerusing and Newbold Stock Watson ariate methods,leading the studies choice lag and orders be achieved information h to misinterpretation of can the (1969), results. Furthermore, the VAR (2014), modelscriteria. are andThere andSmith and Newbold Wallis Nevertheless, there exist someand lim McCracken (2005), andSmith and still Wallis (2009). Nev h and Wallisstudies (2009).byNevertheless, there(2005), still exist some limitations in(2009). the BatesMcCracken andthat Granger and (2014), Stock and Watson Clark a-theoretic, is they(1969), are notGranger based on any economic theory, given there is(2004), no

McCracken (2005), andSmith andeconomic Wallis (2009). Nevertheless, there still exist somea restrictions onofapproach any of thepertaining parameters under estimation. wide array McCracken of applications VARs in empirical and finance literature. Examples VAR to distortions in impulse responses resulting from the VAR approach pertaining tosome distortions in impulse o distortions in impulse responses resulting from the(2009). effects of omitted (2005), andSmith and Wallis Nevertheless, there still exist limitations in effec the

A very important issue related to the poor performance of VAR and VAR approach pertaining toanchored distortions into impulse responses resulting the other stated equivalent models isresiduals, in formal misspecifications variables and or misspecifications being embedded inoffrom the residua variables and measurement errors, or misspecificatio errors, or and misspecifications being embedded in theNewbold leading VAR approach pertaining to measurement distortions in errors, impulse responses resulting from the such effects omitted y Bates Granger (1969), Granger and (2014), Stock and Watson (2004), Clark a as spurious relationship between variables, resulting from the so-called nonvariables anda-theoretic, measurement ornotmore, misspecifications being embedded inthe the misinterpretation the results. Further theAccordingly, VAR areissue a-theoretic, that is VAR they misinterpretation of models the results. Further s. Further more, thestationary VAR models are is errors, they are based processes, as inof Clements and Hendry (1998). of more, variables and measurement errors, orthat misspecifications being embedded in the the residuals, leading tore ken (2005), andSmith and Wallis (2009). Nevertheless, there still exist some limitations in t variable stationarity and cointegration analysis has been extensively investigated onofany economic theory, given there is no restrictions on any of the parameters under estim misinterpretation of the results. Further more, the VAR models are a-theoretic, that is on any economic theory, given there is no restrictions o n there is no misinterpretation restrictions on any of the parameters under estimation. the results. Further more, the VAR models are a-theoretic, that is they since early 1980s by many scholars including Granger (1981), Granger and are not based proach pertaining distortions impulse responses resulting from the(1988), effects of omitt Weissto(1983), Grangerin(1986), Engle and Granger (1987), Johansen on any economic given there ison noany restrictions on any of the parameters under on any economic given theory, is no restrictions of the parameters under estimation. Johansen, theory, Douglas, andthere Nonaka (1985), Banerjee, Dolado, and Mestre (1998), and Harris, McInish, Shoesmith, and Wood (1995). Given theinnon-stationary and measurement the residuals, leading 7 errors, or misspecifications being embedded property of most economic time series as can be verified from their stochastic behaviour, investigation of a genuine long-run relationship in their trended pretation of the trend results. Further more, the VAR models are a-theoretic, that is they are not bas behaviour plays a significant role. This approach calls for 7determination of the validity of the cointegrating series by investigating the order of integration of conomic theory,variables, given there restrictions any of In the parameters estimation. whichisbyno definition shouldon be similar. other words, oneunder may assume an equilibrium long-run relationship to exist between variables if the variables are integrated of the same order. There are numbers of possible alternatives on cointegration testing. Broadly speaking, Engle and Granger (1987) developed an OLS framework to estimate the static version of cointegration model (SOLS). Another approach was proposed 7by Johansen (1988) in which the necessary information on the cointegrating property of variables will be provided while estimating the long-run relationship. As a pre-condition, all variables must be integrated of order 1 (i.e., I(1)) to be estimated in the model.Alternatively, the so-called Autoregressive distributed lag (ARDL) bound testing by Pesaran, Shin, and Smith (1999), and Pesaran et al. (2001) employs a single equation setup wherein a combination of I(0) and I(1) series can be taken into consideration.

