Dissertation Paper. Testing the Informational Efficiency of the Romanian Capital Market

Transcription

1 Academy of Economic Sudies Buchares Docoral School of Finance and Banking Disseraion Paper Tesing he Informaional Efficiency of he Romanian Capial Marke Supervisor PhD. Professor Moisa Alar M. Sc. Suden Adrian Codirlaşu June 2000

2 2 Conens 1. INTRODUCTION 3 2. TESTS OF THE RANDOM WALK HYPOTHESIS Uni Roo Tess and Dependence Tess for he Romanian Capial Marke BET Index BETC Index RASDAQC Index Implicaions of Nonrandom Walks Volailiy Beas TESTS OF THE WEAK FORM OF MARKET EFFICIENCY Seasonaliy of Reurns Tess of he Trading Rules Tess of he Filer Rules Tess of Moving Average Rules TESTS OF THE SEMI-STRONG FORM OF MARKET EFFICIENCY The Response of Sock Prices o he Announcemen of a Sock Spli The Response of he Sock Prices afer a New Company is lised on a Sock Exchange TESTS OF THE STRONG FORM OF MARKET EFFICIENCY ALTERNATIVES TO THE EFFICIENT MARKET HYPOTHESIS 35 BIBLIOGRAPHY 36

3 3 1. Inroducion Several very imporan implicaions for securiy valuaion are derived from he heory of marke efficiency. The main premises of an efficein marke are: firs assumpion is ha he marke consiss of a large number of profi-maximizing agens who operae independenly and, in order o achieve heir goals, hey analize and valuae socks. A second assumpion is ha new informaion concerning securiies comes o he marke in a random way, and anouncemens over ime are generally independen from one anoher. A hird assumpion is ha invesors adjus securiy prices rapidly in order o reflec he effec of arrival of new informaion. The adjusmen of he securiy prices akes place rapidly because of he large number of profimaximizing agens. The join effec of informaion comming in a random, independen way and he numerous invesors who adjus sock prices rapidly in order o reflec he arrival of new informaion is ha he price changes should be independen and random, so, according o he efficien marke hypohesis, hey should reflec all available informaion, including he risk involved. According o Eugene Fama 1, aking ino accoun he way he differen ypes of informaion are refleced in securiies prices, here are hree forms of marke efficiency: he weak form, he semi-srong form and he srong form. Under he weak form of he efficien marke hypohesis, sock prices are assumed o reflec any informaion conained in he hisorical sock price. If his hypohesis is rue, a rader canno obain abnormal profi (profi above he average) by aking ino accoun only he hisorical sock price, because informaion se is already included in he sock price. As a resul, echnical analysis or charing becomes ineffecive. In his siuaion i is said ha he sock price s movemens are random walk. Under he semi-srong form of he efficien marke hypohesis, all publicly available informaion is presumed o be refleced in he securiies price. This includes informaion in he sock price series as well as informaion in he firm s accouning repors, he repors of compeing firms, announced informaion relaing o he sae of he economy, and any oher publicly available informaion relevan o he valuaion of a firm. Again, if a rader akes ino accoun only all available informaion, he/she canno obain abnormal profi, because all 1 Fama, Eugene F. (1970); Efficien Capial Markes: a Review of Theory and Empirical Work ; The Journal of Finance

4 4 available informaion se is already included in he sock prices. As a resul, fundamenal analysis becomes ineffecive. The srong form of he efficien marke hypohesis akes he noion of marke efficiency o he ulimae exreme. If his form of efficiency is rue, all informaion is refleced in he sock prices. This includes privae, or inside informaion as well as ha which is publicly available. Under his form, hose who acquire inside informaion, ac on i and quickly force he price o reflec he informaion. Bu, he iniial acquisiion of new pieces of his informaion is a maer of chance, and since he sock prices already reflec he exising invenory of inside informaion, an invesor canno obain above he average reurns. In he nex chapers I es he validiy of hose hree forms of informaional marke efficiency on he Romanian capial marke. In chaper wo I es he random walk hypohesis and he implicaion of he validiy of his assumpion on esimaing risk and volailiy. The random walk model of sock prices generaes several imporan implicaions for praciioners. Two of hem are: Firs, ha sock reurns should follow a Normal disribuion; and second, ha he risk of a securiy should 1 2 scale proporionally o he square roo of ime (he T rule). The firs implicaion is imporan because invesors need o assume a given disribuion in order o esimae he risk of heir securiies. The second implicaion is imporan because i says ha invesors can esimae he risk of a securiy in any ime inerval, and subsequenly esimae he implied risk in any oher ime inerval hrough a linear rescaling. The hird chaper presens ess of he weak form of marke efficiency: ess for seasonaliy of marke reurns, ess of he efficiency of using ransacions rules. These ess show wheher or no an invesor can obain abnormal profi using only he informaion provided by he hisory of he sock prices. In he forh chaper, here are ess for he semi-srong form of marke efficiency. Some of he bes momens o es wheher or no all he publicly available informaion are refleced in he sock prices are by looking a he reurn of he sock before and afer a sock spli and also by looking a he evoluion of he sock marke prices of a company afer i becomes publicly raded on a sock marke. In he fifh chaper I perform a es of he srong form of marke efficiency. I assume ha, if he marke were srong form efficien, a professional porfolio manager couldn obain abnormal reurns only if he/she used insider informaion.

