Sress Tesing of Commercial Banks Exposure o Credi Risk: A Sudy Based on Wrie-off Nonperforming Loans Wei Lu 1 & Zhiwei Yang 1 1 School of Managemen, Universiy of Science and Technology of China, Hefei, China Correspondence: Zhiwei Yang, School of Managemen, Universiy of Science and Technology of China, Hefei, China. E-mail: yzhw@mail.usc.edu.cn Received: April 12, 2012 Acceped: May 16, 2012 Published: Augus 1, 2012 doi:10.5539/ass.v8n10p16 URL: hp://dx.doi.org/10.5539/ass.v8n10p16 Absrac This sudy inroduced a sress-esing model wih a dummy variable ha refers o wrie-off non-performing loans (NPL) by Agriculural Bank of China. A new variable Y ha indicaed he rae of NPL in major naional commercial banks in erms of logi ransformaion was applied o es sress olerance. This aricle buil a regression model on he basis of four explanaion variables: he growh rae of GDP, indicaor of cusomer price, he growh rae of supplying nominal currency and indicaor of house price. Then we ook advanages of VAR model o esablish he relaionship beween variables. Based on he model, diverse scenario was se up o conduc sress es o NPL of commercial banks. The es covered four quarers and discovered ha lower growh rae of GDP, slump in CPI, slowdown in supply of nominal currency and surging price of house are in charge of shor-erm increase in non-performing loans. From long-erm perspecive, he commercial banks would iniiae inernal sysem o miigae he shock from volaile macro facors. Keywords: credi risk, wrie-off nonperforming loans (NPL), VAR model, sress-es 1. Inroducion Furher opening of domesic financial marke enables closer connecion o inernaional markes and challenges credi risk managemen of naional commercial banks. Sress es is well known o evaluae he sabiliy of commercial banks credi sysem by simulaing poenial risk exposure o exreme shock like financial crisis. The es works especially owards unexpeced macro flucuaion on banking sysem and herefore plays a crucial role in forecasing and reducing sysemaic financial risk, which conribues o he sabilizaion financial sysem. This paper applied macro sress es o he analysis of changes in NPL (nonperforming loan) of a commercial bank under given macro-economic shock. This aricle consiss of hree pars: Secion one where we reviewed he naional and inernaional lieraures concerning macro sress es. Secion wo where we simulaed a macro sress esing model in erms of NPL in Chinese commercial banks and quarerly macroeconomic daa. Finally, we hypohesized cerain scenario on he basis of model and hen forecased he new financial saus of NPL of commercial bank in one year afer exreme shocks. 2. Lieraure Review IMF defined macro sress esing as a range of echniques used o assess he vulnerabiliy of a financial sysem o excepional bu plausible macroeconomic shocks. Inernaional Organizaion of Securiies Commissions (1995) suggesed ha sress es analyzes he impac on porfolio of wors siuaion in marke (rise of ineres rae and crash of sock marke). In 1999, IOSCO specified ha sress es esimaes and measures exreme bu plausible risk of porfolio. Guidelines for he Sress Tesing of Commercial Banks by CBRC (China Banking Regulaory Commission) indicaes ha The erm sress esing as menioned in hese Guidelines refers o such a kind of risk analysis mehod in which he quaniaive analysis mehod is usually adoped, he frangibiliy of a single bank, a bank group and he banking sysem is evaluaed and judged hrough forecasing possible losses from small probabiliy evens or any oher exremely unfavorable circumsances as assumed and analyzing he negaive impac of such losses on he bank s profiabiliy and capial, and necessary measures are hus aken. Sress es can ake advanage of differen daa ses a eiher porfolio or aggregae level. Aggregae-level sress es direcly specifies he relaionship beween macro variables and banking aggregae elemens and analyzes he risk exposure of a group of reporing firms under cerain sress scenario. Porfolio-level sress es emphasizes on 16
