sign in sign up
<<
Home , Albanian lek, Bank of Albania

A Service of

Leibniz-Informationszentrum econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible. zbw for Economics

Çera, Gentjan; Dokle, Eda; Çera, Edmond

Article Do the News Affect the EUR/ALL Exchange Rate Volatility?

Economic Review: Journal of Economics and Business

Provided in Cooperation with: Faculty of Economics, University of Tuzla

Suggested Citation: Çera, Gentjan; Dokle, Eda; Çera, Edmond (2015) : Do the News Affect the EUR/ALL Exchange Rate Volatility?, Economic Review: Journal of Economics and Business, ISSN 1512-8962, University of Tuzla, Faculty of Economics, Tuzla, Vol. 13, Iss. 1, pp. 21-28

This Version is available at: http://hdl.handle.net/10419/193843

Standard-Nutzungsbedingungen: Terms of use:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated licence. www.econstor.eu Economic Review - Journal of Economics and Business, Vol. XIII, Issue 1, May 2015

DO THE NEWS AFFECT THE EUR/ALL EXCHANGE RATE VOLATILITY?

ABSTRACTGentjan ÇERA*, Eda DOKLE**, Edmond ÇERA***positive transition steps undertaken in terms - Since the early 1990s, has adopted the of the exchange transition rate. process, Albania because opted for it had the a float lim- - iteding exchangeamount of rate international regime since reserves the first and years was flexible exchange rate regime. A vast empirical literature on exchange rate is focused on model ing its volatility. In contrast, this paper provides trying to keep away from any exchange rate empirical analysis regarding the news impact Moreover,misalignment in 2009 that thewould Bank be of too Albania costly decided for the threeon the importantEUR/ALL exchange features ofrate asset volatility, return using pro- small and fragile economy (Luci & Vika, 2011). posedTGARCH by model.the theory: We argueunpredictability, that the series fat tails has - portantto adopt precondition. the inflation targeting regime which has ‘free floating exchange rate’ listed as an im and volatility clustering. The results show the existence and importance of news impact on - The general study of exchange rate in Albania is Keywords:exchange rate Albanian return. lek, EUR/ALL, news im- important for several reasons. First, the finan pact, TGARCH cial sector is highly affected by the exchange rate because of the large part of loan portfolio JEL classification: F31, C58 being in foreign , which exposes the banking system towards a high currency risk. 1 INTRODUCTION According to the data from the , by the end of the 2014 first quarter, the loans - total loan portfolio in the country, while by the endin foreign of May currency 2014, the amounted loans in to 60.3% amounted of the to becauseThe exchange of its relevance rate volatility both haswith become the corpo in- creasingly important in developing countries- 39.8% of the total portfolio in the country. cymakers. While the former use it when eval- rate decision making bodies and with the poli ofSecond, its trade international with other trade countries is highly in euro. affected In this by focused more on the impact it has on the macro the exchange rate since Albania conducts most uating their portfolios, the policymakers are- regard, Figure 1.1 shows the progress of GDP level. Exchange rate can be either fixed or float and total trade (measured as sum of exports ing. A fixed exchange rate is decided and not rate is more volatile and may shift up or down and imports) starting from 2004 until 2013.- allowed to move, while the floating exchange In 2012, the total trade amounted to 52.95%- of GDP, highlighting the importance of the ex depending on the economic shocks. change rate in the international trade. In addi tion, other central actors highly affected by the Albania switched from being a closed economy- exchange rate risk are the 339 official exchange tionto an to open an open one economy in the early has ended 1990s. or Although not, it is know a currency trend or a possible forecast. rate offices in Albaniawhich are interested to importantseveral economists to have a argueclose look whether at the the two transi main

*Faculty of Economy and Agribusiness, [email protected]:[email protected] ** Institute of Economic Studies, Charles University in Prague [email protected] ***Faculty of Economy, University of [email protected]

Economic Review - Journal of Economics and Business, Vol. XIII, Issue 1, May 2015 21 G. Çera, E. Dokle, E. Çera the relationship between the “conditional vari- In their study, Engle and Ng (1993) examined

showedance of proof asset of returns asymmetry to exogenous in stock returns.shocks”, named also ‘news impact curve’. The results

In 1963, Mandelbrot introduced the concept of volatility clustering referring to the idea that “large changes tend to be followed by large changesof either sign and small changes tend to be followed by small changes.” Figure 1.1. The international trade (export + import) The leverage effects are largely known as the and annual GDP trends (in millions, EUR) - negative relationship between an asset price- and its return. The term was first introduced whereThe exchange the rates trade between between different arecurren de- by Black (1976) who tried to address the asym cies is performed in exchange rate markets metry found by Engle and Ng (1993). He argued- - that a fall in the stock prices originates from renciesfined and does trade not is facilitated. follow the However, economic the theory ALL a decline in firm’s equity, causing a direct in- () exchange rate with other cur sociatedcrease in the the increase leverage level (debt of to risk equity with ratio) the pres and- signaling the riskiness of the firm. Also, he as since the market equilibrium does not fully and significantly depend on the exchange rate ence of high volatility. This remark is generally (Hoda, 2012). accepted as a conclusion that negative shocks are followed by high volatility, while positive It is generally known that exchange rate affects shocks are followed by low volatility. Moving the behavior of macroeconomic agents in the away from the asset return case and focusing Albanian economy (Istrefi & Semi, 2009 and not found any study focused on the news im- more on the exchange rate volatility, Longmore Vika & Luçi, 2011), but in our research we have and Robinson (2004) noted that the magnitude we initiated a study on the possible predictabil- of exchange rate volatility depends on whether pact on the exchange rate volatility. Therefore the shock is positive or negative. to the use of the ARCH and GARCH models and ity of the EUR/ALL exchange rate volatility and Observations of this type of times series has led the importance of the news in exchange rate conductsvolatility. The with focus Eurozone on the countries EUR/ALL and exchange other their extensions. Mundaca (1991) used ARCH rate was mainly due to the large trade Albania rateand GARCH observed models twice to daily. examine She foundthe Norwegian out that thecurrency conditional basket variance and the was NOK/USD smaller exchange for the countries using euro as a currency in their transactions. The main research questions this study aims to address are: Do the news affect thebasket hour than of the for day the the NOK/USD observation exchange was taken. rate the EUR/ALL exchange rate volatility? Which is Moreand that importantly, these variances she concluded changed dependingthat three seon- 2the LITERATURE effect of bad REVIEW news on it?