5 5 2. Tess of he Random Walk Hypohesis The definiion of a random walk is: y = y 1 + ε, where y is a ime series and ε is a saionary random disurbance erm. The series y has a consan forecas value, condiional on, and he variance is increasing over ime. The random walk is a difference saionary series since he firs difference of y is saionary: y y ) 1 = (1 L y = ε. A difference saionary series is said o be inegraed and is denoed as I(d) where d is he order of inegraion. The order of inegraion is he number of uni roos conained in he series, or he number of differencing operaions i akes o make he series saionary. For he random walk above, here is one uni roo, so i is an I(1) series. Similarly, a saionary series is I(0). This condiion is no sufficien for a ime series o be random walk. The oher condiions are: he error erm is normally disribued and here is no linear or non-linear correlaion beween he error erms. For a series of sock marke prices, y is he logarihm of he sock prices series, and, as a resul, he firs difference y y 1 is he series of reurns. I es in his par wheher sock prices behave as a random walk using he hree main indexes on he Romanian sock markes: he BET Index, he index which ake ino accoun he evoluion of he en mos liquid companies on he Buchares Sock Marke; he BETC Index, which is he composie index of he Buchares Sock Exchange and he RASDAQC Index, he composie index of he Romanian OTC marke RASDAQ. The firs sep in analysis consiss in esing for a uni roo in he ime series. For esing, I used Augmened Dickey-Fuller and Phillips-Perron ess. Because he resuls of hose ess were similar, I will presen only he resul of he Augmened Dickey-Fuller es. y To illusrae he use of Dickey-Fuller ess, consider firs an AR(1) process: = µ + ϕy 1 + ε, where µ and ϕ are parameers and ε is assumed o be whie noise. y is a saionary series if 1 <ϕ < 1. If ϕ = 1, y is a non-saionary series. If he absolue value of ϕ is greaer han one, he series is explosive. The difference beween a uni roo (nonsaionary) series and a saionary series is ha in a saionary series, a random shock will be absorbed in ime, bu in a uni roo series, a random shock will never be absorbed. The es, performed wih Eviews 3.0 will reurn an ADF Tes Saisic, which is he es for rejecing he null hypohesis (he series is uni roo). To rejec he null hypohesis, he value of he saisic mus be less han he criical value for he chosen significance level.

6 6 Then I es if he reurns are normally disribued, and if here is or no linear or nonlinear correlaion beween he error erms (sock reurns). ε For linear dependence I used hose wo regressions: = φ0 + φ1ε 1 1 ln( I ) = µ + ρ ln( ) 1 + ε and I, where I is he value of an index in day. In order o exis linear dependence, φ should be saisical significan. This is equivalen o an ARMA(1,1) process for ln( I ). For nonlinear dependence I esed wheher or no he reurns are a GARCH(p,q) process. The form of a GARCH(p,q) model is: r = β + βl( ) + ε 0 r ε N 0, h ) ( h = α + L) h α( L) ε γ ( The finding ha sock prices do no follow a random walk has implicaions on volailiy and beas. If sock prices do no follow a random walk, he volailiy canno be 1 2 esimaed hrough a linear rescaling (he T rule). In realiy, he monhly volailiy (compued on he monhly reurns) is greaer han he volailiy compued hrough a linear rescaling. The beas, under he random walk hypohesis, should be independen from he frequency of he daa used o compue hem. Bu in realiy he monhly beas are larger han he daily beas, so he invesors would underesimae he sysemaic risk.

7 Uni Roo Tess and Dependence Tess for he Romanian Capial Marke BET Index 2 ADF Tes Level: ADF Tes Saisic % Criical Value* % Criical Value % Criical Value *MacKinnon criical values for rejecion of hypohesis of a uni roo. Augmened Dickey-Fuller Tes Equaion Dependen Variable: D(LOGBET) Mehod: Leas Squares Sample(adjused): Included observaions: 629 afer adjusing endpoins Variable Coefficien Sd. Error -Saisic Prob. LOGBET(-1) D(LOGBET(-1)) D(LOGBET(-2)) D(LOGBET(-3)) D(LOGBET(-4)) C E E R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Firs difference: ADF Tes Saisic % Criical Value* % Criical Value % Criical Value *MacKinnon criical values for rejecion of hypohesis of a uni roo. Augmened Dickey-Fuller Tes Equaion Dependen Variable: D(LOGBET,2) Mehod: Leas Squares Sample(adjused): Included observaions: 628 afer adjusing endpoins Variable Coefficien Sd. Error -Saisic Prob. D(LOGBET(-1)) D(LOGBET(-1),2) D(LOGBET(-2),2) D(LOGBET(-3),2) D(LOGBET(-4),2) C E E R-squared Mean dependen var -3.39E-05 Adjused R-squared S.D. dependen var For analisys were used he daily average values of he BET index beween 09/22/ /28/2000

8 8 S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Linear dependence ARMA(1,1): Dependen Variable: LOGBET Mehod: Leas Squares Sample(adjused): Included observaions: 633 afer adjusing endpoins Convergence achieved afer 20 ieraions Backcas: 1 Variable Coefficien Sd. Error -Saisic Prob. AR(1) MA(1) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Invered AR Roos 1.00 Invered MA Roos -.35 Nonlinear dependence: GARCH(1,1): Dependen Variable: BETRETURN Mehod: ML - ARCH Sample(adjused): Included observaions: 633 afer adjusing endpoins Convergence achieved afer 24 ieraions Backcas: 1 Coefficien Sd. Error z-saisic Prob. MA(1) Variance Equaion C E ARCH(1) GARCH(1) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Invered MA Roos -.34