al. (2001) employs a single equation setup wherein a combination of I(0) and I(1) se p wherein a combination of I(0) and I(1) series can be taken into consideration. Further more, different variables can be assigned different lag-leng 82 be assigned different variables can lag-lengths as they enter into the model. An ARDL (p, q) regression model takes the following form: different model takesFurthermore, the following form:variables can be assigned different lag-lengths as they enter into the model. An ARDL (p, q) regression model takes the following form:

where et is a random “disturbance” term. The model follows an auto regressive representation where yt is explained by lagged values of itself and employs distributed successive lags of the xt explanatory variable. Using this approach, the time series properties concerning serial correlation of residuals and dynamic stability of the model will be controlled. The presence of significant long-run relationship between variables can be verified using the so-called bound testing approach. Hence the model is more appropriate than conventional cointegration techniques. The ARDL has been extensively used in empirical economics and finance literature by many scholars, reasons being more applicability and flexibility of test modelling for economic time-series forecasting. Examples are studies by Narayan (2005), Duasa (2007), Rapach and Strauss (2010), and Barhoumi, Darné, Ferrara, and Pluyaud (2012). In the next section, the data and methodology applied in this study are discussed.

This research was designed to investigate whether the Malaysian economic behaviour is dependent on that of its major trade partners. The data series on variables (Gross Domestic Product and Industrial Production) were from Malaysia, China, and the US, where a long period quarterly data series over 19912014 were employed. The major sources of data were from: The International Financial Statistics (IFS) CD-ROM, and Thomson Reuters DataStream. “What determines the Malaysian time-series economic behaviour” was the research question. Two test models were developed specifying Malaysian GDP and IPI as dependent variables, with distributed lag structure of GDP and IPI from China and the US being the independent variables within bivariate and multivariate regression equations, respectively. It is believed that this approach has yielded new insights on how (i) the Malaysian economic behaviour can be determined concerning its (ii) interdependence on major trading partners’ past economic behaviour in an attempt to (iii) introduce an alternative and reliable approach to Malaysian economic forecasting. 3.1

The central research question was: What determines the behaviour of Malaysian economic time-series? To examine the normality assumption of data, a summary of descriptive statistics is provided in Table 1. The data used for this analysis were over the whole sample period. The variables were transformed into natural logarithmic form.

The measure of dispersion, standard deviation in GDP for China is larger than other relevant metrics in view of high economic fluctuations in China, whereas the Chinese industrial production is the least fluctuating series. The statistics on skewedness and kurtosis fulfil the assumption of normal distribution of data. In order to confirm the order of integration of the time series, this study conducted two unit root tests using the augmented Dickey-Fuller (ADF) (Dickey & Fuller, 1979, 1981), and the Phillips and Perron (1988) (PP) tests. The ADF model can be very useful in identifying higher order serial correlations in conjunction with higher order lags. The PP test allows for relatively weak assumptions regarding the distribution of residuals in the equation. The results reported in Table 2 suggest that most of the series are integrated of order one and the degree of integration of all of the series are not identical. Examining the results in Panel B, it was observed that all tests showed stationarity of the series. The levels data were not stationary or at best, this findings were mixed. Hence these series, to be used for ARDL, satisfies the necessary condition for reliable test results, given that all series are integrated at order zero or one. Table 2 reports the statistics on stationarity of data series. The statistics suggested that most of the data were stationary at first difference, which were judged by the respective ADF and PP tests of unit root. 4.1

Constant With Trend

Constant Without Trend

Constant With Trend -3.36 [10]

Panel B

First Difference

USGDP -3.88*** (1) -7.39*** (0) dings 4.1 from m Findings Structural Findings Structural from from Structural from Breakpoint Structural Breakpoint Structural Breakpoint Test Breakpoint Test Breakpoint Test TestTest ndings gs Structural from 4.1 from Findings Structural Breakpoint Structural from Breakpoint Structural Test Breakpoint Test Breakpoint TestTest

