Predictability and Efficiency of the Philippine Stock Exchange Index

Transcription

1 Journal of Business and Economics, ISSN , USA April 2014, Volume 5, No. 4, pp DOI: /jbe( )/ /009 Academic Sar Publishing Company, 2014 hp:// Predicabiliy and Efficiency of he Philippine Sock Exchange Index Jo-hui Chen, John Francis T. Diaz (Chung Yuan Chrisian Universiy, Chung-li, Taiwan) Absrac: This sudy applied he auoregressive fracionally inegraed moving average (ARFIMA) and he ARFIMA-fracional inegraed general auoregressive condiional heeroskedasiciy (ARFIMA-FIGARCH) models on he daily sock reurns of he Philippine Sock Exchange Index (PSEi) from January 4, 2000 o December 29, 2011 and found significan long-memory processes in boh reurn and volailiy of he PSEi. The Chow breakpoin es was used o idenify he srucural break during he subprime morgage crisis. Posiive dependence in boh reurn and volailiy was eviden before he srucural break, bu predicabiliy was only observed in volailiy afer he breakpoin. The log-likelihood value consisenly showed ha he combined ARFIMA-FIGARCH model beer characerized he PSEi. Key words: long-memory; ARFIMA-FIGARCH; Philippine Sock Exchange Index JEL codes: G11, G14 1. Inroducion The Philippine Sock Exchange (PSE) is he naional sock exchange of he Philippines and is one of he oldes in Souheas Asia, conducing coninuous operaion since is esablishmen on Augus 8, I was he produc of he merging of he Manila Sock Exchange and he Makai Sock Exchange. The PSE Composie Index (PSEi) is he main index of he bourse, which is composed of 30 lised companies and addiional six sub-secors aside from he PSEi, namely, financial, indusrial, holding firms, propery, services, and mining and oil indices. As of December 2010, he PSE had a oal marke capializaion of USD 202 billion, wih 253 companies and 184 rading paricipans lised. The PSEi is he mos moniored index in he Philippines and is also one of he mos observed economic indicaors. The srong poenial of he Philippine economy has been recognized wih is inclusion as one of he op 16 economies of he world by 2050, as prediced by he Hong Kong and Shanghai Banking Corporaion (HSBC). Such forecas increased he confidence on he economic posiion of he counry in he global arena. The Philippine economy s 27-noch improvemen 40 years from now, which ranked 41s in 2010, will be of grea ineres o he inernaional invesing communiy. The PSEi, as one of he economic indicaors of he counry, should gain renewed ineres from boh academicians and invesors. The possible predicion of he Philippine sock marke, paricularly PSEi reurns, is currenly a primary ineres because of is immense possibiliy o propel upwards. Jo-hui Chen, Ph.D. in Economics, Professor, Deparmen of Finance, Chung Yuan Chrisian Universiy; research areas: inernaional finance and regional economics. johui@cycu.edu.w. John Francis T. Diaz, Ph.D. in Business, Assisan Professor, Deparmen of Inernaional Business, College of Business, Chung Yuan Chrisian Universiy; research areas: inernaional finance and invesmens, and inernaional rade and economics. di.jiang@cycu.edu.w. 535

2 The efficien marke hypohesis (EMH) of Fama (1970) is a viable explanaion for he difficuly or even he impossibiliy of predicing sock reurns, especially hose of developed markes. The reason is ha price movemens occur accordingly in he marke. On he conrary, several sudies (Bekaer, 1995; Bekaer & Harvey, 1995; Wrigh, 1999) have cied ha emerging marke reurns, such as he Philippines, could possibly be more persisen han he more developed markes. This condiion may be a sign of marke inefficiency and a probable long-memory propery, and i may be exploied by invesors o earn excess reurns. According o Wrigh (1999), anoher reason why emerging markes are desirable for porfolio diversificaion is heir low correlaion wih developed markes and heir obvious poenial for profiabiliy. These conexs are eviden in he Philippines, as indicaed by HSBC s curren research findings. This sudy conribues o he lieraure on sock marke reurns by examining he