he banks unique credi risk exposure o macro shock and discusses loss in each uni under excepional circumsances. Porfolio-level sress es is more operaional and includes feedback effec in financial insiuions. However, he mehodology requires larger sample size of hisoric daa, while Chinese commercial bank generally lacks long-erm easily-accessible and well-recorded daabase. Therefore, due o low availabiliy of daa, his paper decided o develop a sress es a aggregae level, which generally demands macroeconomic daa of he whole financial sysem. How o build credi model deermines he operaion of macro sress es. Framework from Wilson and Meron s models conribues mos o curren heoreical research of sress-es. Wilson (1997a, 1997b) seleced macro economical facors like he growh rae of GDP and ineres rae as risk facors, defaul raes as measuremen of credi asses, and hen esed in erms of he relaionship beween macro siuaion and defaul rae. Meron s model (1947) applied marke informaion (he sock price and bond price) o evaluae credi risk. As he marke informaion referred o he expecaion of invesors, Meron s model performs more prospecive han oher hisoric-daa based models. However, in realiy, Wilson s model is more operaional and more pracical. Consequenly, mos of macro credi models are in favor of Wilson s Credi Porfolio View. Like Mckinsey & Company, Singapore Minisry of Finance, Hong Kong Moneary Auhoriy, Bank of Finland, Bank of England, mos insiuions applies Wilson s framework o credi risk sress es model. Boss M (2002) developed Wilson s model by adoping aggregae-level defaul raes of firms o es Ausralian bank deparmen and discovered ha indusrial producion, inflaion rae, sock index, nominal shor-erm ineres rae and oil price significanly imposed on defaul raes of firms. Subjec of Virolaninen s (2004) research was Bank of Finland. He used firm s bankrupcy daa in he model concerning he relaionship beween macro facors and defaul raes. Empirical resul in his paper indicaed ha GNP, firm s loans and ineres rae significanly affeced defaul raes. Virolaninen creaed a variey of scenarios o conduc sress es and explored differen credi exposures of Bank of Finland o macro shocks. Jim Wong, Ka-fai Choi and Tom Fong(2008) suggesed he srong relaionship beween he defaul raes of bank loans and major macroeconomic facors, like Hong Kong s GDP, ineres raes and propery prices and he Mainland s GDP, They se wo pressure indicaor o conduc esing o reail banks in Hong Kong in erms of Mone Carlo simulaion. The scenario resembled Asian crisis in year 1998.Tes resul proved ha curren credi risk of banking sysem was moderae. Allan Kearns, Erlen maier and Gersbach H(2004) developed macro credi equaion o assess defaul rae of loans and hen respecively applied esing o Irish reail credi insiuion and Norway Cenral Bank. Recenly, Chinese scholars increasingly ook ineres in he research of macro sress es. From heoreical perspecive, Sun Lianyou (2007), Gao Tongyu and Chen Yuanfu (2006) reviewed relaed inernaional lieraures and heoreically discussed sress es in China. Xu Mingdong and Liu Xiaoxing (2008) compared several macro sress es mehodologies and briefly inroduced esing process as well as how he whole process guaraneed he sabiliy of financial sysem. Form empirical research perspecive, Li Jiang and Liu Liping (2008) regarded defaul raes of loans as he indicaor of evaluaing credi risk of banking sysem and seleced macro economical facors ha imposed on bank credi. They made use of Muli linear regression model o inegrae hese facors and finally generae one comprehensive macro facor. Apar from radiional macro facors, bank s degree of finance leverage was brough in by Zhou Yuan (2010) o build sress models, which argeed a commercial banks in Jiang Su province. Zhou Yuan assumed bad and very bad sress scenarios and laer explored ha under given bad sress, local banks would be in exremely hard ime. Chinese scholars preferred regarding NPL as kind of elemen o measure pressure olerance. Bu considering Chinese banks, especially Agriculure Bank of China, have removed pleny of NPL from balance, empirical research migh suffer from removal of NPL, where a urning poin of financial performance would exis. In order o build a more pracically oriened sress-es model, furher exploraion is required o find ou how macro shock would influence rae of NPL in commercial banks given he removal of NPL. The aricle improved pas model by bringing in a dummy variable o represen wrie-off NPL. This effor would ake he disracion of removed NPL away from he analysis how macro economy impacs credi risk in commercial banks. 