MYGDP 24.38 24.88 25.25 25.65 25.95 Q2 Q2Q DP 4.38 24.38 MYGDP 24.38 24.88 24.88 24.38 24.38 24.88 25.25 25.25 24.88 24.88 25.25 25.65 25.25 25.65 25.25 25.65 25.95 25.95 25.65 25.65 25.95 Q2 Q2 25.95 25.95 Q2 Q2 Q2 Q2 Q1 Q2 Q1 Q2 Q2 Q2 Q1 Q1 Q3 Q2Q3 Q2 Q2 Q3 8YGDP DP 24.38 MYGDP 24.88 24.38 24.88 25.25 24.88 24.38 25.25 25.65 25.25 24.88 25.95 25.25 25.65 25.95 Q2 25.65 25.95 Q2 Q2 25.95 Q1 Q2Q2 Q2 Q2 Q1 Q2Q1 Q1Q2 Q2 Q3 Q1Q1 Q3 Q3 Q3Q2 (0.12) (0.08) 25.65 (0.09) (0.09) (0.04) Q2 1993 1998 2003 2006 2009 Q2

MYIPI 3.66 4.35 Q2Q2Q2Q2 Q2Q3 Q3Q3 MYIPI 3.66 MYIPI 3.66 4.06 4.06 3.66 3.66 4.06 4.35 4.06 4.06 4.35 4.35 4.35break-selection Q2 Q2 Q2 Q2 Q3 Q3 Q3 Q3 6IPI3.66 3.66 MYIPI 4.06 3.66 4.06 4.35 4.06 3.66 4.35 4.35 4.064.06 4.35 Q2Q2 Q2 Q2 Q3 Q2 Q2Q2 Q3Q2 Q3 Q2 The4.35 test applied three criteria toQ2 identify optimum number -- - - -- - - - -- -- - - of breaks.(0.08) The final (0.06) choice was(0.04) made based on the LWZ criteria by Liu,1998 Wu, 1993 0.08) (0.08) (0.06) (0.06) (0.08) (0.08) (0.06) (0.04) (0.04) (0.06) (0.06) (0.04) (0.04) (0.04) 1993 1993 1993 1998 1998 1993 1993 1998 2002 2002 1998 1998 2002 2002 2002 2002 8)(0.08) (0.08) (0.06) (0.08) (0.06) (0.04) (0.06) (0.08) (0.04) (0.04) (0.06) (0.04) 1993 1993 1998 1993 1998 2002 1998 1993 2002 2002 1998 and Zidek (1997), which is robust to serial correlation problems, and the test 2002 performed relatively well. The statistics revealed that there is evidence of three structural breaks in the Malaysian economic series which are in line with several up-turns ever since 1990, given that the coefficients reported are all positive (the The test applied three criteria to identify optimum number ofbreaks. breaks. The fina eThe ed three applied test three test applied break-selection applied three break-selection three break-selection three break-selection criteria break-selection criteria to criteria to identify criteria identify criteria to optimum tooptimum identify to identify optimum number optimum optimum number of of breaks. number breaks. number of breaks. of The final breaks. final breaks. The choice choice The final The final choice final choice cho st ree plied applied The break-selection three test three break-selection applied break-selection criteria three break-selection criteria tobreak-selection identify criteria to identify optimum toidentify criteria identify optimum number tooptimum identify number of breaks. number optimum of breaks. The ofThe number breaks. final The choice of The final final choice The choice final up-turn behaviour of GDP and IPI can also benumber verified from Figures 1of and 2). Although the Malaysian economy witnessed a negative average nominal GDP was made based on the LWZ criteria by Liu, and Zidek (1997) ,serial which is robust toserial serial co ,Wu, which ,and which is,the which iswhich robust ,is which ,isto which to serial is robust isserial to correlation robust serial correlation to serial tocorrelation serial correlat sde was ed on made based on the made the based LWZ on LWZ based the criteria on LWZ criteria on the the LWZ by criteria LWZ by Liu, criteria Liu, Wu, by criteria Wu, Liu, and by and by Liu, Wu, Zidek Liu, Zidek Wu, and (1997) Wu, Zidek and (1997) and Zidek (1997) Zidek (1997) (1997) ,Wu, which ,isto which robust ,robust torobust serial isrobust ,robust which to correlation tois serial robust correlation correlation tocorrelation corr nbased ade the was based LWZ on made the on criteria LWZ the based LWZ by criteria on Liu, criteria the by Wu, LWZ Liu, by and criteria Liu, Wu, Zidek Wu, and by (1997) Zidek and Liu, Zidek (1997) (1997) Zidek (1997) growth rate of minus 11% as of 1985-1986 due collapse of commodity prices leading to the country’s recession since its independence, the country has problems, and the test performed relatively well. The statistics revealed that there evidence dhe problems, blems, s,the and test test performed the and performed and test the the performed test relatively test performed relatively performed relatively well. well. relatively The relatively well. The statistics well. statistics The well. The statistics revealed revealed statistics statistics revealed that that revealed there revealed there is there evidence is that evidence there isline evidence of isthere of three evidence isof three evidence ofis three of three of tho and ms, test problems, the and performed test the performed test and relatively performed the test relatively well. performed relatively The well. statistics relatively well. The The statistics revealed well. statistics revealed The that revealed statistics there that isthat that there evidence revealed there isthat evidence that of isthere evidence three is three of evidence three been growing steadily and substantially atThe the rate of 8% on average, in with structural breaks in the Malaysian economic series which are in line with several up-turns ev uctural snal aks structural in breaks in the the breaks Malaysian in breaks Malaysian in Malaysian the in economic the Malaysian Malaysian economic series economic series economic which series which series are which are in which in line are which line with in are with line are several in several line with inwith line with up-turns several up-turns with several several up-turns ever ever up-turns since up-turns since ever ever since ever since si breaks ral the structural breaks Malaysian in the inthe Malaysian breaks the economic Malaysian ineconomic the economic series Malaysian economic which series economic series are which inseries which line are series with in are line which in several line with are several up-turns in several line up-turns with ever up-turns since several ever ever since up-turns since eve