dual long-memory properies of he Philippine sock marke, paricularly, he PSEi, in is condiional mean and variance. To invesigae he hypoheses, his sudy uilizes he auoregressive fracionally inegraed moving average (ARFIMA) process. The ARFIMA was firs menioned in Granger and Joyeux (1980) and Hosking (1981), in which he difference parameer is allowed o be a non-ineger. Similarly, he curren paper adops he fracional inegraed general auoregressive condiional heeroskedasiciy (FIGARCH) model o verify long memory in reurn and volailiy, as suggesed by Baillie e al. (1996). To he bes of our knowledge, no formal empirical lieraure has sudied he dual long-memory properies of he PSEi and is efficiency based on he aforemenioned models. However, Aquino (2006) concluded ha he Philippine sock marke suppors a weak-form efficien hypohesis using he AR(1) process, which is weak in erms of no considering higher AR lag operaors and lagged values of he error erm provided by he MA process. Lieraure on ARFIMA-FIGARCH models has been proved o capure he long-memory properies of ime series daa. The sudies of Floros e al. (2007), Kang and Yoon (2007), McMillan and Thupayagale (2008), Korkmaz e al. (2009), and Tan and Khan (2010) uilized he models o examine sock marke reurns. Beine e al. (2002) used i o illusrae exchange raes, and Choi and Hammoudeh (2009) adoped i for commodiies. This sudy has he following objecives. Firs, i examines he long-memory properies in he reurns and volailiies of he PSEi. Second, i idenifies he srucural break poin using he Chow (1960) es in he recen subprime morgage crisis and examines if changes exis in he long-memory propery of he exchange before and afer he srucural break. Third, i verifies if EMH exiss in he Philippine sock marke, as long memory in ime series implies ha lagged reurns can be uilized o predic fuure reurns. Lasly, i aims o deermine he exisence of a dual long memory in reurns and volailiies in PSEi reurns, similar o he findings of Kang and Yoon (2007) in he Korean sock reurns. This paper is organized as follows. Secion 2 explains he daa and he ARFIMA-FIGARCH models; Secion 3 discusses he empirical resuls; and Secion 4 presens he conclusion. 2. Daa and Mehodology Daily closing prices of he PSEi sock marke reurns during he las decade from January 4, 2000 o December 29, 2011 were obained from he Yahoo Finance Websie. This sudy divided he original ime series ino wo o deermine changes in he long-memory properies of he index before and afer he srucural break caused by he recen subprime morgage crisis. Period 1 covers January 4, 2000 o Ocober 24, 2008, and period 2 encompasses Ocober 30, 2008 o December 29, Daa modeled by boh ARFIMA and FIGARCH processes are explained below. 536

3 According o Granger and Joyeux (1980) and Hosking (1981), he ARFIMA model is a parameric approach used o examine long-memory properies of ime series. This model is disincive because i enables he difference parameer o be a non-ineger and considers he fracionally inegraed process I(d) in he condiional mean. The ARFIMA (p,d,q) model saisfies boh saionariy and invariabiliy condiions. I is expressed as d 2 ( L)(1 L) Y ( L), ~ (0, ) (1) where d is he fracional inegraion real number parameer, L is he lag operaor, and is he whie noise residual. The fracional differencing lag operaor (1-L) d is expressed in he following equaion: d( d 1) 2 d( d 1)( d 2) 3 (1 L) d 1 dl L L... (2) 2! 3! The process is saionary when he ARFIMA model is -0.5 < d < 0.5, in which he effec of shocks o decays a a gradual rae o zero. The process also has a shor memory if d = 0, in which he effec of shocks decays geomerically. A uni roo process is exhibied when d = 1. A long memory process or he so-called posiive dependence among remoe observaions exiss when 0 < d < 0.5. Inermediae memory or ani-persisence exiss when -0.5 < d < 0 (Baillie, 1996). The process is non-saionary if d 0.5 and is a saionary bu non-inverible process if d -0.5, hus making he ime series impossible o model by any auoregressive process. As proposed by Baillie e al. (1996), he FIGARCH model capures long memory in he reurn and volailiy of economeric ime series. The model provides furher flexibiliy in examining he condiional variance, encompassing he covariance saionary GARCH for =0 and he non-saionary IGARCH for =1. The FIGARCH (p,, q) model is wrien ashhf d 2 L )(1 L) [1 ( L)] ( v (3) where v is he innovaion for he condiional variance. The roo of (L) and [1-(L)] is condiioned o lie ouside he uni roo circle. 