玉镯的保养
The relationship between net interest margin and noninterest income using a system estimation approach
James Nguyen ⇑
Johnson C.Smith University,100Beatties Ford Dr.,Charlotte,NC 28216,USA
黑蜂a r t i c l e i n f o Article history:
Received 24February 2009Accepted 20April 2012
Available online 30April 2012JEL classification:D40G20G21
Keywords:Bank margins Market structure Diversification
Noninterest income
a b s t r a c t
This paper examines the determinants of bank net interest margin (NIM)and non-traditional banking activities (NII).A system estimation approach is employed to control for the simultaneity between NIM and NII for commercial banks in a group of 28financially liberalized countries during the period between 1997and 2004.We find a statistically significant negative relationship between NIM and NII for the period between 1997and 2002.A generally positive but statistically insignificant association between NIM and NII is found for the subquent period (2003–2004).Banks’increasing involvement in non-traditional activities is negatively correlated with risk-adjusted profitability measures in the for-mer subperiod,suggesting no obvious diversification benefits.However,the share of noninterest income is positively related to the return on asts (ROA)and the return on equity (ROE)for the latter subsample.
Ó2012Elvier B.V.All rights rerved.
1.Introduction
The recent decline in NIMs reflects increasing competition on the returns on earning asts and the cost of bank funds.For instance,the growth in core deposits at banks has decread becau customers now have other alternatives that offer similar rvices and pay higher interest rates such as cash management accounts and mutual funds.Loan yields have also declined due to competition from nonbank creditors such as finance and leasing companies.In addition to traditional bank lending,banks engage in non-traditional off-balance-sheet (OBS)rvices.NII now makes up approximately half of all operating income generated by US commercial banks and a significant amount of total income in many mature economies.OBS products provide fee income but are not shown as asts or liabilities on the balance sheet.NII gen-erated by OBS activities includes trading gains and fees,investment banking and brokerage fees,net rvicing fees,insurance commis-sions,net gains on ast sales,fiduciary income,net curitization,rvice charges on deposit accounts,other foreign transactions and other noninterest income.Incread competition has reduced the comparative advantage banks had in obtaining funds and has weakened their position in the loan market.Banks,therefore,have tried to offt this reduction in NIM with NII derived from non-traditional banking activities.
啦啦操串词
The relationship between NII and NIM has important implica-tions for business strategy and regulatory policy.A positive relationship between NII and NIM,for instance,indicates that the shift towards non-traditional activities is beneficial to banks as it contributes to the improvement in NIM.In other words,high performing banks can have high levels of NII and NIM.Therefore,policies aimed at allowing banks enter new markets to engage in new OBS activities are likely to yield desirable outcomes.Also,extensive regulation limiting bank activities may be no longer as esntial since the freedom to offer a variety of rvices may also help banks compete more effectively with non-bank financial insti-tutions.Some authors (Rogers and Sinkey,1999;Smith et al.,2003;Lepetit et al.,2008a;Fu and Heffernan,2010)have obrved that NIIs in some countries em to have incread at the expen of NIMs.One explanation for this negative relationship between NIM and NII is that banks increa their involvements in OBS activ-ities to offt the decline in margin income.It could also be the ca that OBS activities generate higher net interest margins through interest and fee income banks charge to compensate,for instance,for providing the line of credit or options included in the OBS con-tracts (Angbazo,1997).1In a related study,Boyd and Gertler (1994)show that the movement towards off-balance sheet activities does
0378-4266/$-e front matter Ó2012Elvier B.V.All rights rerved./10.1016/j.jbankfin.2012.04.017
人像摄影图片Tel.:+12817236536;fax:+17043783543.
