The Dynamics of Operating Income in the Norwegian Banking Sector

Andersen, Henrik; Berg, Sigbjørn Atle; Jansen, Eilev S. Working Paper The Dynamics of Operating Income in the Norwegian Banking Sector Working Paper, No. 2008/13 Provided in Cooperation with: Norges Bank, Oslo Suggested Citation: Andersen, Henrik; Berg, Sigbjørn Atle; Jansen, Eilev S. (2008) : The Dynamics of Operating Income in the Norwegian Banking Sector, Working Paper, No. 2008/13, ISBN 978-82-7553-451-2, Norges Bank, Oslo, https://hdl.handle.net/11250/2497772 This Version is available at: https://hdl.handle.net/10419/209904 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/deed.no ANO 2008/13 Oslo August 12, 2008 Working Paper Financial Markets Department The dynamics of operating income in the Norwegian banking sector by Henrik Andersen, Sigbjørn Atle Berg and Eilev S. Jansen Working papers fra Norges Bank kan bestilles over e-post: [email protected] eller ved henvendelse til: Norges Bank, Abonnementsservice Postboks 1179 Sentrum 0107 Oslo Telefon 22 31 63 83, Telefaks 22 41 31 05 Fra 1999 og fremover er publikasjonene tilgjengelig på www.norges-bank.no Working papers inneholder forskningsarbeider og utredninger som vanligvis ikke har fått sin endelige form. Hensikten er blant annet at forfatteren kan motta kommentarer fra kolleger og andre interesserte. Synspunkter og konklusjoner i arbeidene står for forfatternes regning. Working papers from Norges Bank can be ordered by e-mail: [email protected] or from Norges Bank, Subscription service P.O.Box. 1179 Sentrum N-0107Oslo, Norway. Tel. +47 22 31 63 83, Fax. +47 22 41 31 05 Working papers from 1999 onwards are available on www.norges-bank.no Norges Bank’s working papers present research projects and reports (not usually in their final form) and are intended inter alia to enable the author to benefit from the comments of colleagues and other interested parties. Views and conclusions expressed in working papers are the responsibility of the authors alone. ISSN 0801-2504 (printed), 1502-8143 (online) ISBN 978-82-7553-450-5 (printed) 978-82-7553-451-2 (online) The macrodynamics of operating income in the Norwegian banking sector Henrik Andersen*, Sigbjørn Atle Berg* and Eilev S. Jansen** August 12, 2008 (Revised March 5, 2010) Abstract The banking literature contains only a handful of studies of how bank revenues vary over the business cycle, and nearly all of these studies look exclusively on the net interest margin. The general conclusion has been that the margin tends to increase during recessions and decrease during booms. In this paper we study the effect of macroeconomic variables on the operating income in the Norwegian banking sector. We contribute to the existing literature by looking at how net interest income as well as fee income varies over the cycle in an error correction framework, which allows us to identify both short term and long term relationships. Our paper also differs from most previous studies by taking into account the volume effect as well as the price effect of the business cycle. * Financial Markets Department, Norges Bank ** Statistics Norway JEL Codes: G21, E32 Keywords: Bank operating income, Business cycles 1 1. Introduction Bank revenues have a time variation pattern over the business cycle. Since revenues are a major determinant of bank capital and lending capacity, the time variation may have an impact on the real economy and may potentially amplify the business cycle. It will therefore be useful to understand how bank revenues vary over time, and in particular what are the relationships with key macroeconomic variables. There are relatively few studies in the banking literature on this question, and the existing research is almost exclusively focused on the net interest margin. The present paper sets out to investigate the relationship between bank revenues and the business cycle more deeply in the context of the Norwegian banking industry. The paper contributes to the existing literature by including volume as well as price effects, and by looking at fee income in addition to net interest income. We are looking at both the short term and the long term effects of the business cycle. These two income variables account for about 85 per cent of total operating income in Norwegian banking. Net interest income was the dominant component with 67 per cent of operating income in the Norwegian banking sector in 2007. Remaining operating income consists of fee income and trading income. Fee income is the more important of these two components and made for 19 per cent of total bank operating income in 2007. The trading income component is very volatile and does not appear to have a stable relationship with macro variables. Section 2 below provides a brief overview of the relevant banking literature. Section 3 sketches the theoretical framework for the analysis. Section 4 introduces the data used, whereas section 5 presents the estimation procedure. The results are presented and discussed in sections 6 and 7. Section 8 concludes the paper. 