3. Empirical Resuls Table 1 shows ha he average PSEi reurns for he enire 11 years gains 1.2%, sub-period 1 has an average loss of -10.9%, and period 2 incurs average gains of 11.5%. Period 1 could have been affeced by significan losses from he subprime morgage crisis, whereas period 2 benefied from he general upward rend afer he crisis. The sample of he enire period posed he highes volailiy, wih a sandard deviaion of Beween sub-periods 1 and 2, he former experienced a slighly higher volailiy of compared wih of he laer. This finding may be evidence of he modern porfolio heory of Markowiz (1952), which indicaes ha he greaer is he dispersion of reurns, he higher he risk of an invesmen possibly aribued o he predominanly downward rends and heighened volailiy of sub-period 1 because of he crisis. All sample periods are negaively skewed and conain lepokuric disribuions. The Jarque-Bera saisics for residual normaliy shows ha he PSEi reurns are under a non-normal disribuion assumpion. Based on he resuls of he Lagrange muliplier (LM) es, samples have no serial correlaion. The minimum value of he Akaike informaion crierion is uilized o idenify he orders of he ARFIMA and ARFIMA-FIGARCH models. This paper used he ARCH-LM process o es he ARCH effec and o show ha we could apply GARCH models in he chosen sample because he null hypohesis was rejeced for PSEi reurns. Table 2 illusraes he resuls for boh ARFIMA and ARFIMA-FIGARCH models. The ARFIMA model shows ha in he pas 11 years, he PSEi exhibied posiive dependence. Therefore, he Philippine index has a long-memory process in is reurns, hus implying marke inefficiency and he possibiliy of invesors exploiing 537

4 such srucure o earn excess reurns is likely. The long-memory process in reurns is similar o hose in he sudies of Kang and Yoon (2007) and Tan and Khan (2010) ha probe ino he Souh Korean and Malaysian sock marke reurns, respecively. The long-memory process was also observed in sub-period 1 or before he subprime crisis srucural break. However, afer experiencing crisis, he PSEi showed ani-persisence in sub-period 2, in which an inermediae memory became eviden. This finding indicaes no only he PSEi s efficiency afer he subprime morgage crisis bu also is endency o rever as well. In erms of reurn and volailiy oucomes, he reurn and volailiy of he PSEi for he enire sample period exhibied long-memory processes. Such findings are similar o hose of Kang and Yoon (2007), Korkmaz e al. (2009), and Tan and Khan (2010), furher implying ha he PSEi reurn and volailiy have predicable srucures and are no a weak-form efficien marke. The period of he curren sudy may have produced conradicing resuls wih he earlier findings of Aquino (2006), who finds ha he Philippine sock marke is a weak-form efficien using he AR(1) process. Long-memory processes in boh reurn and volailiy are also eviden in sub-period 1, bu sub-period 2 is characerized only by posiive dependence in is volailiy. The log-likelihood value consisenly considers ARFIMA-FIGARCH o beer characerize he PSEi compared wih he ARFIMA model. Table 1 The Sample Size and Period of PSEi Reurns PSE Index Sar of Daa End of Dae Obs. Mean Sd. Dev. Skew. Kur. J-Bera LM es ARCH-LM Whole Period Jan. 4, 2000 Dec. 29, *** *** Period 1 Jan. 4, 2000 Oc. 24, *** *** Period 2 Oc. 30, 2008 Dec. 29, *** *** Source: Yahoo Finance January 2000 o December 2011; hp:// Table 2 Summary Saisics of ARFIMA and ARFIMA-FIGARCH Models ARFIMA Log- ARFIMA-FIGARCH Log- PSE Index model d-coeff. AIC likelihood d-coeff. model d-coeff. AIC likelihood Whole Period (2,1) (0.031)** (0.023)** (1,2) (0.000)*** Period 1 (2,1) (0.058)* (0.060)* (2,2) (0.045)** Period 2 (2,2) (0.020)** (0.333) (2,2) (0.000)*** Noe: *, ** and *** are significance a 10, 5 and 1% levels, respecively; p-values are in parenheses. 