ries, out of four, were better fitted by GARCH. - Chong, Chun and Ahmad (2002) used GARCH- kurtosisDifferent and empirical fat tails instudies distribution, on asset which returns models and its modification to study the Ma have noticed the presence of skewness, excess- laysian Ringgit/Sterling exchange rate volatil ity for the period 1990-1997. The results show brings up the assumption that identical and in that the exchange rate volatility is persistent dependently distributed (i.d.d.) errors do not- walkand theas a constant standard variance model, the model GARCH is rejected. models work properly for the asset returns (Longmore They highlight the fact that if using the random and Robinson, 2004). Therefore, three impor tant issues are generally discussed: asymmetry, perform better when forecasting the volatility- volatility clusters and leverage effect. of the exchange rate. Later on, Insah (2013) 22 Economic Review - Journal of Economicsstudied the and nature Business, of the Vol. exchange XIII, Issue rate1, May volatil 2015 GARCH model with volatility breaks. When using the TGARCH model, they observed that GARCHthe analysis model of thewith two volatility types of breaks. news showed When usingthat the the news TGARCH impact model,is asymmetric. they observed that the analysis of the two types of news showed that the news impact is asymmetric. A study was conducted by Jithitikulchai (2005) on the exchange rate volatility of Baht/USD for Athe study period was between conducted 1990 by and Jithitikulchai 2005 based (2005) on weekly on the data. exchange The author rate volatility used TGARCH of Baht/USD to account for thefor period the response between of 1990 the series and 2005 to good based and on bad weekly news. data. The The results author show use thatd TGARCH the negative to account impact forof the badresponse news ofis notthe significantseries to good although and bad the news. direction The seems results coherent. show that However, the negative he found impact out ofalso the that bad TGARCH news is not modelled significant best although the out theof sample direction exchange seems coherent. rate volatility. However, he found out also thatGARCH TGARCH model modelled with volatility best the breaks. out of When sample using exchange the TGARCH rate volatility. model, they observed that Regardingthe analysis the case of theof Albania,two types Çera,of news Çera showed and that Lito the (2013) news impact conducted is asymmetric. a study on the value at Regardingrisk for EUR/ALL the case of exchangeAlbania, Çera, rate. Çera They and used Lito GARCH(1,1) (2013) conducted to analyze a study daily on the data value of theat riskexchange forA EUR/ALLstudy rate. was The conducted exchangestudy addressedby Jithitikulchai rate. They the main(2005) used agentson GARCH(1,1) the exchange conducting rate to analyzevolatilityeconomic of daily Baht/USDactivity data in offor EUR the exchangeand workingthe rate.period in The thebetween Albanianstudy 1990 addressed and economy. 2005 thebased It main is on important weekly agents data. conductingto Theemphasize author economicuse thatd TGARCH it precedes activity to account our in study. EUR and workingfor the in response the Albanian of the series economy. to good It and is importantbad news. Theto emphasize results show that that it the precedes negative ourimpact study. 3 METHODOLOGYof the bad news is not ANDsignificant THEORETICAL although the direction MODELS seems coherent. However, he found out 3 METHODOLOGYalso that TGARCH modelled AND THEORETICAL best the out of sample MODELS exchange rate volatility. 3.1 DifferentRegarding themodels case ofto Albania,measure Çera, risk Çera and Lito (2013) conducted a study on the value at 3.1 Differentrisk for EUR/ALL models to exchange measure rate. risk They used GARCH(1,1) to analyze daily data of the For a exchange long time, rate. econometricians The study addressed were the mainly main agents studying conducting how aeconomic variable activity would in change EUR in Forresponse a longand to working time,a change econometriciansin the in Albanian another economy.variable were mainlyItand is importantthe studyingproper to emphasizemodel how to a variablethatanswer it precedes their would questionsour change study. was in responsethe least tosquares a change model. in another However, variable Engel and (2001) the proper argues model that economists to answer theirwere questionsfaced with was the 3 METHODOLOGY AND THEORETICAL MODELS thechallenge least squares of forecasting model.Do the However, the news size affect of Engelthe theeur/all error(2001) exchange and argues rate finding volatility? that aeconomists proper model were that faced would with allow the challengethem to answer of forecasting questions the about size volatility.of the error At anda time finding when a hetereoscedasticity proper model that waswould becoming allow models to measure3.1 volatility.Different The models later was to measure risk ity inthem a Ghana problem to using answer ARCH(1) of least questions and squares, GARCH(1,1) about it waswhotime volatility. when inRobert 1982 hetereoscedasticity introduced AtEngel a time who a new when was in model becoming1982 hetereoscedasticity called introduced a a new was model becoming called volatility in Ghana.For a long time, econometriciansproblem of were least mainlysquares, it studying was Robert how Engel - a variable would change in founda ARCHto problem be the proper (Autoregressive of model least for squares, exchange Conditional rate it was Robert Heteroscedasticity) Engel who in 1982 which introduced proposes a new a waymodel to called model response to a change in another variable and the proper model to- answer their questions was ARCHheteroscedasticy. (Autoregressive Conditionaleroscedasticy.ARCH (Autoregressive Heteroscedasticity) Conditional Heterosce which proposes a way to model the least squares model. However,Thedasticity) mathematical Engelwhich proposes (2001) form of aargues way the to ARCH modelthat model economists het were faced with the To accommodateheteroscedasticy. the leverage effects, new modelsThe started mathematical challengebeing used: TGARCH,of forecasting form EGARCH, of thethe size ARCH of the model error comesand finding from a proper the AR model (Autoregressive) that would allow model - which is presented as follows: and PARCH. Balathem and Asemotato answer (2003) questions studied about volatility. At a time when hetereoscedasticity was becoming Thewhich mathematical is presented formas follows: of the- comes ARCH from model the AR comes (Autoregres fromsive) model the AR (Autoregressive) model the exchange ratea problem volatility of in Nigeria,least squares, apply it was Robert Engel who in 1982 introduced a new model called ing GARCHwhich models is presented with exogenous as follows: break. Us ARCH (Autoregressive Conditional Heteroscedasticity)d which proposes a way to model ing monthly data on exchange rate returns for Y a d Ya  H models,three foreign and concluded currenciesheteroscedasticy. that (USD, better EUR, results and GBP), were t 0 ¦ iti t Y a  i 1 Ya  H obtainedthey compared with the results application of different of the GARCH GARCH t 0 ¦ iti t (1) The mathematical form of the ARCH model comes from the AR (Autoregressive) model i 1 2 TGARCHTo model, obtainwhich they theobserved is ARCH(presented that theq )analysisasmodel, follows: Engel (1982) suggests expressing the variance (ı )as a linear - 2 ofmodel theTo two with obtain types volatility of newsthe breaks. ARCH(showed When thatq using)themodel, news the Engel (1982)q suggests expressing the variance (ı )as a linear function of past squared innovations,To obtain where the ARCH(q) denotesd model, the Engel lag (1982) order. impactfunction is asymmetric. of past squared innovations,suggests where expressing qdenotes the variance the (σ2)as lag order.a lin Yt a0  ¦ Ya iti  H t ear function of past squaredq innovations, where 2 i 1 2 qdenotes theV t lag order.Z  q a H iti 2 ¦ 2 2 theA study period was between conductedTo obtain 1990 by Jithitikulchaiand the 2005 ARCH( based (2005)q )onmodel, EngelV t (1982) Z  ¦ suggestsi 1a H iti expressing the variance (ı )as a linear weeklyon the exchange data. Thefunction rate author volatility usedof past of TGARCH Baht/USD squared to acfor innovations,- where qidenotes 1 the lag order. This model was largely used to describe possible volatility(2) clustering where the sign of This model was largely used to describe possibleq volatility clustering- where the sign of count changesfor the response in of price the series followed to good and by other changes2 in price2 was unpredictable. Since in practice - V t Z  ¦a H iti bad news.changesARCH( The results q in) price models show that followed the require negative by a otherThis large model changes number was largely in of i used price1 lags to describe was and unpredictable. pos parameters, SinceBollerslev in practice (1986) heimpact foundARCH(introduced of out the also badq) that news models aTGARCH new is not modelmodelled significant require best called al a largeinsible GARCH.price volatility numberwas unpredictable. clustering In theseof lags where Since cases, andthe in practice sign tKH parameters, of WHUP µFRQGLWLRQDO Bollerslev YDULDQFH¶ (1986)is thoughintroduced the directionThis seems a model new coherent. model was However, largely called usedchanges GARCH. to describein price In followed these possible by cases, othervolatility t changes clustering where the sign of is used tochanges refer to in the price forecast followeded byvariance, other changes which inis pricebased was onKH unpredictable.past WHUP information. µFRQGLWLRQDO Since This in YDULDQFH¶ practice model, in the outusedwhich of sample to ARCH(prefer isexchange the toq order) ratethe models volatility. forecast of GARCH requireed variance, aARCH(q) anew(lagged large model models numberwhichcalledvolatility) require GARCH. is ofa largebased Interms lags these number on and cases,and pastof parameters,lags theqis information. the order Bollerslev of This ARCH model, (1986) (lagged in Regardingwhich the pcaseis theof Albania, order Çera, of GARCHÇera and and (lagged parameters, volatility) Bollerslev terms(1986) introducedand qis the order of ARCH (lagged Lito (2013)of squared conductedintroduced error) a study aterms, on new the modelvalue can atbe- called thepresented forecasted GARCH. variance,as follows: In these which cases,is based tonKH past WHUP µFRQGLWLRQDO YDULDQFH¶ is risk of for squared EUR/ALLused exchangeerror) to refer terms, rate. to the They can forecast used be presentedinformation.termed variance, ‘conditional Thisas whichfollows: model,variance’ inis which is based used pis toon the refer pastorder to information. This model, in GARCH(1,1) to analyze daily data of the ex q p which pis the order of GARCH2 (lagged volatility)2 terms and2 qis the order of ARCH (lagged change rate. The study addressed the main V t Z  q¦a H iti  p¦ E V  jtj importantagents conducting to emphasizeof squared economic that error) activityit precedes terms, in EUR our can of be GARCH 2presented (lagged as volatility) follows:2 terms and2 qis the canV t be presentedZ  ¦i 1asa follows:H iti  ¦j 1E V  jtj study.and working in the Albanian economy. It is order of ARCHi (lagged1 q of squaredj 1p error) terms, Later on, the idea that asymmetric Veffect2 Z of shocksa H 2  areE independentV 2 was formulated with the 3 METHODOLOGY AND THEORETICAL t ¦ iti ¦  jtj Later on, the idea that asymmetric effect of ishocks 1 arej 1 independent was formulated with the MODELS (3) Later on, the idea that asymmetric effect of shocks are independent was formulated with the 3.1 Different models to measure risk shocks are independent was formulated with Later on, the idea that asymmetric effect - of