3. Model and Daa This paper develops macro sress-es models by Wilson (1997a,1997b), Boss (2002) and Virolainen (2004) and creaes a new concepual model ha could be applied o Chinese commercial banks. Our framework consiss of wo pars: one par refers o empirical model concerning credi risk of commercial bank as well as macro economical sysem. The oher par ses up scenarios abou fuure and uses empirical model born in par one o conduc sress es. 3.1 Model Non-linear relaionship beween defaul raes and macro facors enables ransformaion from rae of NPL o Y in erms of logi model. Equaion (1) explains he logi-relaionship beween Y as mediaor variable and NPL: Published by Canadian Cener of Science and Educaion 17
Y 1 N P L ln ( ) N P L (1) Where NPL presens raios of non-performing loans a ime, Y is a aggregae grouping economical indicaor a ime. Equaion (1) implies ha Y is negaively relaed o NPL. More defaul raes lead o less Y Wih he mediaor variable, we can reasonably esablish sress model: Y m A X... A X Y... Y D v (2) 1 1 s s 1 1 k k X n B X... B X C Y... C Y (3) 1 1 p p 1 k k Where X is macro economical facor, D a new dummy variable ha defines he removal of NPL and is assgined 0 over he firs hree quarers in 2008, 1 since Q4. Associaion beween NPL and equaion (1) generaes a group componen of macro facors. If we pu he componen ino equaion (2), we can esimae coefficien of each macro facor and herefore he defaul-model. When we conduc sress esing and laer assume sress scenario, we can use (3) o esimae each macro facor. Associaion of esimaion wih equaion (2) leads o value of Y under given scenario. Y could be used in equaion o calculae raes of NPL. Equaion (2) reflecs he relaionship beween macro economical facors and group componen. The concepual model in his aricle includes he funcion of dummy variable D and he associaion of macro economical daa wih previous year s macro economical daa wih he use of lagged erms. Equaion (3) repors he ineracion effec among macro economical facors. This paper mainly uses VAR model o describe how hese facors are relaed o each oher. Vecor auo regression (VAR) is an economeric model used o capure he evoluion and he inerdependencies beween muliple ime series. All he variables in a VAR are reaed symmerically by including for each variable an equaion explaining is evoluion based on is own lags and he lags of all he oher variables in he model. As for he predicion of ime series of macro facors, mos lieraures applies auo regression (AR) model o individually predic each macro economical facor. Considering hese facors are highly sensiive o he flucuaion of one facor, we should build VAR model like equaion (3) o include he inerdependency among facors. And he key role of commercial banks in he economy of China deermines he inclusion of banking sysem s feedback effec on economy developmen. According o he underlying philosophy of model, shock from one single macro economical facor can be ransmied by model and hen impacs oher facors,. Finally, raio of NPL will be changed. Therefore, banking sysem could make use of lagged NPL o impose on macro economy. One Advanage of his model comes from he logic ha changes in macro facors and changes in NPL barely happen simulaneously, and we inroduces a dummy variable o be a proxy of wrie-off NPL in Agriculure Bank of China; Addiional advanage follows he idea ha VAR model should be a beer opion o reflec he ineracion of facors and should have he feedback effec of commercial bank on macro economy. Given he perspecive over he fuure and he role of commercial bank in economy developmen, he model is a good fi o pracice in China. 