9 9 Disribuion of reurns: Series: BETRETURN Sample Observaions 634 Mean Median Maximum Minimum Sd. Dev Skewness Kurosis Jarque-Bera Probabiliy According o he uni roo es, BET is an I(1) series. According o he linear dependence es, here is linear dependence beween he reurns. Also, BET is a GARCH(1,1) process. The disribuion of he reurns in no Normal, bu Lepokuroik. So, he BET series is no random walk BETC Index 3 ADF es Level ADF Tes Saisic % Criical Value* % Criical Value % Criical Value *MacKinnon criical values for rejecion of hypohesis of a uni roo. Augmened Dickey-Fuller Tes Equaion Dependen Variable: D(LOGBETC) Mehod: Leas Squares Sample(adjused): Included observaions: 491 afer adjusing endpoins Variable Coefficien Sd. Error -Saisic Prob. LOGBETC(-1) D(LOGBETC(-1)) D(LOGBETC(-2)) D(LOGBETC(-3)) D(LOGBETC(-4)) C E E R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion For analisys were used he daily average values of he BETC index beween 04/17/ /28/2000

10 10 Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Firs difference: ADF Tes Saisic % Criical Value* % Criical Value % Criical Value *MacKinnon criical values for rejecion of hypohesis of a uni roo. Augmened Dickey-Fuller Tes Equaion Dependen Variable: D(LOGBETC,2) Mehod: Leas Squares Sample(adjused): Included observaions: 490 afer adjusing endpoins Variable Coefficien Sd. Error -Saisic Prob. D(LOGBETC(-1)) D(LOGBETC(-1),2) D(LOGBETC(-2),2) D(LOGBETC(-3),2) D(LOGBETC(-4),2) C E E R-squared Mean dependen var 1.22E-05 Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Linear dependence ARMA(1,1) Dependen Variable: LOGBETC Mehod: Leas Squares Dae: 06/21/00 Time: 14:07 Sample(adjused): Included observaions: 495 afer adjusing endpoins Convergence achieved afer 28 ieraions Backcas: 1 Variable Coefficien Sd. Error -Saisic Prob. AR(1) MA(1) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Invered AR Roos 1.00 Invered MA Roos -.35

11 11 Nonlinear dependence GARCH(2,2) Dependen Variable: BETCRETURN Mehod: ML - ARCH Sample(adjused): Included observaions: 493 afer adjusing endpoins Convergence achieved afer 30 ieraions Backcas: 3 Coefficien Sd. Error z-saisic Prob. AR(2) MA(1) Variance Equaion C 7.99E E ARCH(1) ARCH(2) GARCH(1) GARCH(2) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Invered AR Roos Invered MA Roos -.54 Disribuion of reurns: Series: BETCRETURN Sample Observaions 496 Mean Median Maximum Minimum Sd. Dev Skewness Kurosis Jarque-Bera Probabiliy According o he uni roo es, BETC is an I(1) series. According o he linear dependence es, here is linear dependence beween he reurns. Also, BETC is a GARCH(2,2) process. The disribuion of he reurns in no Normal, bu Lepokuroik. So, he BETC series is no random walk.

12 RASDAQC Index ADF Tes Level ADF Tes Saisic % Criical Value* % Criical Value % Criical Value *MacKinnon criical values for rejecion of hypohesis of a uni roo. Augmened Dickey-Fuller Tes Equaion Dependen Variable: D(LOGRASDAQC) Mehod: Leas Squares Sample(adjused): Included observaions: 416 afer adjusing endpoins Variable Coefficien Sd. Error -Saisic Prob. LOGRASDAQC(-1) D(LOGRASDAQC(-1)) D(LOGRASDAQC(-2)) D(LOGRASDAQC(-3)) D(LOGRASDAQC(-4)) C E E R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Firs difference: ADF Tes Saisic % Criical Value* % Criical Value % Criical Value *MacKinnon criical values for rejecion of hypohesis of a uni roo. Augmened Dickey-Fuller Tes Equaion Dependen Variable: D(LOGRASDAQC,2) Mehod: Leas Squares Sample(adjused): Included observaions: 415 afer adjusing endpoins Variable Coefficien Sd. Error -Saisic Prob. D(LOGRASDAQC(-1)) D(LOGRASDAQC(-1),2) D(LOGRASDAQC(-2),2) D(LOGRASDAQC(-3),2) D(LOGRASDAQC(-4),2) C E E R-squared Mean dependen var 2.25E-05 Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic)

13 13 Linear dependence ARMA(1,1) Dependen Variable: LOGRASDAQC Mehod: Leas Squares Sample(adjused): Included observaions: 420 afer adjusing endpoins Convergence achieved afer 21 ieraions Backcas: 1 Variable Coefficien Sd. Error -Saisic Prob. AR(1) MA(1) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Invered AR Roos 1.00 Invered MA Roos.23 Nonlinear dependence GARCH(1,1) Dependen Variable: RASDAQRETURN Mehod: ML - ARCH Sample(adjused): Included observaions: 419 afer adjusing endpoins Convergence achieved afer 39 ieraions Backcas: 2 Coefficien Sd. Error z-saisic Prob. AR(1) MA(1) Variance Equaion C 8.42E E ARCH(1) GARCH(1) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Invered AR Roos -.74 Invered MA Roos -.63