- the introduction of the TARCH (Threshold Au- For a long time, econometriciansintroduction were of mainly the TARCH toregressive (Threshold Conditional Autoregressive Heteroscedasticity) Conditional Heteroscedasticity) studying how a variable would change in re model by Zakoïan (1994)model and by Glosten, Zakoïan Jaganathanand (1994) and Glosten, Runkle Jaga (1993)- independently. The sponse to a change in another variable and general specification of thlows:nathanandis model is Runkle as follows: (1993) independently. The the proper model to answer their questions general specification of this model is as fol was the least squares model. However, Engel q p r 2 2 2 2 (2001) argues that economists were faced with ǔt Z  ¦a dž iti  ¦ ǔǃ  jtj  ¦ džDŽ  I ktktk the challenge of forecasting the size of the error i 1 j 1 k 1 and finding a proper model that would allow (4) whereIt is a dummy variable taking value 1 for İt< 0 or 0 otherwise. This model introduces themEconomic to answer Review questions - Journal of about Economics volatility. and Business, At a Vol. XIII, Issue 1, May 2015 23 the concepts of good news (İt-i> 0) and bad news (İt-i< 0), which have different effects on the conditional variance. The good news effect is measured by aicoefficient while the bad news effect is measured by ai + Ȗi because It = 1. In this case, if Ȗi> 0, the bad news causes an increase in variance and has a leverage effect of order i, but if Ȗi   WKH QHZV LPSDFW LV DV\PPHWULF7KHµQHZV¶FRQFHSWUHIHUVWRWKHLQIRUPDWLRQRISDVWSHULRGVZKLFKDIIHFWVWKH current period. Themodel can be written as TGARCH(r,p,q), where q, p and r are the orders of ARCH, GARCH and TGARCH terms. It is easily noticeable that GARCH model is a specific case of TARCH with an It equal to 0. The ARCH family models include other models with specific restrictions like the IGARCH model which is estimated by removing the free constant and where the sum of coefficients a andȕ is 1. By making other forms of restrictions we can obtain other ARCH models like: EGARCH, PARCH, CGARCH, GJARCH etc.

3.2 The data The data employed consists of average daily data of the EUR/ALL exchange rate presented in the form of dynamic series over a time horizon between January 01, 2002 and July 10, 2014.The data is taken from the official webpage of the Bank of Albania where daily ALL exchange rates with other currencies are archived1. The total number of observations is 3,131. The daily exchange rate return is defined as:

EUR / ALLt ' ln(EUR / ALLt ) ln EUR / ALLt1

wKHUHǻOQ (85$// LVWKHUHWXUQRQH[FKDQJHUDWHZKLOH(85$//t is the exchange rate at time t, and EUR/ALLt-1 is the exchange rate at time t± 1. Our empiric analysis is computed by using EViews7, a professional computer program for econometric analysis.

4 FINDINGS

4.1 Descriptive statistics Figure 2 shows time series plots of EUR/ALL exchange rate in the upper graph and the exchange rate return in the below graph. From the graphs in Figure 2, three periods stand out: 1) The first period between January 2002 and June2004, is characterized by an average exchange rate of 134. During this period the financial system was not fully consolidated and the Albanian economy was already showing signs of growth.