1990, given that the coefficients reported are allbehaviour positive (the up-turn behaviour ofalso GDP andcan IPIa 90, hat ven the given the coefficients that given coefficients the that coefficients the reported the coefficients reported coefficients reported are are all reported all positive reported are positive all are positive (the are all (the positive all up-turn up-turn positive (the up-turn (the behaviour (the up-turn up-turn behaviour of behaviour of GDP GDP behaviour of and and GDP IPI of IPI GDP of can and can GDP also IPI and and can IPI also IPI can also ngiven e1990, that coefficients 1990, the that coefficients given thethat reported coefficients that the reported are coefficients reported all positive are all are reported positive (the all positive up-turn are (the all up-turn (the behaviour positive up-turn behaviour (the of behaviour GDP up-turn of and GDP behaviour of IPI GDP and can also IPI and of can IPI GDP also canand also IPI c its extensive foreign and domestic investment and manufactured goods export activities. The first break date (1993Q2) reported in Table 3 was associated with the introduction of a number of significant tariff and fiscal policy changes that took effect immediately leading to the growth of the economy. In mid-1997 the Asian Financial crisis led to severe deteriorations in the performance of rate of minus 9% until the second quarter of 1998, mainly due to substantial credit squeeze and Malaysian financial markets, leading to a decline in GDP growth rate of minus mounting the corporate sector. after the due secondtoquarter of 1998, credit the GDPsqueeze and IPI 9% untildebts the insecond quarter ofHowever, 1998, mainly substantial and debts corporate sector.control However, after the second1998. quarter of beganmounting growing in line within thethe introduction of exchange measures, as of September 1998, the GDP and IPI began growing in line with the introduction of exchange control measures, as of September 1998.

and improved export performance. Furthermore, the low interest rate policy stimulated higher access to