14 14 Disribuion of reurns: Series: RASDAQCRETURN Sample Observaions 421 Mean Median Maximum Minimum Sd. Dev Skewness Kurosis Jarque-Bera Probabiliy According o he uni roo es, RASDAQC is an I(1) series wih 5% significance level. According o he linear dependence es, here is linear dependence beween he reurns. Also, RASDAQC is a GARCH(1,1) process. The disribuion of he reurns in no Normal, bu Lepokuroik. So, he RASDAQC series is no random walk Implicaions of Nonrandom Walks Volailiy If he sock prices do no follow a random walk, esimaing volailiy hrough a linear 1 2 rescaling (he T ) may be badly misleading. The able below 4 repors a quanificaion of he misakes an invesor could make if he or she esimaes monhly risk on he basis of daily daa. Symbol Daily volailiy Average day/monh Square roo of ime Imply volailiy Monhly volailiy Relaive difference ALR % ARC % ATB % AZO % DAC % ELJ % OIL % OLT % PCL % TLV % 4 For he esing were used en of he mos liquid companies raded on he Buchares Sock Exchnage in 1999

15 15 Imply volailiy is he produc beween square roo of ime and daily volailiy. Generally, imply volailiy was lower han he observed volailiy. These findings show ha, in shor horizons, volailiy scales a a faser rae han implied by he random walk, and he invesors who misakenly assuming ha sock prices follow a random walk process will underesimae risk. These findings, also shows ha volailiy has a erm srucure. For he Buchares Sock Exchange index BET 5, he erm srucure of volailiy is presened in he able below: Square roo Day of ime Volailiy Imply volailiy Relaive difference % % % % % % % % % % % % % % % % % % % % % % % % % % % % % % 5 For analisys were used he daily average values of he BET index beween 09/22/ /28/2000

16 Term srucure of volailiy volailiy volailiy imply volailiy days Beas The able below repors observed beas for he mos liquid companies on he Buchares Sock Exchange compued on he basis on boh daily daa and monhly daa. Under he random walk hypohesis, he beas should be independen from he frequency of he daa used o compue hem. However, he able shows ha, generally, monhly beas are larger han daily beas. Hence, use of daily daa o compue monhly beas will lead invesors o underesimae he sysemaic risk. Symbol Daily bea Monhly bea Relaive difference ALR % ARC % ATB % AZO % DAC % ELJ % OIL % OLT % PCL % TLV %

17 17 3. Tess of he Weak Form of Marke Efficiency 3.1. Seasonaliy of Reurns If he random walk hypohesis is valid, here should no be any consisen paerns in securiy reurns. Some sudies deec evidence of sysemaic paerns in sock reurns. In he picures below are presened he January Effec and he Weekly Effec for he Romanian capial marke. The January Effec refers o he fac ha sock reurns in January are greaer han reurns in oher monhs. An explanaion of ha fac is ax-selling hypohesis: in December, individuals sell socks ha have declined in value during he year in order o realize a capial loss for ax purposes. Then, in January hey reinves heir money, he demand rises and he reurns are greaer. Bu, his effec appeared in some counries wih differen ax legislaion. For he Buchares Sock Exchange indexes (BET, BETC) and for he OTC marke index (RASDAQC) he average monhly reurns are shown below: 1.20% BET - average monhly reurns 1.00% 0.80% 0.60% 0.40% 0.20% 0.00% -0.20% jan feb mar apr may jun jul aug sep oc nov dec -0.40% -0.60% -0.80% -1.00%

18 BETC - average monhly reurns jan feb mar apr may jun jul aug sep oc nov dec % RASDAQC - average monhly reurns 0.20% 0.00% -0.20% jan feb mar apr may jun jul aug sep oc nov dec -0.40% -0.60% -0.80% -1.00% -1.20% Alhough, he ax legislaion is differen from he counries in which he January Effec was discovered, here is a January Effec on he Buchares Sock Exchange. Bu, on he OTC marke, he January reurns are he lower han any oher monh of he year. The Weekly Effecs refers o he unusual behavior of he sock reurns on Monday versus oher days of he week. On he evolved capial markes, evidence shows ha Monday sock reurns are subsanially lower, on average, han hose on oher days of he week. An explanaion is ha firms release he bad news o he public on Friday.

19 19 The average daily reurn for he Romanian capial marke are presened below: 0.20% BET - average daily reurns 0.10% 0.00% -0.10% Mon Tue Wed Thu Fri -0.20% -0.30% -0.40% -0.50% 0.10% BETC - average daily reurns 0.05% 0.00% -0.05% Mon Tue Wed Thu Fri -0.10% -0.15% -0.20% -0.25% -0.30% -0.35% 0.20% RASDAQC - average daily reurns 0.10% 0.00% -0.10% Mon Tue Wed Thu Fri -0.20% -0.30% -0.40% -0.50%