1 The Bank of Albania publishes the exchange rates four to five times a week. G. Çera, E. Dokle, E. Çera 4 FINDINGS

whereIt is a dummy variable taking value 1 for 4.1 Descriptive statistics εt< 0 or 0 otherwise. This model introduces the concepts of good news (εt-i> 0) and bad- news (εt-i< 0), which have different effects on - the conditional variance. The good news ef Figure 2 shows time series plots of EUR/ALL fect is measured by aicoefficient while the bad exchange rate in the upper graph and the ex introduction of the TARCHnews effect (Threshold is measured Autoregressive by ai + Conditional γi because- It change Heteroscedasticity rate return ) in the below graph. From = 1. In this case, if γi> 0, the bad news causes the graphs in Figure 2, three periods stand out: model by Zakoïan (1994) and Glosten, Jaganathanand Runkle (1993) independently. The - an increase in variance and has a leverage ef general specification of this model is as follows: 1) The first period between January 2002 and- fect of order i, but if γi ≠ 0 the news impact is nancialJune2004, system is characterized was not fully by consolidated an average and ex currentasymmetric. period.q The Themodel ‘news’p concept can r be refers written to the as 2 2 2 2 change rate of 134. During this period the fi informationǔt Z  ¦ ofa dž pastiti  ¦ periodsǔǃ  jtj  which¦ džDŽ affects I ktktk the i 1 j 1 k 1 of ARCH, GARCH and TGARCH terms. It is easily the Albanian economy was already showing TGARCH(r,p,q), where q, p and r are the orders whereIt is a dummy variable taking value 1 for İt< 0 or 0 otherwise. Thissigns model of growth. introduces the concepts of good news (İt-i> 0) and bad news (İt-i< 0), which have different effects on the noticeable that GARCH model is a specific case 2) The second period lies between July 2004 conditional variance. TheThe good ARCH news family effect models is measured include byother aicoefficient models while the bad news of TARCH with an It equal to 0. and December 2008 and is noticeable for a effect is measured by ai + Ȗi because It = 1. In this case, if Ȗi> 0, the- badstable news exchange causes rate an around 124. Continuing increase in variance and has a leverage effect of order i, but if Ȗi - the WKH revival QHZV of LPSDFW the first LV period, the economy was with specific restrictions like the IGARCH mod DV\PPHWULF7KHµQHZV¶FRQFHSWUHIHUVWRWKHLQIRUPDWLRQRISDVWSHULRGVZKLFKDIIHFWVWKHperforming better with a growth of around el which is estimated by removing the free con current period. Themodel can be written as TGARCH(r,p,q), where q, p was6%.Theand duer are to need the lack orders of of borrowersprevious opportunities for funds and to in of- stant and where the sum of coefficients a andβ of ARCH, GARCH and can TGARCH obtain other terms. ARCH It is models easily noticeable like: EGARCH, that GARCHbanks for model investing is a in profitable instruments is 1. By making other forms of restrictions we specific case of TARCH with an It equal to 0. vest and generate substantial profits. The ARCH family models3.2PARCH, include The CGARCH, data other modelsGJARCH with etc. specific restrictions like the IGARCH - model which is estimated by removing the free constant and where the sum3) The of coefficientslast period startsa in January 2009 and andȕ is 1. By making other forms of restrictions we can obtain other ARCHofcontinues 138,which models to nowadays. we like: believe The to graphbe also shows due toa rap the EGARCH, PARCH, CGARCH, GJARCH etc. id increase of the exchange rate to an average theThe form data of employed dynamic series consists over of a average time horizon daily crisis within the European Union. 3.2 The data betweendata of the January EUR/ALL 01, exchange 2002 and rate July presented 10, 2014. in global financial crisis and the sovereign debt The data employed consists of average daily data of the EUR/ALL exchange rate presented in the form of dynamic series over a time horizon between January 01, 2002 and July 10, The data is taken from the official webpage of 2014.The data is taken fromrates the with official other webpagecurrencies of are the archived Bank of . TheAlbania where daily ALL thetotal Bank number of Albania of observations where daily is 3,131. ALL exchange exchange rates with other currencies are archived1. The total number of observations is 3,131. The daily exchange rate return is defined as:

The daily exchange rate returnEUR is/ definedALLt as: ' ln(EUR / ALLt ) ln EUR / ALLt1 (5) wKHUHǻOQ (85$// LVWKHUHWXUQRQH[FKDQJHUDWHZKLOH(85$//t is the exchange rate at Figure 4.1. EUR/ALL (LHS axis) and Δln(EUR/ALL) time t, and EUR/ALLt-1 is the exchange rate at time t± 1. (RHS axis) for January 2002 to July 10, 2014 whereΔln(EUR/ALL)is the return on exchange Our empiric analysis is computed by using EViews7, a professional computer program for rate, while EUR/ALLt is the exchange rate at econometric analysis. time t– 1. time t, and EUR/ALLt-1 is the exchange rate at The second graph in Figure 4.1 shows the EUR/ 4 FINDINGS ALL exchange rate return which is centered on ratethe zero sometimes value along increased the entire and in time other horizon times 4.1 Descriptive statisticseconometricOur empiric analysis. analysis is computed by using under study, despite the fact that the exchange EViews7, a professional computer program for Figure 2 shows time series plots of EUR/ALL exchange rate in the upper graph and the  The Bank of Albania publishes the exchange rates decreased. It can be noticed that the degree of exchange rate return in thefour below to five graph. times aFrom week. the graphs in Figure 2, threereturn periods is continuouslystand out: changing, taking both high and low values. This effect is measured by 1) The first period between January 2002 and June2004, is characterized by an average 24 Economic Review - Journal of Economics and Business, Vol. XIII, Issue 1, May 2015 exchange rate of 134. During this period the financial system was not fully consolidated and the Albanian economy was already showing signs of growth.