In 2002, the Malaysian economy strengthened its GDP growth by 4.2%, financing and significant improvement in commodity prices leading to a considerable growth in the due to strong domestic demand and improved export performance. Furthermore, private The Malaysian economic performance alsoaccess grew into2006 with its and nominal domestic the lowsector. interest rate policy stimulated higher financing significant improvement in commodity prices leading to a considerable growth in the private product expanding by 6% in line with the positive global economic environment that led to a strong sector. The Malaysian economic performance also grew in 2006 with its nominal demand for electronics and primary commodities, triggering private consumption growth and private domestic product expanding by 6% in line with the positive global economic environment that led to athese strong demand for electronics primary commodities, investment expansion to meet demands. Following the advent ofand global financial crisis in 2007triggering private consumption growth and private investment expansion to 08, in the first half of 2009, the global economy experienced a sharp contraction leading to the credit meet these demands. Following the advent of global financial crisis in 2007-08, crunch the private low household consumption, declining employment production cut by in theinfirst half sectors, of 2009, the global economy experienced a rate, sharp contraction leading to the credit crunch in the private sectors, low household consumption, mainstream businesses, and the associated collapse of world trade. Accordingly, the Malaysian economy declining employment rate, production cut by mainstream businesses, and the faced an economic downturn with its GDP declining by 6.2% in the first half of the year, while associated collapse of world trade. Accordingly, the Malaysian economy faced experiencing an accelerated in the second half due toby the6.2% implementation of fiscal an economic downturnrecovery with its GDP declining in the first halfstimulus of the year, while experiencing an accelerated recovery in the second half due to tothe measures, aggressive easing of monetary policy, and continued access to financing, therebyleading a implementation of fiscal stimulus measures, aggressive easing of monetary revival of the private sector sentiment and improvement of labour market conditions and the resulting policy, and continued access to financing, there by leading to a revival of the private sentiment and of labour market economy conditions the expanded sector private consumption.In viewimprovement of such fluctuations in the Malaysian (as isand evident resulting expanded private consumption. In view of such fluctuations in the Malaysian economy (as is evident from Figures 1 and 2), the question of its economic interdependency on its major trading partners needs some detailed investigation and testing to verify an alternative approach for the determination of Malaysian economy.

some detailed investigation and testing to verify an alternative approach for the determination of Malaysian economy.

Hence, there is a need to develop a model that could establish linkages between the Malaysian economy

Hence, there is a need to develop a model that could establish linkages between the Malaysian economy and those of its trading partners. In the next section, this paper reports and discusses the from findings associated the associated with the determination of Malaysian economy arising its relationship with with its trading determination of Malaysian economy arising from its relationship with its partners, particularly USA and China. trading partners, particularly USA and China.

and those of its trading partners. In the next section, this paper reports and discusses the findings

4.2Findings Findings ARDL 4.2 from thefrom ARDL the test results test results

This section presents and discusses the results associated with the ARDL test. Table 4 reports statistics

This section presents and discusses the results associated with the ARDL test.

on the ARDL test from the regression Equation test 7 onfrom pair-sampled-country equations;having Table 4 reports statistics on the ARDL the regression Equation 7 the on pair-sampled-country having theseparately, Malaysian on the Malaysian GDP regressed onequations; the US and Chinese GDP, as in GDP the firstregressed and second columns

US and Chinese GDP, separately, as in the first and second columns and on both countries’ GDP series, as in the third column. The empirical results were reparameterisation of different estimated ARDL models reported in the upper panelmodels of the table. The test based on the reparameterisation of different estimated ARDL reported in the upper panel of the table. The test procedure applied a large number of procedure applied a large number of model iterations (72 and 648) to produce fully specific models with model iterations (72 and 648) to produce fully specific models with lagged lagged parameters being identified appropriately in presence of no serial correlation problem. The results parameters being identified appropriately in presence of no serial correlation problem. The results affirmed evidence of a significant contemporaneous and lagged impact of the US GDP on Malaysian GDP at four quarters or a year ahead from the actual Malaysian GDP growth. Similarly, the Chinese GDP determined the behaviour of Malaysian GDP contemporaneously and a year ahead, when taken into consideration jointly with the US GDP. However, the information on the Chinese GDP in isolation can only have a contemporaneous impact on the Malaysian GDP, with coefficient size being very negligible (0.045) and significant at 10%. Results from Table 5 is also evident of the fact that Malaysian IPI was determined mainly by the US contemporaneous and one-quarter lagged IPI, when taken in isolation and jointly with the Chinese IPI. The Chinese IPI however was found to have no impact on the Malaysian IPI, even when estimated jointly with the US IPI. Hence, the role of China in determination of the Malaysian IPI as a proxy for earnings and income was not significant. The contemporaneous impact of the US IPI on that of Malaysia was positive while the one-quarter lagged impact was negative with relatively lower magnitude.