2 2. Existing literature Most of the existing research on bank operating income focuses on the impact of competition in the deposit and loan markets. Traditionally this research has been based on what is known as the structure-conduct-performance paradigm. The literature has mainly studied how market concentration allows banks the market power to set interest rates on loans and deposits, using other relevant factors as controls. In some more recent research market structure has been treated as endogenous. This literature on bank competition has recently been surveyed by Degryse, Kim and Ongena (2009). It is of limited relevance for the present paper, where we focus on the dynamics of bank operating income and in particular on the link between bank income and macro variables over the business cycle. A more relevant strand of literature seeks to identify the determinants of the net interest margin, i.e. the difference between the average loan rate and the average funding rate of banks, in different markets. This margin is generally approximated as net interest income relative to total assets. Most studies build on the model framework of Ho and Saunders (1981), and explain the net interest margin by structural characteristics of national banking industries. Recent examples include papers by Demirgüç-Kunt and Huizinga (1999), Saunders and Schumacher (2000), Demirgüç-Kunt, Laeven and Levine (2004), Maudos and de Guevara (2004), Claeys and van der Vennet (2005), Schwaiger and Liebig (2008) and Hawtrey and Liang (2008). The studies typically find that the net interest margin depends positively on market concentration, the average ratio of equity to assets, the volatility of interest rates and the level of non-interest operating costs. An early analysis of the dynamics of the net interest margin is found in McShane and Sharpe (1985), but they did not include macroeconomic variables in their model specification. More relevant for our purpose are a few papers that explicitly investigate the cyclical behaviour of the net interest margin. This literature builds on the well documented fact that 3 the price mark-up in most non-financial industries exhibit a counter-cyclical pattern, with higher mark-ups in recessions and lower mark-ups in booms. Dueker and Thornton (1997) and Aliaga-Diaz and Olivero (2005) for the US, da Silva, Oreiro, de Paula and Sobreira (2007) for Brazil, Turgutlu (2010) for Turkey and Mandelman (2006) for a panel of more than 100 countries find the same cyclical pattern for banks’ net interest margin: Margins tend to be higher during recessions and lower during booms. There are a number of possible explanations for this counter-cyclicality of net interest margins: Banks may have a preference for smoothing total income and thus compensate for lower volumes by charging higher margins during recessions. Furthermore, credit risk and adverse selection may be more severe problems for banks during recessions and thus require higher mark-ups. Also, the loan markets may be less contestable during recessions, meaning that incumbents who resort to limit pricing may maintain higher margins without encouraging potential entrants. All of these explanations rely on banks having some market power, and that market power may itself be stronger during recessions. Aliaga-Diaz and Olivero (2007) present a theoretical model of counter-cyclical margins, where the market power of banks is derived from customer switching costs and asymmetrical information among lenders. They note that the counter-cyclical behaviour of margins will act as a financial accelerator, amplifying the effects of any shocks on the macro economy. A fourth stream of papers tries to identify the interest rate risk faced by banks by using gap analysis. This involves measuring the duration (repricing) mismatch between assets and liabilities on the bank balance sheet, and has a long tradition in bank analysis; see e.g. Wright and Houpt (1996). Bank assets and liabilities are sorted into duration buckets, and each item is assumed to be repriced with changes in market interest rates and with the appropriate lag. The implicit assumption is that this duration mismatch between assets and liabilities captures the substance of interest rate risk. Recent experience has taught us that time variation in credit or 4 liquidity premiums may also be significant risk factors. One further problem is that accurate data on repricing periods are almost never available to the researcher: Securities must be allocated to a limited number of duration buckets. Furthermore, information on off-balance sheet hedges, such as interest rate swaps, is rarely available. Gap analysis performed without detailed inside information on bank assets and liabilities may therefore tend to overestimate the sensitivity of bank income to interest rate changes. A recent example of a gap analysis approach is found in Drehmann, Sorensen and Stringa (2010), who proposes a bank model for the stress testing exercises at the Bank of England. Their motivation for choosing this approach is that it facilitates a very transparent exposition of banks’ dependence on market interest rates. Their framework also allows for the introduction of shocks on liquidity and credit premiums. Their model parameters are calibrated rather than estimated. Other papers have tried to identify banks’ exposure to interest rate risk by looking at the time series relationship between bank profitability and interest rates. Early examples are papers by Flannery (1981, 1983), who found that his samples of US banks were not much affected by interest rate changes. Flannery measured profitability from accounting data. A more recent paper also using accounting data is Maes (2004), who looked at Belgian banks without finding much impact of interest rate changes. A closely related parallel literature looks at stock market prices as indicators of profitability. A significantly negative relationship is generally found between unanticipated