1 The Bank of Albania publishes the exchange rates four to five times a week. above, we observeTable 4.1andTable 4.2, which present several statistical data. The last two columns of Table 1 show results on the exchange ratereturn. However, the second column in the Table takes intoDo theconsideration news affect the the eur/all whole exchange period, rate whilevolatility? the third column shows results for the ARCH familythe period models, untilknown January also as vola 19,- 2010.then the kurtosis should be three. The kurtosis - - Table 4.1: Descriptive VWDWLVWLFVRI(85$//DQGǻOQ (85$// tility clustering. The two graphs are Seriesshown to itof is the about Δln(EUR/ALL)EUR/ALL 10. As a result, for the distributionsampleǻOQ (85$// “until of Jan the gether to present an important statementSample that uary 19, 2010”All is 7.2, whileAll for the fullUntil sample Jan. 19, 2010 Mean 132.114 4.15E-5 5.69E-5 when the exchange rate falls, the volatilityStd. Dev. of 7.3591 0.0028 0.0033 exchange rate return is higher. TakingSkewness a look at- exchange rate-0.1780 EUR/ALL return0.2756 has fat tails. 0.2557 the graph, the period between 2003Kurtosis and 2005 1.3180 10.0362 7.2171 The unpredictability of the EUR/ALL exchange confirms this statement, which we will ana the idea of asymmetric volatility described by rate returns can be shown by observing the sign lyze in the next section. Also, we can observe The mean is almost zero in bothof the columns, exchange return but slightly in Figure h4.1.igher The graphin the first one with data from the full period included in the study.confirms Standard that the returndeviation sometimes and smoveskewness up, are different in both cases. Nelson with his EGARCH model (Engel, 2004) The most interesting feature causingis the kurtosis for a positive which sign, measur and otheres the times magnitude of the extremes. If stating that volatility tends to increase when returns are normally distributed,down, changing then the to a kurtosis negative sign. should However, be three. The kurtosis of the the markets are in decline.These graphs give a ǻOQ (85$// IRUWKHVDPSOH³XQWLO-DQXDU\´LVthe moment when this change happens cannotZKLOHIRUWKHIXOOVDPSOHLWLV first hint of a possible presence and effect of the about 10. As a result, the distributionbe defined, of proving the exchange the unpredictability rate EUR/ALL of the return has fat tails. bad news in the exchange rate in the EUR/ALL - EUR/ALL exchange rate return. exchange rateThe volatility, unpredictability which we will discussof the EUR/ALL exchange rate returns can be shown by observing the Returns are almost unpredictable, they have in the followingsign section. of the exchange returnThevolatility in Figure clustering 4.1. The can graph be concisely confirms ob that the return sometimes moves up, causing for a positiveserved bysign, looking and atother autocorrelation times down, results. changing to a negative sign. However, the moment when this change happens cannot be defined, proving the Volatility clustering will show up as significant- surprisingly large numbers of extreme values, clustered inunpredictability time. These features, of forthe whichEUR/ALL autocorrelations exchange in squaredrate return. or absolute returns and both the extremes and quiet periods are Thevolatility clustering can- (Engel, be concisely 2004). Table observed 4.2 presents by the lookingautocor at autocorrelation results. scribed as unpredictability,Volatility clustering fat tails, and volatil will - showrelation up coefficients as significant for the series autocorrelations of squared in squared or absolute the ARCH models were designed, are often de returns (Engel, 2004). Table- returns 4.2 presents which are thesignificant autocorrelation above the 99% coefficients for the series of squared returns which are significantlevel, indicating above a clear the evidence 99% of level, volatility indicating a clear evidence of ity clustering. Engel’s (2004, 407) important re volatility clustering in this series.clustering in this series. mark on the topic remains the following: ‘When returns the volatility is high, it is likely to remain high,Table 4.2: Autocorrelations of EUR/ALL squared returns To check whether the return of the EUR/ALL Table 4.2. Autocorrelations of EUR/ALL squared and when it is low, it is likely to remain low.’ Lag AC Q-Stat. Prob. 1 0.265 220.59 0.000 above, we observeTable 4.1andTable 4.2, which 2 0.224 377.71 0.000 presentexchange several rate has statistical the three data. features The last mentioned two col- 3 0.141 440.26 0.000 4 0.154 514.33 0.000 5 0.087 538.29 0.000 ratereturn. However, the second column in the 6 0.112 577.52 0.000 Tableumns of takes Table into 1 show consideration results on the the whole exchange pe- 7 0.083 599.36 0.000 above, we observeTableriod, while 4.1andTable the third4.2, which column present shows several resultsstatistical fordata. The last two 8 0.145 665.10 0.000 9 0.102 697.65 0.000 columns of Tablethe 1 show period results until on the January exchange 19, ratereturn. 2010. However, the second column in10 0.155 773.59 0.000 the Table takes into consideration the whole period, while the third column shows results for11 0.092 800.07 0.000 the period until January 19, 2010. 12 0.092 826.71 0.000 Table 4.1: Descriptive VWDWLVWLFVRI(85$//DQGǻOQ (85$// Table 4.1. Descriptive statistics of EUR/ALL and The above analysis showed that the EUR/ALL Δln(EUR/ALL)Series TheEUR/ALL above analysisǻOQ (85$// showed that the EUR/ALL exchange rate has three features of the series Sample presentedAll Allby theUntil theory: Jan. 19, 2010 unpredictability, fat tails and volatility clustering. Mean 132.114 4.15E-5 5.69E-5 presented by the theory: unpredictability, fat Std. Dev. 7.3591 0.0028 0.0033 exchange rate has three features of the series Skewness -0.1780 0.2756 0.2557 Kurtosis 1.3180 10.0362 7.2171 4.2tails Evidence and volatility of news clustering. impact The mean is almost zero in both columns, but The mean is almost zero in both columns, but slightly higher in the first one with data from the - full period includedthe in fullthe study.period Standard included deviation in andthe s kewnessstudy. Standardare different in both cases. The most interestingslightly feature higher is the kurtosis in the which first measur onees with the magnitude data from of the extremes.The econometric If modeling procedure of differ returns are normally distributed, then the kurtosis should be three. The - kurtosisent of time the series requires the need to perform ǻOQ (85$// IRUWKHVDPSOH³XQWLO-DQXDU\´LVdeviation and skewness are differentZKLOHIRUWKHIXOOVDPSOHLWLV in both addressa stationarity the unit test root before of the moving series aheadand results with about 10. As a result, the distribution of the exchange rate EUR/ALL return has fat tails. cases. The most interesting feature is the kur arethe presentedanalysis. We in use Table the 4.3 Dickey-Fuller for both EUR/ALL test to The unpredictabilitytosis of the which EUR/ALL measures exchange therate returns magnitude can be shown of the by observing the sign of the exchangeextremes. return in If Figure returns 4.1. are The graphnormally confirms distributed, that the return sometimes moves up, causingEconomic for a positive Review sign, - andJournal other oftimes Economics down, changing and Business, to a negative Vol. sign.XIII, Issue 1, May 2015 25 However, the moment when this change happens cannot be defined, proving the unpredictability of the EUR/ALL exchange rate return. Thevolatility clustering can be concisely observed by looking at autocorrelation results. Volatility clustering will show up as significant autocorrelations in squared or absolute returns (Engel, 2004). Table 4.2 presents the autocorrelation coefficients for the series of squared returns which are significant above the 99% level, indicating a clear evidence of volatility clustering in this series. Table 4.2: Autocorrelations of EUR/ALL squared returns Lag AC Q-Stat. Prob. 1 0.265 220.59 0.000 2 0.224 377.71 0.000 3 0.141 440.26 0.000 4 0.154 514.33 0.000 5 0.087 538.29 0.000 6 0.112 577.52 0.000 7 0.083 599.36 0.000 8 0.145 665.10 0.000 9 0.102 697.65 0.000 10 0.155 773.59 0.000 11 0.092 800.07 0.000 12 0.092 826.71 0.000