E-mail address:acafi
1
A positive association between NIM and NII is also consistent with the underinvestment hypothesis by James (1988)which suggests that non-traditional activities create a more diversified ‘‘margins-generating’’ast ba than under equity or deposit-financing.
not indicate that banks are abandoning their traditional lines of busi-ness.Rather,there has been‘‘at most’’a slight decrea in commercial banks’share offinancial intermediation.Market-share intermedia-tion numbers fail to capture the relative increa infinancial interme-diation.In fact,their data suggest that banking as a component of gross product actually incread in importance and that traditional banking is not‘‘dead’’yet.Davis(2002)finds that the relationship be-tween NII and NIM is positive for most of the countries in his study. Stiroh(2004)obrves an incread correlation between net interest income growth and noninterest income growth at US banks,possibly due to an incread focus on cross-lling different products to the same customer ba leading to reduced diversification benefits.Valv-erde and Fernandez(2007)show that,under certain assumptions,NII is ne
gatively related to NIM.Empirically,however,theyfind that the income from off-balance-sheet activities is positively related to one of the proxies for traditional activities.In short,the empirical evi-dence about the NIM versus NII link is rather weak.In this context, the main contribution of this paper is to revisit this issue with a large panel data t and recent empirical methodologies.
In particular,a thorough investigation of the factors influencing NIMs must explicitly account for the potential impact of NII on bank spreads.Prior studies establishing a link between NII and NIM estimate the variables parately.By contrast,we jointly estimate NIM and NII to better account for the simultaneity be-tween NIM and NII.To our knowledge,this is thefirst attempt to u this approach to maximize statistical efficiency and to exploit the benefits offered by perhaps the largest panel data t to date. While some authors(Stiroh,2006;Laeven and Levin,2007;Baele et al.,2007)have addresd the potential issue of endogeneity be-tween NIM and NII,we attempt to tackle this issue with a simulta-neous-equation approach.Since endogeneity is effectively controlled by the framework of simultaneous equations approach, we simultaneously estimate NIM and NII with a system GMM (Generalized Method of Moments)methodology which exploits the interactions among their innovations.Endogeneity in the bank-ing literature refers to the possibility that banks u diversification benefits to engage in riskier OBS activities,leading to biad results due to obscurity in
the interpretation between diversification and risk(Laeven and Levin,2007;Baele et al.,2007;Elsas et al.,2010; and others).The endogeneity in the prent study is an economet-ric problem relating to the simultaneity between NII and NIM and is discusd in Section2.3.
In addition,we also study the determinants of NIM and examine the factors commonly found in banks heavily involved in non-tra-ditional activities.It is not clear whether the potential benefits of diversification outweigh the costs.Previous empirical studies mostly examine the relationship between risk and the shift to-wards non-traditional banking activities using either North Amer-ican or European data.While many studiesfind that incread reliance on non-interest income is associated with higher volatility of the accounting profits(Stiroh,2004;Stiroh and Rumble,2006; Calmès and Liu,2009),others show a strongly positive relationship between diversification and bank profitability(Baele et al.,2007; Elsas et al.,2010;Calmès and Théoret,2010).Therefore,we aim to check if the results from our28-country sample differ from tho of prior studies.In particular,we study whether OBS activi-ties lead to higher risk-adjusted profit measured by risk-adjusted return on equity(RAROAE)and risk-adjusted return on ast (RAROAA)by regressing a t of control variables such as ast growth,ast size,the share of non-traditional activities,the ratio of loan loss provision to total ast and the equity over ast ratio on RAROAE and RAROAA.
Our mainfinding is that NIM is not always inverly related to NII.Similar to many prior results from studies examining NIM and NII parately,a statistically significant negative relationship between NIM and NII is obrved for the period between1997and2002,implying evidence of cross-subsidization.However,the association between NIM and NII is positive but insignificant for the subquent period.Similarly,banks’increasing involvement in non-traditional activities is negatively correlated with risk-adjusted profitability measures(risk-adjusted return on ast and risk-adjusted return on equity)for the period between1997and 2002.Further,the coefficients of the noninterest income share become positive for the subquent period(after2002)for both return equations.While the Kendall’s tau rank-order correlation test confirms the results from thefirst subperiod,there exists a positive but statistically insignificant correlation between nonin-terest income share and return on asts in the latter subsample. The paper is structured as follows.Section2.1prents the defini-tions of the variables ud in our study.Section2.2discuss the data.Section2.3prents the methodology.Results and robustness are detailed in Section3.Section4concludes the study with some additional remarks.