4. Conclusion The economic significance of predicing sock reurn and volailiy of he Philippine sock exchange should be of ineres o raders, especially as HSBC prediced he Philippines o be he 16h larges economy by 2050, ascending 27 noches from is 43rd posiion in This sudy addressed his concern by providing evidence on he sock reurn and volailiy predicabiliy of he PSEi. The ARFIMA model shows ha he Philippine index exhibis long-memory process in is reurns, which can be exploied by invesors o earn excess reurns. I is a clear sign of marke inefficiency. Posiive dependence was observed in sub-period 1. However, afer he srucural breakpoin of he crisis, he PSEi in sub-period 2 showed o have inermediae memory, indicaing improved efficiency. The ARFIMA-FIGARCH model demonsraed long-memory processes in boh reurn and volailiy hroughou he sample period. This predicable srucure of he PSEi using lagged index reurns conradics he iniial findings of Aquino s (2006) sudy, which characerizes i as a weak-form efficien. Posiive dependence in boh reurn and volailiy is eviden in sub-period 1; sub-period 2 is predicable only in is volailiy. Beween he 538

5 ARFIMA and he ARFIMA-FIGARCH models, he log-likelihood value consisenly indicaes ha he laer bes characerizes he PSEi. References: Aquino R. (2006). Efficiency of he Philippine sock marke, Applied Economic Leers, Vol. 13, No. 7, pp Baillie R. (1996). Long memory processes and fracional inegraion in economerics, Journal of Economerics, Vol. 73, No.1, pp Baillie R., Bollerslev T. and Mikkelsen H. (1996). Fracionally inegraed generalized auoregressive condiional heeroskedasiciy, Journal of Economerics, Vol. 74, No. 1, pp Bekaer G. (1995). Marke inegraion and invesmen barriers in emerging equiy markes, World Bank Economic Review, Vol. 9, No.1, pp Bekaer G. and Harvey C. (1995). Time-varying world marke inegraion, Journal of Finance, Vol. 50, pp Beine M., Lauren S. and Lecour C. (2002). Accouning for condiional lepokurosis and closing days effecs in FIGARCH models of daily exchange raes, Applied Financial Economics, Vol. 12, No. 8, pp Choi K. and Hammoudeh S. (2009). Long memory in oil and refined producs markes, The Energy Journal, Vol. 30, No. 2, pp Chow G. (1960). Tess of equaliy beween ses of coefficiens in wo linear regressions, Economerica, Vol. 28, No. 3, pp Fama E. (1970). Efficien capial markes: A review of heory and empirical work, Journal of Finance, Vol. 25, No. 2, pp Floros C., Jaffry S. and Lima G. (2007). Long memory in he Poruguese sock marke, Sudies in Economics and Finance, Vol. 24, No. 3, pp Granger C. and Joyeux R. (1980). An inroducion o long memory ime series models and fracional differencing, Journal of Time Series Analysis, Vol. 1, No. 1, pp Hosking J. (1981). Fracional differencing, Biomerika, Vol. 68, No. 1, pp Kang S. H. and Yoon S. M. (2007). Long memory properies in reurn and volailiy: Evidence from he Korean sock marke, Saisical Mechanics and Is Applicaions, Vol. 385, No. 2, pp Korkmaz T., Cevik E. and Ozaac N. (2009). Tesing for long memory in ISE using ARFIMA-FIGARCH model and srucural break es, Inernaional Research Journal of Finance and Economics, Vol. 26, pp McMillan D. and Thupayagale P. (2008). Efficiency of he Souh African equiy marke, Applied Financial Economics Leers, Vol. 4, No. 5, pp Markowiz H. (1952). Porfolio selecion, Journal of Finance, Vol. 7, No. 1, pp Nouira L., Ahamada I., Jouini J. and Nurbel A. (2004). Long-memory and shifs in he uncondiional variance in he exchange rae Euro/US Dollar reurns, Applied Economics Leers, Vol. 11, No. 9, pp Tan S. and Khan M. (2010). Long memory feaures in reurn and volailiy of he Malaysian sock marke, Economics Bullein, Vol. 30, No. 4, pp Wrigh J. (1999). Long memory in emerging marke sock reurns, Inernaional Finance Discussion Papers, number