and on both countries’ GDP series, as in the third column.The empirical results were based on the

MYGDP (-2)

Intercept

MYGDP (-1)

MYGDP (-2)

UDGDP (-1)

UDGDP (-2)

UDGDP (-3)

UDGDP (-4)

CHGDP (-1)

CHGDP (-2)

CHGDP (-3)

CHGDP (-4)

Obs.

# of Models Evaluated

Obs. # of Models Evaluated Adjusted R-Squared

Obs.

# of Models Evaluated

Adjusted R-Squared

Adjusted R-Squared

Intercept

Obs.

Obs.

Obs.

# of Models Evaluated

# of Models Evaluated

# of Models Evaluated

Adjusted R-Squared

Adjusted R-Squared

Adjusted R-Squared

Tables 6.1 and 6.2 report statistics on bound tests of Pesaran et al. (2001) to verify presence of long-run relationship between Malaysian GDP and IPI, and those of its trading partners. The results were indicative of presence of significant long-run (cointegrating) relationship for all six model types discussed in previous sections. The computed F-value, the likelihood ratio, and Lagrange multiplier were used for testing the long-run relationship. Pesaran et al. (2001) provided critical values for the bound tests.

In Table 6.1, k was the number of variables; the maximum lag identified was 2; the tests is identified the upper and lower bounds at three levels of is significance. The calculated F-statistics (e.g., F (1, 89) = 9.465 for the first equation), was greater than upper bound at 1 percent degree of significance. Results reported in Table 6.2 also verify presence of significant long-run relationship between Malaysian IPI and that of US and China, given that the associated F-statistics was bigger than the upper bound of all three relevant models.

In Table 6.2, k was the number of variables; the maximum lag identified was 2; the tests is identified the upper and lower bounds at three levels of is significance. Table 7 and 8 summarise the results from the ARDL long-run relationship tests for GDP and IPI, respectively. Results from Equation 1 revealed that 1 unit increase in the US GDP will lead to an increase of 1.94 units in the Malaysian GDP over the long-run. Similarly, a 1 unit increase in Chinese GDP will lead to an increase of 0.59 unit in the Chinese GDP when considering a long-run relationship. Results from the joint model (i.e., Equation 3 in Table 7) also showed that there is a relatively larger impact of the US GDP than that of China on the Malaysian GDP over the long-run. On a similar basis, the results reported in Table 8 showed that US IPI played a more contributing role in determination of

Malaysian IPI over the long-run compared to the IPI of China, which was very negligible when estimated jointly with the US IPI.

Cointegration Equation 1 = MYGDP - (1.9432*USGDP) - 33.4908) [13.58***]

Cointegration Equation 2 = MYGDP (0.5908*CHGDP) + 18.8947) Cointegration Equation 3 = MYGDP (1.1913*USGDP + 0.2746*CHGDP) - 13.5037) [-2.35**]

Cointegration Equation 1 = MYIPI - (1.6459*USIPI) - 3.5895) [3.89***]

Cointegration Equation 2 = MYIPI - (0.5908*CHIPI) + 18.8947) [12.32***]

Cointegration Equation 3 = MYIPI - (1.6433*USIPI – 0.0257*CHIPI) - 3.4572) [3.60***]

Finally, the results from the Error Correction model are reported, to identify the speed of adjustment or convergence of the variables to the longrun (cointegrating) equilibrium. A necessary condition for the verification of the presence of long-run equation is to have an error correction term (ECT) to be negative and significant at 5% or lower degree of significance. In other words, this test will identify the time to revert to the equilibrium, provided the coefficient is negative and significant. Tables 9 and 10 report results on the six error correction models in line with the results obtained from previous sections. It was found that the condition was met for all six equations. The US and Chinese GDPs were found to have an almost identical speed of adjustment (0.071, 0.076), implying that the disequilibrium between Malaysian GDP and the US GDP, as well as Malaysian GDP, and Chinese GDP can be corrected by 7% within each period (i.e. per quarter). In other words, it takes approximately 3.5 years (14 quarters) for the total equilibrium to be adjusted; while considering both countries GDP jointly, the speed of adjustment will tend to be 1 year (4 quarters = 100 % / 24.2 %). The statistics in Table 10 also revealed evidence of significant error correction terms