20 20 On he Buchares Sock Exchange he lowes reurn are in he firs days of he week, bu on he OTC marke on Thursday Tess of he Trading Rules Anoher group of ess of he weak form are he es of he rading rules. Advocaes of an efficien marke hypohesized ha invesors using any echnical rading rule could no derive raes of reurn greaer han reurns from any buy and hold policy if he rading rule depended only on pas marke informaion. In a rading rule es, a simple buy and hold sraegy is compared wih he invesmen resuls of a rading rule simulaion Tess of he Filer Rules A filer rule is a mahemaical rule ha can be applied o produce buy and sell signals. In a filer rule, he sock is raded when is price change exceeds he filer se for i. One ype of filer rule: assuming an x% filer, when he sock price has risen x% from some base, he echnical analys hinks ha his movemen indicaes a breakou, meaning ha sock prices will coninue o rise (so echnical raders would acquire he sock o ake advanage of he rise). An x% decline for some peak price would be a breakou on he downside, meaning he prices will coninue o decline (so raders would sell he sock acquired previously). In he able below are he reurns of he rading rules (applied on he mos liquid shares raded on he Buchares Sock Exchange in 1999) for a range of filers from 2% o 10%. The rading commission is se o 0.5%. Symbol Buy and hold 2% 3% 4% 5% 6% 7% 8% 9% 10% ALR 89% 69% 50% 40% 23% 19% 30% 28% 28% 29% ARC 12% 4% 10% 13% 1% 1% 6% 2% 7% 17% ATB -8% -3% -3% -8% -8% -13% -24% -26% -27% -17% AZO 120% 167% 164% 171% 184% 184% 141% 126% 110% 110% DAC 20% 40% 43% 41% 37% 31% 29% 18% 25% 21% ELJ -41% -46% -45% -34% -41% -39% -34% -34% -42% -39% OIL 21% 17% 17% 11% 11% 11% 11% 11% 10% 10% OLT -12% 24% 22% 26% 30% 40% 25% 25% 25% 25% PCL -7% -42% -52% -51% -48% -53% -56% -60% -52% -53% TLV 5% 17% 9% 12% -2% 1% 1% 1% 5% 4% BETC 2% 3% 4% 3% 1% -4% -4% 5% 2% 4% Average 20% 25% 21% 22% 19% 18% 13% 9% 9% 11%

21 21 The reurns of he 2%, 3% and 4% filer are above he reurns of a buy and hold sraegy. This means ha he marke is no efficien in he weak form Tess of Moving Average Rules The moving average rule is: if he sock s price moves above is moving average by x%, buy and hold unil he price moves x% below is moving average and hen sell. In he ables below are he reurns of he differen moving average rules (applied on he mos liquid shares raded on he Buchares Sock Exchange in 1999) for 60, 100, 150 and 200-day moving average and a range of filers from 2% o 10%. The rading commission is se o 0.5%. 60-day moving average rule: Symbol Buy and hold 2% 3% 4% 5% 6% 7% 8% 9% 10% ALR 89% 41% 47% 42% 42% 41% 23% 20% 20% 20% ARC 31% 56% 49% 60% 50% 12% 9% 6% 4% 24% ATB 1% -16% -19% -29% -24% -14% -1% -1% -1% -13% AZO 122% 113% 107% 130% 122% 122% 122% 122% 120% 160% DAC 22% -25% -23% -28% -6% -7% -11% -16% -16% -20% ELJ -40% -2% -3% -6% -7% -9% -9% -13% -14% -14% OIL 25% -18% -10% -22% -24% -16% -11% -13% -13% -15% OLT -8% 43% 41% 47% 44% 44% 42% 40% 27% 27% PCL -7% -34% -34% -35% -36% -44% -45% -37% -35% -43% TLV 7% -5% 11% 11% 15% -10% -11% -13% -13% 9% BET 31% 2% 4% 9% 4% 16% 14% 8% 6% 6% Average 24% 15% 17% 17% 18% 12% 11% 10% 8% 13% 100-day moving average rule: Symbol Buy and hold 2% 3% 4% 5% 6% 7% 8% 9% 10% ALR 89% 50% 37% 36% 28% 28% 23% 23% 22% 17% ARC 31% 36% 45% 44% 36% 36% 35% 28% 32% 23% ATB 1% -19% -10% -10% 0% -3% -7% -10% -10% -16% AZO 122% 127% 127% 105% 105% 135% 135% 135% 135% 129% DAC 22% -7% -8% -11% -18% -22% -9% -9% -9% -12% ELJ -40% -12% -9% -11% -13% -14% -14% -18% -18% -2% OIL 25% -34% -36% -30% -34% -28% -2% -2% -2% -2% OLT -8% 16% 27% 30% 29% 23% 23% 23% 12% 8% PCL -7% -30% -25% -27% -34% -34% -38% -23% -24% -5% TLV 7% 26% 26% 26% 9% 9% 9% 9% -9% -9% BET 31% 12% 8% 6% 3% 1% 13% 10% 6% 6% Average 24% 15% 17% 15% 11% 13% 16% 16% 13% 13%