3.2 Daa Defaul raes ermed as NPL in his analysis are drawn from CBRC s daabase of defaul raes. CBRC daa covers from Q1 in year 2004 o Q4 in 2010. In addiion, we collec daa of macro facors over he same period from Rui Si s and Zhong Hong s daabases. 3.2.1 Dependen Variable Credi risk refers o he poenial possibiliy ha bank borrower or dealer fails o comply wih he original conrac. NPLis one of imporan indicaors o evaluae saus of porfolio in commercial bank. As NPL rae ges higher, bank akes higher risk o call in a loan. Therefore, NPL rae can reasonably represen he defaul risk of commercial bank. 3.2.2 Independen Variable As China redefined NPL of commercial banks in year 2004, daa of NPL is limied. In virue of he low availabiliy of daa, we refer o he variable from inernaional lieraures. A prior es helps us o selec following variables in Table 1 o represen he macro economy. 18
Table 1. Definiion of variable variable Name Descripion Real GDP Growh GGDP I reflecs economical level in a specified period and is a basic indicaor o reflec he healh saus of economy. I measures he flucuaion in he price level of cusomer goods and services purchased by households Cusomer Price Index and play a core role in marke aciviy and moneary CPI policy. Housing loan accouns for a large par in banking House price index PI loans and many loans require house o be morgage. The flucuaion of housing price affecs NPL grealy. Nominal Moneary growh GM2 I reflecs he presen moneary policy in China Dummy Variable D In erms of volaile naional NPL over years, he whole naion or each province s commercial NPL shows a urning poin. 4. Empirical Resul 4.1 Model Esimaion 4.1.1 Daa Preprocess Based on equaion (1), we found ou Y as he proxy of NPL. Resuls from Dickey-Fuller es prove ha no all ime series are as sable as hose lagged on year. Johansen and Juselius es for mulivariae coinegraion is used o disclose ha coinegraion makes sense owards he inerdependency among variables and he relaionship beween variables and mediaor. 4.1.2 Empirical Research Based on equaion (2), we associae all economical variables wih he equaion and make use of Eviews 6.0 o conduc muli-linear regression. Afer sep-by-sep regression and selecion in erms of he goodness of fi, AIC, SC guidelines, Table 2 repors more reliable regression resuls. The resuls indicae ha GGDP, CPI, PI, GM2, Y(-1),PI(-1) and dummy variable D are significan. Table 2. Muli-linear regression and vecor auoregression resuls Dependen Independen Variables variables Y GGDP CPI GM2 PI Inercep -3.266220* 14.40576-17.6452 77.76263*** 4.727057 GGDP 0.069473*** CPI 0.032249*** GM2 0.031535* PI -0.040289*** GGDP(-1) 0.852448*** 0.395101*** -0.48173*** 0.554839*** CPI(-1) -0.22641*** 0.785986*** GM2(-1) 0.339725*** 0.629976*** PI(-1) 0.037683** 0.292577** -0.58382*** 1.028439*** Y(-1) 0.551769*** -0.57868*** -0.76251*** 2.099909*** -1.58407*** D 0.844701*** R-squared 0.994219 0.785658 0.824484 0.944714 0.890078 ADJ.R-squared 0.99209 0.734624 0.782695 0.93155 0.863906 AIC -2.003143 2.717686 3.290962 3.237944 3.5488 obs 27 27 27 27 27 Noe: *, **, *** indicae significance a he 10%, 5% and 1% levels respecively. Then, we ake advanage of equaion (3) o complee auo regression owards each dependen variable. Empirical Published by Canadian Cener of Science and Educaion 19