2.Data and method
2.1.Variable definitions
Bad primarily on the theoretical models of Valverde and Fernandez(2007),Maudos and Guevara(2004),Wong(1997), Angbazo(1997),Zarruk and Madura(1992),and Ho and Saunders (1981),NIM is shown to be related to market structure,operating costs,level of risk aversion,interest rate risk,credit risk,covariance of interest rate risk and credit risk,size of deposits and loans gen-erated,total volume of credit granted,capital adequacy,and nonin-terest income.We employ the following variables in our study to proxy for the above factors.
NIM it¼C iþb1BANKHI itþb2NIEAA itþb3ETA itþb4LACSTF it þb5LLRGL itþb6Co v itþb7LNTA itþb8LNLO it
þb9CFTA itþb10OEATA itþe itð1Þwhere NIM it is net interest margin which reprents the net interest income expresd as a percentage of earning asts and is the proxy for traditional banking activities ud in recent studies.The higher this ratio,the higher the margin the banks earn.Market structure (BANKHI)is calculated as the square of the ratio of each bank’s total deposits to total deposits within the banking ctor of the country in which the bank is bad.Numerically,BANKHI=(TDi/TD)2where TD i is bank i’s total deposits and TD is total deposits within the banking system.The BANKHIs obtained from this method range from0to1in the ca of a completely concentrated market.The ra-tio of noninterest expen to total asts(NIEAA)is ud to proxy for operating costs.
Noninterest expen plus provisions for loan loss provide a measure of the cost side of the banks’performance relative to the asts invested.The ratio of equity to total asts is ud to proxy for the level of risk aversion(ETA)as in Maudos and Guevara(2004).This capitalization measure is an imperfect proxy for risk aversion and should be interpreted accordingly.Flan-nery and James(1984)show that interest rate risk exposure is in-verly related to the average maturity of asts.The higher the level of short-term asts,for instance,the smaller the nsitivity to near-term interest rate changes which may lead to lower interest rate premium.The ratio of liquid asts to customer and short-term funding(LACSFT)is ud as a proxy for inver interest rate risk. The ratio of loan loss rerves to gross loans is ud as a proxy for credit risk(LLRGL).The ratio indicates how much of the total loan portfolio has been provided for,but not charged off.It is a rerve for loss expresd as a percentage of total loans.The covariance of credit risk and interest rate risk(COV)is calculated by taking the product of the values of credit risk and interest rate risk.LNTA,
2430J.Nguyen/Journal of Banking&Finance36(2012)2429–2437
the logarithm of total asts,is ud to proxy for the size of opera-tions.LNLO(in logarithm)proxies for the volume of loans(credit) granted is ud for this variable.The ratio of capital funds to total asts is employed to measure capital adequacy(CFTA).Consistent with Valverde and Fernandez(2007),the
ratio of other earning as-ts to total asts(OEATA)is included to capture the extent of diversification towards OBS activities.OEATA is also less prone to the problems pointed out by Clark and Siems(2002).2First,accord-ing to Clark and Siems(2002),noninterest income,a measure com-monly ud to proxy for OBS activities in many studies,may inflate the amount of OBS activities becau fees and commissions are drawn from on-balance-sheet activities.Second,the credit-equiva-lent measure of OBS activities can substantially underestimate the level of non-traditional activities.Third,the revenue-bad ast-equivalent measure can also distort the measurement.
A review of the literature suggests that noninterest income is a function of bank size,the bank’s debt structure,credit risk,interest rate risk,liquidity risk,overheads,loan loss provisions,and before-tax profit.3The following proxies are ud to estimate the model: OEATA it¼C iþb1LNTA itþb2LNTD itþb3LLRGL itþb4LACSTF it
þb5NLTD itþb6OVTA itþb7LLPTA itþb8BTPTA it
þb9NIM it V itð2ÞWe u the ratio of other earning asts to total asts(OEATA) as our main proxy for non-traditional activities for reasons de-scribed in the preceding paragraph.For the explanatory variables, the following measures are ud.Bank size is measured by the log-arithm of total asts(L
NTA).The logarithm of total deposits is ud to proxy for the bank’s debt structure(LNTD).The ratio of loan loss rerves to gross loans is ud to reflect credit risk (LLRGL).The ratio of liquid asts to customer and short-term funding is employed to proxy for interest rate risk(LACSTF).Net loans to total asts(NLTD)is a proxy for liquidity risk since it measures the percentage of bank asts tied up in loans.The ratios of Overheads(OVTA),Loan Loss Provision(LLPTA),and Before-tax-Profit(BTPTA)to total asts are constructed from the databa.