22 day moving average rule: Symbol Buy and hold 2% 3% 4% 5% 6% 7% 8% 9% 10% ALR 89% 38% 44% 60% 60% 43% 43% 37% 36% 28% ARC 31% 47% 44% 39% 34% 13% 13% 9% 9% 9% ATB 1% -11% -12% -16% -16% -21% -24% -11% -15% -15% AZO 122% 125% 125% 121% 121% 121% 107% 107% 107% 107% DAC 22% -13% -16% -20% -10% 0% 0% -3% -3% -6% ELJ -40% -21% -20% -13% -2% -2% -2% -2% -2% -2% OIL 25% -23% -30% -35% -41% -10% -4% -4% -4% -15% OLT -8% -7% -22% -22% -12% -2% -2% -2% -2% -2% PCL -7% -21% -17% -17% -17% -24% -18% -18% -17% -17% TLV 7% -18% -13% -13% -5% -5% -3% -10% -10% -10% BET 31% 23% 19% 17% 12% 8% 8% 3% 3% 29% Average 24% 9% 8% 8% 11% 11% 11% 10% 10% 8% 200-day moving average rule: Symbol Buy and hold 2% 3% 4% 5% 6% 7% 8% 9% 10% ALR 89% 68% 68% 68% 68% 68% 66% 68% 68% 68% ARC 31% 17% 16% 16% 16% 13% 13% 13% 13% 9% ATB 1% -12% -16% -20% -24% -25% -29% -29% -28% -15% AZO 122% 103% 103% 95% 95% 95% 95% 87% 87% 87% DAC 22% -9% -4% -4% 10% 1% 1% 1% 21% 21% ELJ -40% -2% -2% -2% -2% -2% -2% -2% -2% -26% OIL 25% -28% -25% -19% -28% -29% -24% -28% -28% -8% OLT -8% -2% -2% -2% -2% -2% -2% -2% -2% -2% PCL -7% -20% -26% -27% -38% -28% -34% -34% -34% -34% TLV 7% -10% -10% -10% -10% -10% -12% -12% -12% -14% BET 31% 35% 35% 35% 31% 31% 31% 27% 27% 27% Average 24% 11% 10% 10% 9% 8% 7% 6% 8% 9% The reurns of he moving average rules are below he reurns of a buy and hold sraegy, and, according o his es a rader canno obain higher han average profis using moving average rules.

23 23 4. Tess of he Semi-srong Form of Marke Efficiency 4.1. The Response of Sock Prices o he Announcemen of a Sock Spli The mehodology for sudying he response of sock prices o he announcemen of a sock spli was employed for he firs ime by Fama, Fisher, Jensen and Roll (FFJR) in According o he CAPM, sock reurns are affeced by boh aggregae-marke and companyunique informaion. In an aemp o isolae ha par of a securiy s reurn, which was unique, o company evens alone, FFJR examined he residual errors from he marke model: ~ ~ ~ R = a + b( RM ) + e~, where R ~ is he reurn on day ; a is he consan average daily reurn; b he bea esimae for he sock; R ~ M ~ - he reurn of he aggregae marke porfolio during period and e~ - he residual error in period, he proporion of he reurn due o firm-unique evens. Esimaes of a and b can be developed using a regression equaion relaing sock s hisorical reurn o hisorical marke reurn. The e~ values for each sock spli during he period before he sock spli and afer he sock spli, are calculaed by using esimaes of he ei, i= a s and b s. Then, he average marke model residual in monh is calculaed: AR = 1, N where AR is he average firm-unique reurn for monh. Second, a cumulaive average firmunique reurn (CAR) was calculaed for each monh: CAR = AR K. K = 29 Tesing virually all splis on he Unied Saes capial marke beween 1927 and 1959, FFJR found ha: (1) Socks ha spli appear o have had a dramaic increase in price during he 29 monhs prior o he spli. This is refleced in he subsanial growh in he CAR prior o he spli dae. However, hese price increases canno be aribued o he evenual spli, since rarely was a spli announced more han four monhs prior o he effecive dae of he spli. (2) Afer he spli dae, he CAR is remarkably sable. This implies ha from he spli dae forward, firm-unique reurns were zero. The spli had no immediae or long run impac on securiy prices. On he Buchares Sock Exchange, only four sock splis ook place (companies Banca Agricola, Impac and Imsa spli heir shares). The cumulaive average firm-unique reurns are presened below. According o hese ess, Romanian capial marke is no semi-srong form efficien. N

24 24 Banca Agricola (AGR) 6 The sock spli ook place on Sepember 29, ~ ~ ~ R = ~ before he sock spli: Regression a + b( RM ) + e Dependen Variable: AGRRET Mehod: Leas Squares Sample: Included observaions: 203 AGRRET=C(1)+C(2)*BETCRET Coefficien Sd. Error -Saisic Prob. C(1) C(2) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) ~ ~ ~ R = ~ afer he sock spli: Regression a + b( RM ) + e Dependen Variable: AGRRET Mehod: Leas Squares Sample: 1 72 Included observaions: 72 AGRRET=C(1)+C(2)*BETCRET Coefficien Sd. Error -Saisic Prob. C(1) C(2) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Table of average firm-unique reurn (AR) and cumulaive average firm-unique reurn (CAR): MonhAR CAR % -3.09% % -1.34% % -0.90% % -0.69% % -2.13% % -4.99% % -4.81% % -3.45% % -3.31% % -1.22% 6 For he calculaion of he CAR were used he daily average prices beween December 1998 January 2000

25 % -3.62% % -1.51% % -3.76% % 0.08% 1.00% CAR of AGR 0.00% -1.00% % -3.00% -4.00% -5.00% -6.00% The sock spli ook place in he enh monh. Before he spli, he CAR rose, and afer he spli CAR flucuaed, bu in a cerain limis. The evoluion of CAR afer he spli may no conradic he efficien marke hypohesis. Imsa (IMS) 7 The sock spli ook place on he February 28, ~ ~ ~ R = ~ before he sock spli: Regression a + b( RM ) + e Dependen Variable: RETIMS Mehod: Leas Squares Sample: Included observaions: 462 RETIMS=C(1)+C(2)*RETBETC Coefficien Sd. Error -Saisic Prob. C(1) C(2) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) For he calculaion of he CAR were used he daily average prices beween April 1998 April 2000