The above analysis showed that the EUR/ALL exchange rate has three features of the series presented by the theory: unpredictability, fat tails and volatility clustering. G. Çera, E. Dokle, E. Çera Apart from the existence of the two types of news, the general - literature on threshold 4.2 Evidence ofand news Δln(EUR/ALL) impact series. They confirm that- Apart from the existence of the two types the EUR/ALL exchange GARCH rate is non-stationary discusses the differentof news, weights the general in the literature ARCH family on thresh models (Engel, 2004). The The econometric modeling procedure of differentTGARCH(1,1,2) time series requires can the needbe written to perform as: a stationarity test beforeat level moving while ahead its withfirst thedifference analysis. We becomes use the Dickey-Fullersta old GARCH test to discusses the different weights in address the unit roottionary of the andseries highlyand results significant. are presented However, in Table 4.3 the for both the EUR/ALL ARCH2 family2 models2 (Engel,2 2004).2 The ǔt Z  a1džt1  a2džt2  ǃǔt1  DŽdžt1It1 and ǻOQ (85$// theEUR/ALL series. data. They exchange confirm thatrate the return EUR/ALL is stationary exchange rate at is non-stationaryTGARCH(1,1,2) can be written as: at level while itslevel first differenceand we do becomes not need stationary to first and highlydifferentiate significant. However, the (6) EUR/ALL exchangeTable rate return 4.3. Unitis stationary root at test Balalevel forand and we EUR/ALL Asemota do not need and(2003) to first differentia explainte that these weights are calculated on the long run average, the data. previous forecast, symmetric news and negative news. The weights presented in this model Bala and Asemota (2003) explain that these- Table 4.3: Unit root test for E are: (1 ± a1 ± a2± ȕ ±-FullerȖ2, test)ȕ, a1 + a2) and Ȗ/2. When estimated in our model, the weights of 85$//DQGǻOQ (85$//  $XJPHQWHG'LFNH\ weights are calculated on the long run average, Δln(EUR/ALL) (Augmentedexchange Dickey-Fuller rate returns test) are: ±0.01835, 0.869, 0.1363 and 0.01305. In the TGARCH model, Series in Lag length t-Statistic Prob. previous forecast, symmetric news and nega TheEUR/ALL next step ofLevel our analysisµJRRG3 QHZV¶ is-1.43727 to DQG evaluate 0.5654 µEDG QHZV¶ KDYH GLIIHUHQW HIIHFWV RQ FRQGLWLRQDO YDULDQFH 7KH LPSDFW RI 1st difference 2 -35.3055 0.0000 tive news. The weights presented in this model µJRRGQHZV¶LVPHDVXUHGE\ a1 + a2)ZKLOHWKHLPSDFWRIµEDGQHZV¶E\ Ȗ +a1 + a2). For the ǻOQ (85$// Level EUR/ALL2 -35.2924 exchange 0.0000 rateare: return, (1 – gooda1 – a2–news β –has γ/2, an β, impact a1 + a2) of and 0.1363 γ/2. while bad news has an Note: Null Hypothesis: Series has a unit root; When estimated in our model, the weights of Exogenous: Constant; Automatic lagimpact length - basedof 0.1624. on SIC. These values prove that good news and bad news do not have the same exchange rate returns are: –0.01835, 0.869, impact on the exchange rate returns, confirming the asymmetry of the- news impact on our 0.1363 and 0.01305. In the TGARCH model, The next step of our analysis is to evaluate the news impact on the EUR/ALL exchangefects on rate conditional variance. The impact of series. ‘good news’ and ‘bad news’ have different ef volatility through estimating two models: TGARCH(1,1,1) and TGARCH(1,1,2).- The results in Table 4.4 showthe that news apart impact from Ȗ, on all other theTable EUR/ALL coefficients 4.5 shows areexchange positive correlogram and significant of standardized residuals and correlogram of squared inTGARCH(1,1,1). In this case, the ARCH test suggests the usage of other lag orders‘good in news’order is measured by (a1 + a2), while rate volatility through standardized estimating two residuals mod for TGARCH(1,1,2) model. The results of the squared standardized to exclude all the possible ARCH effects. the impact of ‘bad news’ by (γ +a1 + a2). For els: TGARCH(1,1,1) andresiduals TGARCH(1,1,2). correlogram The (performed until the 12 lag) show once again the absence of ARCH For this reason, we went further to estimate the TGARCH(1,1,2) model with twothe lags EUR/ALL in the exchange rate return, good news results in Table 4.4 showeffect that inapart the frommodel. γ, allMoreover, since the coefficient of standardized residuals autocorrelation error term which implies two a coefficients. Adding an additional lag to thehas error an impact term of 0.1363 while bad news has an other coefficients are positiveare significant and significant while thenews coefficients and bad news for the do not autocorrelation have the same of im the- standardized squared improves4.2 Evidencethe quality of newsof the impact model in general, making all the coefficients significant.impact The of 0.1624.Ȗ These values prove that good coefficientThe econometric is significantinTGARCH(1,1,1). modeling at 99% procedure level and of In differentstrengthens thisresiduals case,time theseries the ideaare requires ARCH thatnot news significant,the test need is important to perform it provesin a the that TGARCH(1,1,2) was properly chosen. exchange ratereturn, implying that news is asymmetric. The positive sign showsthe asymmetry that bad of the news impact on our se- stationarity testsuggests before moving the usage ahead with of other the analysis. lag orders We use inApart the order Dickey-Fuller from the existence test to of the two types of news, the general literature on threshold newsaddress increases the unittheFor rootvariance this of the reason, and series has and wea leverage results went are furthereffect presentedTable in tothe 4.5: in estimate proper Table Correlogram 4.3order. forthe bothWe of pact canEUR/ALL standardized conclude on the exchangeresiduals and rate correlogram returns, ofconfirming standardized residuals squared for to exclude all the possible ARCH effects.GARCH discussesries. the different weights in the ARCH family models (Engel, 2004). The that andthe ǻOQ (85$// second model isseries. better They than confirm the first that one the for EUR/ALL two reasons. exchange First, rate the isARCH non-stationary test shows TGARCH(1,1,2) model at level while its first difference becomes stationary and highly significant.TGARCH(1,1,2)- However, can be the written as: that EUR/ALLthis model exchange does not ratehave return ARCH is stationary effects in at it, level and andsecond, we do it notreflects need ato lower first differentia AIC statisticste - Correlogram2 Z of 2  2  2Correlogram 2 of standardized and thehigher data. log likelihood. ǔt a1džt1 a2džt2 ǃǔt1 DŽdžt1It1 - TGARCH(1,1,2) model with two lags in the er standardized residuals residuals squared Table 4.3: Unit root test for E85$//DQGǻOQ (85$//  $XJPHQWHG'LFNH\Bala and-Fuller Asemota test)Table (2003) 4.5 explain shows that correlogramthese weights are ofcalculated standardized on the long run average, rorTabl eterm4.4: Results which for TGARCH(1,1,1)implies two and a coefficients.TGARCH(1,1,2)Lag models Add Series in Lag length t-Statistic previousProb. forecast,ACresiduals symmetricPAC and newsQ-Stat correlogram and negativeProb. news. ofAC squared The weightsPAC standard presentedQ-Stat in thisProb. model ing EUR/ALLan additional Levellag toTGARCH(1,1,1) the error3 term-1.43727TGARCH(1,1,2) improves are:0.5654 1(1 ± a0.1651 ± a2 ± ȕ 0.165± Ȗ2, ȕ , a85.0711 + a2) and 0.000 Ȗ/2. When -0.004 estimated -0.004 in our model,0.052 the weights0.820 of 1st difference Coef. 2 Prob.-35.3055 Coef. exchange0.0000 Prob. rate ized returns residuals are: ±0.01835, for 0.869, TGARCH(1,1,2) 0.1363 and 0.01305. model. In the The TGARCH model, theǻOQ (85$// quality of theLevel model in 2general, -35.2924 making 0.00002 all0.037 0.010 89.445 0.000 0.002 0.002 0.066 0.968 Ȧ 5.9E-8 0.0000 2.9E-8µJRRG 0.0000 QHZV¶ DQGresults µEDG QHZV¶ of the KDYH squared GLIIHUHQW HIIHFWV standardized RQ FRQGLWLRQDO residuals YDULDQFH 7KH LPSDFW RI theNote coefficients: Null Hypothesis: significant. Series has a unit Theroot; γ coefficient is Exogenous:Į1 Constant; Automatic0.2982 lag length 0.0000 - based on SIC.0.4205µJRRGQHZV¶LVPHDVXUHGE\ 3 0.0000-0.025 -0.033 a91.3611 + a2)ZKLOHWKHLPSDFWRIµEDGQHZV¶E\ 0.000 0.008 0.009 Ȗ0.291 +a1 + a20.962). For the EUR/ALL exchangecorrelogram rate return, (performed good news has anuntil impact the of 12 0.1363 lag) while show bad- news has an significantĮ2 at 99% level± and ± strengthens-0.2842 4 0.0000 the0.010 0.020 91.705 0.000 -0.014 -0.014 0.893 0.926 The next step of our analysisȕ is to evaluate the0.7478 news 0.0000impact on the0.8690 impactEUR/ALL 0.0000 of 0.1624. exchangeonce These againrate values the prove absence that good of news ARCH and bad effect news do innot -havethe the same idea that news is important in the exchangeimpact5 on0.017- the exchange 0.014 rate returns,92.629 confirming 0.000 the-0.022 asymmetry -0.022 of the news2.372 impact 0.796 on our volatility through estimatingȖ two models: TGARCH(1,1,1)0.0239 0.1926 and TGARCH(1,1,2).0.0261 0.0082 model.The results Moreover, since the coefficient of stan ratereturn, implying that news is asymmetric.series. - in Table 4.4Akaike show info that criterion apart (AIC) from Ȗ, all-9.5902 other coefficients-9.60604 are positive 6 -0.014 and significant -0.022 93.288 0.000 0.002 0.002 2.385 0.881 inTGARCH(1,1,1). In this case, the ARCH test suggests the usage of other lag ordersdardized in order residuals autocorrelation are signifi LogproperThe likelihood positive order. sign We showscan15,012.67 conclude that bad that news15,038.46 the increas second 7 -0.029 -0.024 95.975 0.000 -0.026 -0.026 4.552 0.714 to exclude all the possible ARCH effects. Table 4.5 showscant correlogramwhile the ofcoefficients standardized for residuals the autocorrela and correlogram of squared Durbin-Watsones the variance stat and has1.8763 a leverage effect1.8763standardized in8 the-0.025- residuals -0.014 for TGARCH(1,1,2)97.878 0.000 model. -0.013 The results -0.013 of the squared5.067 standardized 0.750 For this reason,ARCH we effect:went furtherChi-sq.(lag=1) to estimate 4.9300the TGARCH(1,1,2) 0.0264 0.0515 modelresiduals 0.8204with correlogram two tion lags inof (performedthe the standardized until the 12 lag) squared show once residuals again the absenceare of ARCH error term which implies two a coefficients. Adding an additional lag9 to0.037 the was error 0.045properly term 102.20 chosen. 0.000 -0.007 -0.008 5.221 0.815 Convergencesons. First, achieved the after ARCH test12 iterations shows that this18 effectiterations model in the model. Moreover, since the coefficient of standardized residuals autocorrelation improves the quality of the model in general, making all the coefficients10 significant.0.023not The significant,0.009 Ȗ 103.80 it proves0.000 that-0.004 TGARCH(1,1,2) -0.004 5.267 0.873 coefficient isNote significantmodel'HSHQGHQWYDULDEOHǻOQ (85$// 6DPSOH DGMXVWHG ± is at better99% level than and strengthensthe first theone idea for that twoare news significant rea3,131; is important while in thethe coefficients for the autocorrelation of the standardized squared exchange ratereturn,Method: ML implying - ARCH that (Marquardt) news is - asymmetric.Normal distribution. The positive residuals11 sign-0.002 are shows not significant, that-0.010 bad it 103.81proves that 0.000TGARCH(1,1,2) -0.013 was-0.013 properly chosen.5.793 0.887 news increases the variance and has a leverage effect in the proper order. We can conclude does not have ARCH effects in it, and second,12Table -0.0404.5: CorrelogramTable -0.036 4.5. of standardized Correlogram108.73 residuals 0.000 and of correlogram standardized-0.014 of standardized-0.016 residuals residuals6.422 squared 0.893 for that the secondlikelihood. model is better than the first one for two reasons. First, the ARCH test shows TGARCH(1,1,2) model that this modelit does reflects not have a ARCH lower effects AIC instatistics it, and second, and it reflectshigher a lowerlog AICand statistics correlogram of standardized residuals squared and higher log likelihood. for TGARCH(1,1,2)Correlogram model of Correlogram of standardized standardized residuals residuals squared Table 4.4: Results for TGARCH(1,1,1)Figure and TGARCH(1,1,2) 4.2illustrates models theLag distribution AC PAC ofQ-Stat the Prob. error AC termPAC inQ-Stat the Prob. TGARCH(1,1,2) model. Table 4.4. Results for TGARCH(1,1,1)TGARCH(1,1,1)Although aandTGARCH(1,1,2) first TGARCH look at the1 0.165 graph 0.165 may 85.071 suggest 0.000 -0.004 that -0.004 the distribution0.052 0.820 of the error term is (1,1,2) models Coef. Prob. Coef. Prob. 2 0.037 0.010 89.445 0.000 0.002 0.002 0.066 0.968 Ȧ 5.9E-8normal, 0.0000 based2.9E-8 on 0.0000 the previously3 -0.025 discussed-0.033 91.361 features0.000 0.008 of 0.009 distribution 0.291 0.962 (skewness, kurtosis and Į1 0.2982 0.0000 0.4205 0.0000 4 0.010 0.020 91.705 0.000 -0.014 -0.014 0.893 0.926 Į2 ± ± -0.2842 0.0000 probability of Jarque-Bera5 statistic)0.017 0.014 we 92.629 conclude 0.000 -0.022 that -0.022 the error2.372 0.796 term in this model is not ȕ 0.7478 0.0000 0.8690 0.0000 Ȗ 0.0239normally 0.1926 distributed.0.0261 0.0082 6 -0.014 -0.022 93.288 0.000 0.002 0.002 2.385 0.881 Akaike info criterion (AIC) -9.5902 -9.60604 7 -0.029 -0.024 95.975 0.000 -0.026 -0.026 4.552 0.714 Log likelihood 15,012.67 15,038.46 8 -0.025 -0.014 97.878 0.000 -0.013 -0.013 5.067 0.750 Durbin-Watson stat 1.8763 1.8763 9 0.037 0.045 102.20 0.000 -0.007 -0.008 5.221 0.815 ARCH effect: Chi-sq.(lag=1) 4.9300 0.0264 0.0515 0.8204 10 0.023 0.009 103.80 0.000 -0.004 -0.004 5.267 0.873 Convergence achieved after 12 iterations 18 iterations 11 -0.002 -0.010 103.81 0.000 -0.013 -0.013 5.793 0.887 Note'HSHQGHQWYDULDEOHǻOQ (85$// 6DPSOH DGMXVWHG ± 3,131; Method: ML - ARCH (Marquardt) - Normal distribution. 12 -0.040 -0.036 108.73 0.000 -0.014 -0.016 6.422 0.893