resuls in able 2 show ha variable lagged on year performs bes and herefore we should build firs-order VAR Laer AR uni es is used o in he previous model. Since all 4 characerisic roo lie inside uni circle, he VAR model is sable. And able 2 demonsraes ha independen variables are lagged o cerain degree. And we can see he independency among lagged variables. Moreover, all he macro economical variables are impaced by previous grouping componen. 4.2 Regression Analysis Table 2, elaboraes ha dummy variable D is significan (0.844701) in he model. In Q4, 2008, remarkable removal of NPL from balance of Agriculure Bank of China conribues grealy o reducion of NPL in naional banking sysem. Four macro economical variables are significanly relaed o grouping componen. And NPL is mos sensiive o flucuaion of GGDP (0.069437). High GDP growh means he boom of naional economy. In he bullish years, firm makes high profi and has beer abiliy o make paymen, which makes NPL decline. Movemen of GM2 predics changes in moneary policy. When cenral bank acceleraes money supply, governmen is in favor of deregulaion. Sufficien money supply lowers cos of finance, increase profiabiliy and reduce non-performing loans. Loose moneary policy leads o inflaion. The process could be seen from VAR model where lagged GM2 should ake responsible for par of changes in CPI. Moderae rise in CPI indicaes adequae money supply and is a good signal o NPL on he balance of commercial banks. PI and Pi -1 are negaively significan o Y. Rise in housing price lagged one year means ha real esae marke is in prosper, which indicaes ha capain chain in he marke is in good saus and solvency is srong. Due o he increasing price index, a large amoun of funds are araced ino real esae marke. Overcapialized housing marke akes risk of producing economic bubble, which enables more defaul. Therefore, our model is pracically oriened. 5. Simulaion of Macro Shock and Analysis This paper applies sensiiviy analysis and scenario analysis o sress esing. Given independen variables in he aricle, we se up wo sress scenarios. Figure 1 depics how NPL responds o flucuaion of risk facors in he firs scenario. In order o accuraely perceive he resul, we assume ha each risk facor will be changed by he same proporion. The second scenario simulaes he wors siuaion when all he exreme shocks happen in one year. We will analyze NPL in unanicipaed ough ime. Y Figure 1. Scenario 1 sress esing graph 5.1 Sressed Scenarios Firs scenario: Suppose growh rae of GDP, CPI and GM2 will be reduced by 3% in Q1 2011 while PI will increase by 3%, he new esimaion should be 7.3% (GDP), 100.3 (CPI), 17.8% (GM2), 112.4 (PI), which are respecively refleced in Figure 1 ermed S1, S2, S3 and S4. Second scenario: Suppose here will be a fall in GDP by 3.1% and CPI by 6.5% in Q1 2011, which resembles Asian crisis over he period from Q4, 2008 o Q1 2009. Meanwhile, we reain he supple of moneary. Then, afer GDP and CPI are reduced by he same proporion, we injec more money ino marke like wha Chinese governmen decided in Asian crisis. 5.2 Sress Tes Resul Based on VAR model, we look ino he fuure and hen obain baseline scenario. Firs of all, we analyze predicaion in firs scenario. The resuls show ha grouping componen will be lowered o a cerain degree due o he poor performance of macro facors. And in our aricle grouping componen is conras wih NPL. In a shor 20
ime, NPL rae is more sensiive o changes in PI han o oher macro facors. In Q1 2011, he movemen of NPL experienced unimaginably drasic flucuaion, bu he ineracion among macro facors in model is able o smooh NPL curves and he urning poin appears quickly. In a long ime, movemens of GDP and GM2 are he mos volaile and have he longes influence. Then, we sar o analyze he scenario wo. Table 3 gives more deails abou he resul. In he wors scenario, we can see an obvious rise in NPL of Chinese commercial banks when here is downside in naional economy and slump in CPI.. Comparison beween 1 s scenario and 2 nd scenario leads o he discovery ha QE (Quaniaive Easing) program in economic depression generaes more money supply and efficienly sops he increase in NPL. QE helped Chinese governmen survive in he las financial crisis. Therefore, in case of high CPI, Chinese governmen could decrease money supply moderaely, which minimizes he negaive influence on NPL of commercial bank and saisfies he requiremen of moneary policy. Table 3. Sress es resuls in scenario 2 baseline 1s scenario 2nd scenario Y NPL Y NPL