for both countries separately and when considered jointly. The speed of adjustment was found to be slower compared to the results obtained in Table 9. For example, the disequilibrium between the Malaysian IPI and that of the US would be corrected over almost 4.5 years. Likewise, the disequilibrium between Malaysian IPI and Chinese IPI would be corrected over a longer period of approximately 10 years. Hence, although the results showed presence of significant long-run relationship between the sampled countries’ IPIs, they tended to be cointegrated on a relatively slower pace than GDP. A reason can be due to the fact that countries have certain comparative advantages over each other, being relatively distinct in different sectors. For example, China has higher degrees of comparative advantage in agricultural products than Malaysia, while Malaysia’s economic activities in industrial sector are higher. China’s manufacturing growth rate is ranked 7th while that of Malaysia is 87th with relatively divergent trend behaviour. Likewise, the rate of growth in industrial value added products and services such as manufacturing, construction, electricity, water, and gas was almost two times higher than that of the US. Furthermore, US and Malaysia have a record of dissimilar manufacturing growth rates, and production of car, machinery and transport equipment; with Malaysian growth rate being higher.

D(MYIPI(-1)) 0.170 (1.69)*

ECT(-1) -0.057 (-4.09)**

ECM 2: MYIPI = f (CHIPI) D(CHIPI) -0.139 (-1.10)

ECT(-1) -0.023 (-4.55)***

ECM 2: MYIPI = f (USIPI, CHIPI) MYIPI

D(USIPI) 1.195 (3.69)***

D(USIPI(-1)) -0.498 (-1.51)

D(CHIPI) -0.018 (-0.15)

ECT(-1) -0.057 (-4.06)***

This paper conceived an alternative approach with novelty in application compared to previous studies to examine the interrelationship between Malaysian economic time series behaviour and those of its trading partners (USA and China). This study applied new and more appropriate econometric approaches; namely (i) the structural break test of Bai and Perron (2003) to identify the structural breakpoints in the behaviour of Malaysian GDP and IPI, and (ii) the ARDL cointegration approach of Pesaran et al. (2001) to estimate the interdependency of the Malaysian economy on its trading partners. The Malaysian, the US, and Chinese data were used since long-length data are available for these economies readily. The ARDL bound testing was used, which satisfies the long-length equilibrium on paired-country as well as multi-country testswherein Malaysian GDP and IPI is regressed on the US and Chinese GDP and IPI. The results revealed that both the GDP and IPIof Malaysia are strongly cointegrated with those of the US and China. The US GDP and IPI as proxies for economic growth and earnings were found to have contemporaneous and 1-year lagged impact on those of Malaysia, while the Chinese GDP has a lagged impact only when estimated jointly with the US. Although there was evidence of longrun relationship between Malaysian IPI and the US and Chinese IPIs, the speed of adjustment was found to range between 4.5 to 10 years, which, in our view, is due to relatively distinct behaviour of countries in their industrial production activities concerning the manufacturing, construction, electricity, and water and gas sectors, as well as production of cars and other machineries.The econometric tests conducted in this study, in our view, made these results reliable and robust compared to earlier studies, given the controls for the number of lag parameters are provided in presence of no serial correlation leading to high model fitness and model specification. Hence, this study has certain implications on how the Malaysian economy is likely to behave based on past information obtained from the US and Chinese economic time series. Perhaps the research process followed in this study can provide a new approach, which may be useful to study other economies to reveal if the same relationship between key fundamental economic time-series hold in more economies than in Malaysia. Acknowledgment The authors are grateful to Sunway University for permission to access the data bases in the University. A number of people have contributed ideas that helped in this study. Among them, we would like to thank Mohamed Ariff and Law Siong Hook, who had collaborated at different times on this research topic with the first author. The authors are solely responsible for any errors in the paper.