Figure 4.2illustrates the distribution of the error term in the TGARCH(1,1,2) model. Economic Review - JournalAlthough of a firstEconomics look at the and graph Business, may suggest Vol. XIII, that theIssue distribution 1, May of2015 the error term is normal, based on the previously discussed features of distribution (skewness, kurtosis and 26 probability of Jarque-Bera statistic) we conclude that the error term in this model is not normally distributed. Do the news- affect the eur/all exchange rate volatility? -

Figure 4.2 illustrates the distribution of the er the EUR/ALL exchange rate return series is un- distributionror term in the of TGARCH(1,1,2)the error term ismodel. normal, Although based predictable, has fat tails and confirms volatility ona first the lookpreviously at the discussedgraph may features suggest of that distri the- rateclustering. volatility Our andempiric provides analysis actual proves values the ex for istence of news impact on EUR/ALL exchange- - ly, the bad news has a considerable impact on rorbution term (skewness, in this model kurtosis is not and normally probability distrib of- the good news and bad news. More important- uted.Jarque-Bera statistic) we conclude that the er the EUR/ALL exchange rate volatility, confirm A possible topic for further research on the is- 900 ing previous researches. Series: Standardized Residuals 800 Samplesue 2 3131 remains the study of news impact of the Observations 3130 700 - Mean 0.016720 600 - Median 0.034068 Maximum 5.681568 500 ALL exchange rate with other foreign curren Minimum -7.654804 400 Std.cies, Dev. allowing 1.001961 for comparison of empirical re Skewness -0.215203 300 Kurtosisforsults further between 7.567525 research different would time be frequencies: the study of daily,the