26 26 ~ ~ ~ R = ~ afer he sock spli: Regression a + b( RM ) + e Dependen Variable: RETIMS Mehod: Leas Squares Sample: 1 41 Included observaions: 41 RETIMS=C(1)+C(2)*RETBETC Coefficien Sd. Error -Saisic Prob. C(1) C(2) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Table of average firm-unique reurn (AR) and cumulaive average firm-unique reurn (CAR): Monh AR CAR % 0.11% % -0.38% % -1.03% % -4.10% % -4.14% % -4.09% % -3.87% % -2.97% % -2.10% % -2.08% % 1.61% % 3.12% % 2.56% % 4.60% % 5.49% % 4.31% % 4.96% % 4.66% % 5.07% % 6.42% % 4.75% % 4.43% % 5.78% % 8.46%

27 % CAR of IMS 8.00% 6.00% 4.00% 2.00% 0.00% -2.00% % -6.00% Sock spli ook place in he wenysecond monh. Afer he spli, CAR raised, which conradics he efficien marke hypohesis. A possible explanaion is ha he company ook he decision o spli he shares in order o make he shares more accessible o he general public. Impac (IMP) 8 The sock splis ook place on Augus and November ~ ~ ~ R = ~ before he sock spli on Augus : Regression a + b( RM ) + e Dependen Variable: IMPRET Mehod: Leas Squares Sample: 1 72 Included observaions: 72 IMPRET=C(1)+C(2)*BETCRET Coefficien Sd. Error -Saisic Prob. C(1) C(2) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) For he calculaion of he CAR were used he daily average prices beween April 1998 April 2000

28 28 ~ ~ ~ R = ~ beween Augus and November : Regression a + b( RM ) + e Dependen Variable: IMPRET Mehod: Leas Squares Sample: Included observaions: 327 IMPRET=C(1)+C(2)*BETCRET Coefficien Sd. Error -Saisic Prob. C(1) C(2) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) ~ ~ ~ R = ~ afer November : Regression a + b( RM ) + e Dependen Variable: IMPRET Mehod: Leas Squares Sample: Included observaions: 104 IMPRET=C(1)+C(2)*BETCRET Coefficien Sd. Error -Saisic Prob. C(1) C(2) R-squared Mean dependen var Adjused R-squared S.D. dependen var S.E. of regression Akaike info crierion Sum squared resid Schwarz crierion Log likelihood F-saisic Durbin-Wason sa Prob(F-saisic) Table of average firm-unique reurn (AR) and cumulaive average firm-unique reurn (CAR): Monh AR CAR % -0.05% % -0.56% % 0.71% % -0.08% % -3.63% % -4.40% % -4.51% % -4.61% % -3.54% % -1.52% % -1.50% % -2.08% % -5.59% % -5.29% % -4.08% % -2.57% % -0.22%

29 % 0.22% % -0.07% % -3.96% % -3.41% % -5.45% % -3.01% % -2.58% % -2.05% 2.00% CAR of IMP 1.00% 0.00% -1.00% % -3.00% -4.00% -5.00% -6.00% Splis ook place in he fourh monh and in he nineeenh monh. Afer he firs sock spli, CAR decreased and afer he second spli he CAR raised. Boh reacions conradic he efficien marke hypohesis The Response of he Sock Prices afer a New Company is lised on a Sock Exchange Anoher economic even ha is expeced o have a significan impac on a firm and is sock is he decision o become lised on a naional exchange. There are wo quesions of ineres. Firs, does he lising on a major exchange permanenly increase he value of he firm? Second, given he change in expecaions or percepions surrounding he lising, i is possible o derive abnormal reurns from invesing in he sock a he ime of acual lising? According o he efficien marke hypohesis, an invesor canno obain abnormal profis from his even. In he picures below is shown he evoluion of he share price of five Financial Invesmen Companies (SIF) afer heir lising on he Buchares Sock Exchange. Those companies are closed end funds, and because of ha, he real value of heir socks is easy o esimae, because heir porfolio is accessible o he general public.

30 30 SIF Transilvania Brasov nov nov nov nov dec dec dec ian ian feb feb feb feb mar mar mar apr apr apr.00 mil ROL ROL volume price SIF Olenia Craiova nov nov nov dec dec dec ian ian feb feb feb mar mar mar apr apr apr.00 mil ROL ROL volume price SIF Munenia Bucuresi nov nov nov nov dec dec dec ian ian feb feb feb feb mar mar mar apr apr apr.00 mil ROL ROL volume price

31 31 SIF Moldova Bacau volume price mil ROL mil. ROL nov nov nov dec dec dec ian ian feb feb.00 SIF Bana Crisana 24.feb mar mar mar apr apr apr.00 volume ROL price nov nov nov nov dec dec dec ian ian feb feb feb feb mar mar mar apr apr apr ROL Afer he lising, he sock price of all companies rose sharply, and hen i decreased sharply and became sable. These evens sugges ha i was an over-reacion of he marke, which conradics he semi-srong form of marke efficiency. The abnormal price changes could generae large abnormal profis (afer aking ino accoun he ransacion coss), which also conradic he semi-srong form of marke efficiency.