2.2.Sample
We u a sample of commercial banks from the28countries classified as‘‘financially liberalized’’by Kaminsky and Schmukler (2003).Unlike most previous studies,which lump countries to-gether bad on geographical regions and focus on broad issues, our study concentrates exclusively onfinancially liberalized mar-kets.This sample lection is important becau it eliminates some of the concerns about significant differences infinancial liberaliza-tion policies among banks in many less advanced nations.4The source of data is Bankscope,a standardized collection of bank state-ments put together from reports that banks issue in accordance with reporting requirements established in the countries.Our data con-tains an unbalanced panel of3593commercial banks and all ratios regarding bank-specific characteristics are calculated using the stan-dardized global reporting format
to ensure that they are comparable. Prior to the estimation,we test the variables for the prence of non-stationarity using the Panel Unit Root tests developed by Choi(1999),
Maddala-Wu(2001),and Im et al.(2003).The results,significant at
the1%significance level,generally indicate that tested ries do not
contain a unit root.The ries are therefore estimated in levels.
2.3.Model specification
Eqs.(1)and(2)are estimated jointly for the following reasons.
On the surface the equations may appear to be emingly unre-
lated to each other.However,since they are using the same data,
the error terms between the two equations may be related.If unac-
counted for,the apparent simultaneous-equation bias from Eqs.(1)
and(2)can lead to biad and inconsistent estimators becau of咎狗之血动漫
the correlation between the random errors and the endogenous
variables.The contemporaneous correlation between e it and v it is also due to the fact that the errors contain the influence of fac-
tors that have been omitted from the equations.Since thefirms
are similar in many respects,it is likely that the effect of the omit-
ted factors on NIM and NII for onefirm is similar to the impact on
NIM and NII for anotherfirm.If this is the ca,e it and v it are cap-turing similar effects and will be correlated.One potent solution to this problem is to estimate the two equations jointly using the pa-nel Generalized Method of Moments(Baltagi,2008).In panel data, the GMM estimator has been shown to be more efficient than the fixed effects or random effects estimators if the strict exogeneity assumption of the regressors fails or if rial correlation is prent.5 Since endogeneity is effectively controlled by the framework of simultaneous equations approach(Greene,2008)all estimations in Section3.1are done with a system GMM approach which exploits the interactions among the innovations in Eqs.(1)and(2).Woold-ridge(2001)shows that the GMM estimator brings efficiency gain
s in the prence of endogenous explanatory regressors.We further control for heteroskedasticity and arbitrary autocorrelation by employing the Newey–West methodology(Newey and West, 1987)in estimating Eqs.(1)and(2).
3.Results and robustness
3.1.Preliminary results
Table1shows the summary statistics for the main variables in the study.Averaging across all obrvations,the mean NIM is4.37. The sample variation is substantial as indicated by large standard deviations.The mean value for NIRAA is3.68and its sample vari-ation is smaller.WJNIMP and PNIMP average3.65and4.38,respec-tively.The average OOIAA for the sample is2.39with a relatively high dispersion.OEATA is relatively small and has little sample variability.All ries are normally distributed,according to Jac-que–Bera statistics(not shown).Table2displays the graphical rep-rentations of the data.The data in Table2generally confirm the common obrvation in the literature that the level of NII has been steadily increasing in recent years.NIM gradually decreas from 3.3in1997to2.7in2003before skyrocketing in2004.There ap-pears to be a weak inver relationship between NII and NIM dur-ing the period.This obrvation is in accordance with the data
on a few European countries indicating that NIM and NIM move in opposite directions.6However,Granger causality tests for NIM and NII indicate a two-way causal relationship between NIM and NII. That is,bad on the data we cannot reject the hypothesis that NIM Granger caus NII and vice versa.Also,structural breaks em to exist in the data around2002for many data ries.The breaks em to have occurred around the date that thefinal consultative
2We also u another proxy for non-traditional activities that is not prone to the problems:the ratio of other operating income to total asts(OOIAA).The empirical results are generally similar and are not reported to conrve space.