Y NPL 2011q1 4.510099 0.010878 4.166432 0.015271 4.296774 0.01343 2011q2 4.509237 0.010887 4.116071 0.016047 4.188899 0.014936 2011q3 4.482648 0.011177 4.12787 0.015862 4.194392 0.014856 2011q4 4.475791 0.011253 4.195042 0.014846 4.270543 0.013782 In 1s scenario, daa from he end of 2010 indicaes ha accumulaed amoun of loans by 2010 in China reaches 31704.3 billion. Reserves agains loan loss and depreciaion are 795.943 billion. In depression, NPL peaks a 1.6047% and he dollar amoun is 509.33659B. Obviously, in hese wo sressed scenario, bank sysem as a whole can depend on reserves o cover risk exposures. The asses are relaively safe. 6. Conclusion This paper improves he sress-esing model prevailing globally. We use pas demographic dae of China o consruc pracically oriened macro sress-esing model. Iniially, we ake non-linear relaionship beween macro economical facors and defaul raes ino consideraion. Therefore, logi funcion is applied o find he proxy of NPL. Under he condiion ha Chinese governmen wroe off pleny of bad asses from he balance of ABC in Q4 2008, his paper inroduces a dummy variable o represen he removal of NPL and enables a corresponding funcion. Finally, we make use of ineracion beween macro facors in VAR model o predic a baseline in he fuure.. As for he selecion of macro facors, we follow he idea from inernaional and domesic model and reasonably choose GGDP, CPI, GM2 and PI as macro facors. And he selecion akes ino consideraion he limiaion like less ransparency in economical daa saisics and disclosure han wesern counries. This paper specifies macro facors and NPL rae of commercial bank from Q1 2004 o Q4 2010, oally 28 quarers, and hen uses Eviews 6.0 sofware o analyze. Resuls show ha dummy variable is significanly relaed o NPL rae wih a comparaively higher coefficien, which means ha NPL of naional commercial bank as a whole exremely benefis from removal of NPL in ABC. If we go rid of he wrie-off NPL in our model, i is impossible o ell he influence of removal from ha of macro shocks. A he same ime, four macro facors have song relaionship wih NPL rae. Afer we apply sensiiviy analysis and scenario analysis o sress esing credi risk in naional commercial banks, we observe ha from shor-erm perspecive, NPL is more sensiive o PI han o oher hree macro facors. Movemen of NPL rae has quie winding pah, bu he ineracion among macro facors in model framework is able o smooh NPL curves. In long erm, GGDP and GM2 impacs NPL mos and has longes-lasing influence. Based on he daa from he end of year 2010, his paper simulaes scenario ha we experienced in 2008 financial crisis and suggess ha when bearish happens and NPL surges, naional reserves agains loan loss and depreciaion can resis o such shock. Meanwhile, in depression, loose moneary policy simulaes economy and efficienly preven furher increase in NPL in beer economical environmen. References Allan Kearns. (2004). Loan Losses and he Macro economy: A Framework for Sress Tesing Credi Insiuions Financial Well-Bein. Financial Sabiliy Repor, 111-121. Boss M. (2002). A Macroeconomic Credi Risk Model for Sress Tesing he Ausrian Credi Porfolio. Financial Sabiliy Repor, (4), 64 82. Published by Canadian Cener of Science and Educaion 21
Meron R. (1974). On he Pricing of Corporae Deb :The Risk Srucure of Ineres Raes. Finance, 29(7), 449-470. Sun Lianyou. (2007). Sress Tesing of Commercial Bank. Guangxi Financial Research, (11), 19-22. Virolainen, K. (2004). Macro Sress-Tesing wih a Macroeconomic Credi Risk Model for Finland. Bank of Finland Discussion Paper. Wilson T C. (1997b). Porfolio Credi Risk II. Risk, 10(10), 56-61. Wilson T. C. (1997a). Porfolio Credi Risk I. Risk, 9(10), 111-117. Wong J., Choi K. F., & Pak-Wing Fong. (2008). A Framework for Sress-Tesing Banks Credi Risk. The Journal of Risk Model Validaion, (SPRING), 3 23. Xu Mingdong, & Liu Xiaoxing. (2008). Financial Sysem Sabiliy Assessmen: Inernaional Comparison Based on Macro - Sress Tesing. Sudies of Inernaional Finance, (2). 22