32 32 5. Tess of he Srong Form of Marke Efficiency One caegory of individuals who could have monopolisic access o he informaion, and heir invesmen performance can be measured are porfolio managers. If he srong form of marke efficiency is valid, he reurn of a muual fund canno exceed he reurn of a buy and hold sraegy. For esing he srong form of marke efficiency for he Romanian capial marke, I compared he performance of he muual funds in 1999 wih he reurn of a buy and hold sraegy. The excess reurn is he difference beween he realized reurns of he muual funds and he reurn of a buy and hold sraegy. If he excess reurn is significan and posiive, he porfolio managers could have monopolisic access o informaion. The muual funds used were: - Acive Clasic - Acive Dinamic - Acive Junior - Ardaf - Armonia - Capial Plus - FCE - FIDE - Foruna Clasic - Sabilo - Tezaur - Transilvania

33 33 Aggregae porfolio srucure of muual funds: Monh Cash (%) Treasury Bonds (%) Deposis (%) Socks (%) Oher insrumens (%) Jan Feb Mar Apr May Jun Jul Aug Sep Oc Nov Dec Muual funds aggregae realized reurn: Monh Monhly realized reurn Annualized reurn Jan % 68% Feb % 103% Mar % 131% Apr % 127% May % 134% Jun % 130% Jul % 117% Aug % 91% Sep % 68% Oc % 59% Nov % 54% Dec % 61% Annualized raes of reurn for a buy and hold sraegy: Monh Curren Accoun Treasury Bonds Bank Deposis Socks Oher Annual Reurn Insrumens (buy and hold) Jan 5.00% 70.36% 68.06% 1.68% 70.36% 66.01% Feb 5.00% 89.57% % -2.16% % % Mar 5.00% 78.51% % -5.98% % % Apr 5.00% % % -5.54% % % May 5.00% % 84.97% 13.47% % 91.12% Jun 5.00% % 88.84% 16.47% % 95.09% Jul 5.00% 74.40% 72.54% -3.28% 74.40% 72.29% Aug 5.00% 68.99% 54.63% 10.91% 68.99% 64.35% Sep 5.00% 55.85% 42.74% -7.45% 55.85% 51.14% Oc 5.00% 52.11% 46.40% 0.50% 52.11% 50.29% Nov 5.00% 62.79% 50.61% -4.36% 62.79% 59.62% Dec 5.00% 73.47% 65.78% -4.90% 73.47% 71.70%

34 34 The rae of reurn for oher insrumens is se o he maximum beween he raes of reurn for curren accoun, reasury bonds, bank deposis and socks. The rae of reurn for socks is se o he rae of reurn of he Buchares Sock Exchange Composie Index (BETC). Excess reurn of he muual funds: Monh Reurn Buy and Hold Reurn Excess Reurn Jan 67.60% 66.01% 1.59% Feb % % 2.87% Mar % % 28.92% Apr % % 6.47% May % 91.12% 42.72% Jun % 95.09% 34.96% Jul % 72.29% 44.75% Aug 90.81% 64.35% 26.46% Sep 68.45% 51.14% 17.31% Oc 58.96% 50.29% 8.66% Nov 53.94% 59.62% -5.67% Dec 61.16% 71.70% % Average 95.27% 78.73% 16.54% 100% Buy and Hold Reurn Excess Reurn 80% 60% 40% 20% 0% -20% Jan Feb Mar Apr May Jun Jul Aug Sep Oc Nov Dec Average The excess reurns are significan, which conradics he srong form of marke efficiency.

35 35 6. Alernaives o he Efficien Marke Hypohesis Peers in proposed a heory Fracal Marke Hypohesis - in which he said ha informaion didn have a uniform impac on sock prices; each invesor according o his invesmen horizon assimilaes i differenly. Peers' heory proposes he following: 1. The marke is sable when i consiss of invesors covering a large number of invesmen horizons. 2. The informaion se is more relaed o marke senimen and echnical facors in he shor erm han in he longer erm. As invesmen horizons increase, longer-erm fundamenal informaion dominaes. Thus, price changes may reflec informaion imporan only o ha invesmen horizon 3. In an even occurs ha makes he validiy of fundamenal informaion quesionable, long-erm invesors eiher sop paricipaing in he marke or begin rading based on he shor-erm informaion se. When he overall invesmen horizon of he marke shrinks o a uniform level, he marke becomes unsable. There are no long-erm invesors o sabilize he marke by offering liquidiy o shor-erm invesors. 4. Price reflecs a combinaion of shor-erm echnical rading and long-erm fundamenal valuaion. Thus, shor-erm price changes are likely o be more volaile, or noisier, han long erm rades. The underlying rend in he marke is reflecive of changes in expeced earnings, based on he changing economic environmen. Shor-erm rends are more likely he resul of crowd behavior. There is no reason o believe ha he lengh of he shor-erm rends is relaed o he long-erm economic rend. 5. If a securiy has no ie o he economic cycle, hen here will be no long-erm rend. Trading, liquidiy and shor-erm informaion will dominae. 9 Peers, Edgar E. (1994); Fracal Marke Analysis. Applying Chaos Theory o Invesmen and Economics ; John Wiley & Sons, Inc.