3For details,e Diamond(1984),Hunter and Stephen(1986),Hunter and Timme (1986),James(1988),Demtz and Strahan(1997),Angbazo(1997),Claesns et al. (2001),Davis(2002),Koch and MacDonald(2003),DeYoung and Rice(2004),and Valverde and Fernandez(2007).
4The countries in our study are Argentina,Brazil,Canada,Chile,Colombia, Denmark,Finland,France,Germany,Hong Kong,Indonesia,Ireland,Italy,Japan,
Korea,Malaysia,Mexico,Norway,Peru,Philippines,Portugal,Spain,Sweden,Taiwan, Thailand,Venezuel
a,United Kingdom,and the United States.5See Im et al.(1999),Wooldridge(2001),and Baltagi(2008)for details. 6See,for instance,Smith et al.(2003)and Lepetit et al.(2008a).
J.Nguyen/Journal of Banking&Finance36(2012)2429–24372431
document of the Bal II Accord was published by the Bal Committee.
元气淋漓Table3shows the Granger causality tests of the main variables in the study.NIM is defined as the difference between a bank’s interest income and interest expen expresd as a percentage of average interest-earning asts.NIRRA is the ratio of net interest revenue to average asts((Net Interest Revenue/Average Earning Asts)Ã100).WJNIMP is calculated as the ratio of net interest margin to total asts as employed in Williams(2007).PNIM is constructed as the ratio of net interest income to total earning as-ts as ud in Lepetit et al.(2008a).For ea of prentation,only relevant information is reported since we are particularly inter-ested in the effects of NIM on NII.The tests are done using one lag and two lags as commonly suggested in the Econometric liter-ature.7The full Granger-causality test results are later confirmed by most of thefindings prented in Tables4and5.In particular,prox-ies for traditional activities(NIM,NIRAA,WJNIMP,and PNIMP)gen-erally Granger-cau OEATA at a5%significance level or better.Since the variables are causally relat
ed in both directions,it is plausible that some third influence is causing them.To illustrate,according to Table3,for OEATA to Granger cau NIM the F-test for the null hypothesis that OEATA does not Granger cau NIM should be signif-icant and the F-test for the null hypothesis that NIM does not Gran-ger cau OEATA should be insignificant.In most cas,however, there are bidirectional causal relationships among the variables, implying that the variables are significantly related.Overall,we can reject the null hypothesis that NIM,NIRRA,PNIM or WJNIMP has no impact on OEATA and vice versa.It is important to control for the feedback problems using a system estimation method since this simultaneous equation bias can lead to inconsistent estimators.83.2.Regression results
Table4reports the regression results for Eq.(1)using four dif-ferent proxies for traditional banking activities.Since the Chow’s tests indicate a structural break near the end of2002,the sample is subdivided into two periods,1997–2002and2003–2004,and is re-estimated.The structural break occurs in the same year that thefinal consultative document of the Bal II Accord was pub-lished by the Bal Committee in2003.The break can also be en from many graphs prented in Table2.All versions are estimated with Panel GMM method.We employ the Newey–West(Newey and West,1987)methodology to account for heteroskedasticity and autocorrelation.For ea of exposition,we attempt to provide general interpretations of the significant or interesting variables prented in the following tables.
According to Table4,market concentration as proxied by BANK-HI is positively correlated with NIM,especially during thefirst sub-sample.Operating cost,NIEAA,has a significant direct effect on NIM at1%significant level,suggesting that banks with high oper-ating costs choo to work with higher NIM.More risk-aver banks,as measured by ETA,charge higher margins to compensate for the higher costs of equityfinancing(Berger,1995).LACSTF is in-verly related to NIM,suggesting that banks may require lower margins if they can obtain more short term asts.This relation-ship is not statistically significant for thefirst period,however. The coefficient on credit risk,LLRGL,has an insignificantly negative sign on NIM for thefirst period.It becomes insignificantly positive after2002.Theory(Maudos and Guevara,2004)suggests thatfirms with higher credit risk should work with higher NIM to maintain the same profit level.While higher provisions indicate the proba-bility of potential future loan loss,they could also signal a timely recognition of bad loans by prudent banks.The relationship be-tween NIM and LLRGL,therefore,is ambiguous.In a related study, Brock and Suarez(2000)find a significantly positive relationship between spread and credit risk for Colombia but a negative associ-ation between the variables for other Latin American countries. One explanation for the direct correlation between LLRGL and NIM,bad on the authors’reasoning,is that banks with adequate
Table1
Descriptive statistics.