200 Jarque-Beraweekly, 2744.949 and monthly. Another interesting idea- Probability 0.000000 100 0 ratereturn risk of ofa portfolioALL with composed other currencies. by different cur -8 -6 -4 -2 0 2 4 6 900 rencies, while minimizing the level of exchange Figure 4.2: HistogramSeries: ofStandardized the error termResiduals distribution of the TGARCH(1,1,2) model 800 Sample 2 3131 6 BIBLIOGRAPHY Observations 3130 700 5 CONCLUSION ANDMean FURTHER 0.016720 RESEARCH 1. 600 Median 0.034068 The importance of studyingMaximum exchange 5.681568 rate has increased significantly with the continuous 500 Bala, D. A., & Asemota, J. O. (2013) Exchange process of globalization Minimum affecting -7.654804the entire world. Despite the considerable amount of 400 Std. Dev. 1.001961 rate volatility in Nigeria: Application of studies on the issue of analyzingSkewness the -0.215203 source of exchange rate volatility and modeling it, 300 several challenges arise Kurtosis constantly. 7.567525 The lack of other studies onGARCH the predictability models with of exogenous break. CBN- EUR/ALL exchange rate and the news impact on exchange rate volatility,Journal with of Applied a specific Statistics, 4(1), pp. 89- 200 Jarque-Bera 2744.949 focus on Albania, was theProbability main drive 0.000000 of this research. Following a previous116.Bank study of conducted Albania (2014)Monetary poli 100 by Çera et al. (2013) to assess the value at risk of the EUR/ALL exchangecy: Offical rate, Exchange this paper Rates[Online]. Avaibble 0 focuses mainly on the news impact. from:http://www.bankofalbania.org/web/ -8 -6 -4 -2 0 2 4 6 The results of descriptive statistics show that the EUR/ALL exchangeExchange_Rates_2014_2.php rate return series is [Accessed: 10 Figure 4.2: Histogram of the error term distribution of the TGARCH(1,1,2) model unpredictable,Figure has 4.2. fat Histogram tails and confirmsof the error volatility term distributionclustering. Our empiricJuly analysis 2014]Bank proves of the Albania (2014)Statistics: existence of news impact on EUR/ALL exchange rate volatility and Timeprovides series actual [Online].Avaibble values from: http:// 5of CONCLUSIONthe TGARCH (1,1,2) AND model FURTHER RESEARCH for the good news and bad news. More importantly, the bad news haswww.bankofalbania.org/web/Statistika_ a considerable impact on the EUR/ALL exchange rate volatility, confirming previous researches.2. - 5 CONCLUSION AND FURTHER RESEARCH 230_1.php [Accessed:22 July 2014] A possible topic for further research on the issue remains the study of news impact of the Black, F. (1976) Studies in Stock Price Vol The importance of studying exchange rateALL has exchange increased rate significantly with other foreign with currencies, the continuous allowing for comparison of empirical - The importance of studying exchange rate has atility Changes. Proccedings of the 1976 process of globalization affecting the entireresults world. between Despite different time the frequencies: considerable daily, amount weekly, and of monthly.nomics Another Statistics interesting Section, American Statis- idea for increasedfurther rese arch significantly would be the with study of the the continuousreturn of a portfolio Businesscomposed Meetingby different of the Business and Eco studies on the issue of analyzing the sourcecurrencies, ofprocess exchange while minimizing of globalizationrate volatility the level of affecting andexchange modeling rate the risk entire it,of ALL with other currencies. 3. - several challenges arise constantly. The lack world. of other Despite studies the on considerable the predictability amount of of- tical Association. pp. 177-181. EUR/ALL exchange rate and the news impact6 BIBLIOGRAPHY onstudies exchange on the rate issue volatility, of analyzing with the a source specific of Bollerslev, T.(1986) Generalized Autore focus on Albania, was the main drive of thisBala, research. D. exchangeA., & FollowingAsemota, rate J.volatility O. a (2013)previous and Exchange modeling study rate conducted volatility it, sever in Nigeria: Application of GARCH - gressive Conditional Heteroskedasticity. by Çera et al. (2013) to assess the value at risk modelsalof challengesthe with EUR/ALL exogenous arise break. constantly.exchange CBN Journal Therate, lackof this Applied of paper other Statistics 4., 4(1), pp. 89-116. Journal of Econometrics 31, pp. 307-327. focuses mainly on the news impact. Bank ofstudies Albania on (2014)the predictabilityMonetary policy: of EUR/ALL Offical Exchange ex Ratesmodel[Online]. approach Avaibble to calculate the value at from:changehttp://www.bankofalbania.org/web/Exchange_Rates_2014_2.php rate and the news impact on exchange Çera, [Accessed: G., Çera, 10 E., July& Lito, G. (2013) A GARCH- The results of descriptive statistics show that the2014]rate EUR/ALL volatility, exchangewith a specific rate focus return on series Albania, is unpredictable, has fat tails and confirms volatilityBank of Albaniawas clustering. the (201 main4)Statistics: Our drive empiric of Time this series research.analysis [Online].Avaibble Followingproves the from a : risk of Albanian lek exchange rate. Europe - 5. existence of news impact on EUR/ALL exchangehttp://www.bankofalbania.org/web/Statistika_230_1.phpprevious rate volatility study conducted and provides by Çera actual et al. values(2013) [Accessed:22 an July Scientific 2014] Journal, 9(25), pp. 250-260. - for the good news and bad news. More importantly,newsto assess the impact. bad the news value has at riska considerable of the EUR/ALL impact ex Chong, C., Chun, L., & Ahmad, M. (2002) on the EUR/ALL exchange rate volatility, confirmingchange previous rate, this researches. paper focuses mainly on the The results of descriptive statistics show that Modeling the volatility of currency ex A possible topic for further research on the issue remains the study of news impact of the change rate using GARCH model. Pertanika J.Soc.Sci.Hum, 10 (2), 85-95. ALL exchange rate with other foreign currencies,Economic allowing Review for - Journal comparison of Economics of empirical and Business, Vol. XIII, Issue 1, May 2015 results between different time frequencies: daily, weekly, and monthly. Another interesting idea for further research would be the study of the return of a portfolio composed by different 27 currencies, while minimizing the level of exchange rate risk of ALL with other currencies.

6 BIBLIOGRAPHY Bala, D. A., & Asemota, J. O. (2013) Exchange rate volatility in Nigeria: Application of GARCH models with exogenous break. CBN Journal of Applied Statistics, 4(1), pp. 89-116. Bank of Albania (2014): Offical Exchange Rates[Online]. Avaibble from:http://www.bankofalbania.org/web/Exchange_Rates_2014_2.php [Accessed: 10 July 2014] Bank of Albania (2014)Statistics: Time series [Online].Avaibble from: http://www.bankofalbania.org/web/Statistika_230_1.php [Accessed:22 July 2014] G. Çera, E. Dokle, E. Çera -

6. Engel, R.(2004) Risk and Volatility: Econo vectors of autoregression. Working paper,- 405-420.metric Models and Financial Practice. The elBank comparison: of Albania. nonparametric Jithitikulchai, T. approach. (2005)A American Economic Review, 94(3), pp. Facultystudy on of Thai Economics, exchange Thammasat rate volatility Univer mod- sity, 1. 7. Engel, R., & Ng, V.(1993) Measuring dhe 10. - testing the impact of news on Volatility. The - Jurnal of Finance, 48(5), pp. 1749-1778.- namics:Longmore, An R.,Application & Robinson, of Asymmetric W. (2004)Mod Vol- Glosten, L. R., Jaganathan, R., & Runkle, D.- atilityelling Models. and Forecasting Exchange . Rate Dy (1993) On the Relation between the Ex- 11. pected Value dhe the Volatility of the Nor mal Excess Return on Stocks. Jurnal of Fi 8. - Luci, E., & Vika, I. (2011) The equilibrium- nance, 48, pp. 1779-1801. per,real Bankexchange of Greece. rate of lek vis-a-vis euro: is it Hoda, B. (2012), The equilibrium real ex much misaligned? Speciali Conference Pa 12. change rate in a Developing Economy.- Working paper, Bank of Albania. Insah, B. Mundaca, G. B.(1991) The Volatility of the (2013) Modelling Real Exchange Rate Vola Norwegian Currency Basket. Scandinavian- tility in a Developing Country. Journal of Journal of Economics, vol.93 (1), pp. 53-73.- Economics and Sustainable Development, 9. icsZakoïan, and Control, J. M. (1994) 18, pp. Threshold 931-944. Heteroske Vol.4, No.6. dastic Models. Journal of Economic Dynam Strefi, K., & Semi, V. (2009)Exchange rate pass through in Albania, evidencefrom

28 Economic Review - Journal of Economics and Business, Vol. XIII, Issue 1, May 2015