Variable Mean Median Maximum Minimum Std.dev.Obrvations
NIM 4.3664 3.0120918.3090À30010.262718,240 NIRAA 3.6765 2.7200100.9800À35.1350 4.492918,258 WJNIMP 3.6496 2.6422647.8102À35.13537.552918,057 PNIMP 4.3776 2.91292133.3330À300.621118.309218,036 OEATA0.37920.3285 1.093500.238418,787 OOIAA 2.3868 1.1305302.6040À101.15608.228518,228 NIEAA 5.1416 3.2690287.5000À27.13008.304418,273 ETA12.42498.1515100À531.946016.427118,994 LACSTF31.298317.6790988.5710À5050.293316,236 LLRGL 4.9409 2.4580750À0.674010.546914,896 BANKHI0.0013 1.13EÀ06100.014018,820 LNLO 6.7318 6.821013.3299À4.1352 2.512518,585 LNTD7.10687.177013.5380À6.2146 2.371418,716 COV163.679835.674296610.5000À978.19021144.439012,567 LNTA7.54037.513514.0258À0.6812 2.183919,002 CFTA13.40819.7500100À56.779013.24906768 NLTD52.225856.850099.9900À20.754024.208418,787 OVTA0.042660.02689.1642À0.09770.105518,273 LLPTA0.00100.0036 5.7847À1.79240.061517,241 PBTTA0.01010.0091 5.8299À3.14620.085518,377辣木
Notes:NIM=net interest margin,NIRAA=ratio of net interest revenue to average asts,WJNIMP=ratio of net interest margin to total asts,PNIMP=ratio of net interest income to total earning asts,OEATA=ratio of other earning asts to total asts,OOIAA=ratio of other operating income to average asts,NIEAA=ratio of noninterest expen to total asts,ETA=ratio of equity to total asts,LACSTF=ratio of liquid asts to customer and short term funding,LLRGL=ratio of loan loss rerves to gross loan, BANKHII=bank concentration index,LNLO=log of loans,LNTD=log of total deposits,COV=covariance of credit risk and interest rate risk,LNTA=log of total asts, CFTA=ratio of capital funds to total asts,NLTD=ratio of net loans to total asts,OVTA=ratio of overheads to total asts,LLPTA=ratio of loan loss provisions to total asts,PBTTA=ratio of before-tax profit to total asts.
7See Wooldridge(2001).
8We also perform the Granger Causality test using OOIAA,the ratio of other
operating income to total asts,as a proxy for non-traditional activities.Once again,
there exists a bidirectional linkage between the measures of OBS and traditional
activities.The results reinforce the need to estimate the model simultaneously as
Eqs.(1)and(2)are not only theoretically related but also exhibit bidirectional
Granger causality.
2432J.Nguyen/Journal of Banking&Finance36(2012)2429–2437
provisioning of loan loss may experience incread margins.In other words,in the prence of sufficient loan loss rerves higher non-performing loans do not necessarily lower bank spreads.Fu and Heffernan(2010)similarlyfind a positive association between NIM and LLRGL for Chine banks.Both LNTA and LNLO are signif-icantly related to NIM following thefirst period only.It is also interesting to note that the impact of ETA on NIM becomes statis-tically insignificant after2002.
A centralfinding from this study is that OBS activities,as mea-sured by OEATA,are inverly associated with NIM for the period between1997and2002.This result is consistent with most prior findings from data for most European banks that NII has been growing at NIM expen.It appears that banks price loans to stim-ulate the sales of non-traditional rvices,supporting the loss-lea-der hypothesis.However,this association is positive but statistically insignificant after2002.In particular,the coefficients of OEATA on all four proxies for NIM become positive and insignif-icant for the cond subperiod.A possible explanation for this phe-nomenon is that banks may have adapted t
o new non-traditional business lines and noninterest income no longer increas at the detriment of net interest margin.This may be explained by the notion of‘‘learning by doing’’or cost adjustments suggested by
微课模板