interest rate changes and financial stock returns. A recent example is Fraser, Madura and Weigand (2002) who also explored how the sensitivity of individual banks to interest rate changes depends on a number of bank characteristics. They found that banks with little equity capital, little non-interest income, high loan volumes and high dependence on demand 5 deposits are more sensitive. Surveys of the literature on financial stock prices and interest rates are found in Staikouras (2003, 2006). A few papers examine the diversification benefits of introducing more non-interest income activities as a supplement to the net interest income from credit intermediation. Recent papers include Smith, Staikouras and Wood (2003), Stiroh (2004), Chiorazzo, Milani and Salvini (2008), Lepetit, Nys, Rous and Tarazi (2008) and Calmès and Liu (2009). Most of these papers find that interest and non-interest income correlate positively both in time series and in cross section dimensions, and that there is consequently no diversification benefit. In the present paper our concern is to model the dynamics of net interest and fee income for the banking sector. We choose to look at the total net interest and fee income components rather than the margins and volumes separately. One reason is that the income variables can be precisely measured from accounting data, whereas the margins can only be approximated. The common practice of measuring the net interest margin as the ratio between net interest income and total assets may be misleading because the composition of total assets may change over time and because total assets always refer to specific points in time whereas income is a stream variable. As discussed below there may also be an endogeneity problem. As far as we are aware the only published paper focussing on the dynamics of bank income components rather than interest margins is by Albertazzi and Gambacorta (2009). They used panel data for ten countries over the period 1981-2003, and found that GDP have a positive effect on both net interest and non-interest income. This is in contrast to the short term negative impact of GDP growth generally found on the net interest margin, see above. This may be explained both by the different time horizons and by the volume effects of GDP growth. 6 There is no guidance in existing literature on the relevant determinants for the fee income. We shall assume that the fee income equation involves the same variables as the net interest income equation, except for the housing wealth variable that is presumably less relevant. The estimation procedure is also the same. ΔlnFEE = α0 + α1*lnFEE-1 + α2*lnGDP -1+ α3*R3M-1+ α4*R5Y-1 + ∑β1k*ΔlnRNET-k + ∑β2k*ΔlnFEE-k + ∑β3k*ΔlnGDP-k + ∑β4k*ΔR3M-k+ ∑β5k*ΔR5Y-k + β6*ΔFORB-1 + ∑β7k*Qk + η (2) For notational simplicity we have used the same parameter symbols in equations (1) and (2), but they will naturally not be assumed equal in the estimation procedure. 6. Co-integrating relationships We have analysed several information sets for Norwegian banks from the period 19902007, when we have data for both income variables. The results we report below are based on a VAR in the two endogenous variables net interest income and fee income, where we consider GDP, housing wealth, and the short and long interest rate variables R3M and R5Y as exogenous variables that are candidates to enter the long run relationships for the bank income variables. We have looked at different lag lengths for the VAR (1, 2 and 4), and we find that even a VAR of the first order gives a satisfactory description of the data1, when we also include seasonals and the change in the market share of foreign branches (ΔFORB ) as unrestricted variables in the VAR. Univariate unit root tests suggest that all variables are I(1) and hence we can carry out a Johansen test for the co-integration rank, i.e. to determine the number of 1 The system tests are: Vector AR(1-5) test: F(20,94) = 0.97 [0.51], Vector Normality test: χ2(4) = 3.06 [0.55], Vector Hetero test: F(36,127) = 0.65 [0.93], Vector Hetero-X test: F(81,84) = 0.90 [0.67], where p-values are given in square brackets. 13 co-integrating vectors (Johansen, 1988). The trace tests of co-integration rank give the results shown in table 1. Tests based on a VAR(1) in lnRNET, lnFEE conditional on lnGDP, lnHW, R3M, R5Y (restricted) and ΔFORB (unrestricted). Data is from 1990Q2 – 2007Q2. _______________________________________________________________________ Hypothesis Trace statistics Approximate p-value* Eigenvalue r = 0 56.278 p < 0.05 0.42 r ≤ 1 18.667 p ≈ 0.07 0.24 *) Critical values are intraand extrapolated from Table IV in MacKinnon et al (1999) and Table 6(c) in Pesaran et al (2000). Table 1: Tests of co-integration rank Since we are considering a partial VAR, i.e. conditioning on a set of exogenous variables, the standard critical values for the Johansen test, as found e.g. in Osterwald-Lenum (1992), do not apply. Instead we refer to critical values from MacKinnon et al (1999) and Pesaran et al (2000); confer also Harbo et al (1998). The second co-integration vector is somewhat doubtful according to the p-value even though the eigenvalue is estimated as high as 0.24. Adopting a higher order VAR yields a similar picture. We have, however, imposed on the VAR two co-integration vectors which in our case can be written as (normalising on the two endogenous variables): lnRNET + β10lnFEE + β11lnGDP + β12lnHW + β13*R3M + β14*R5Y (3) β20lnRNET + lnFEE + β21lnGDP + β22lnHW + β23*R3M + β24*R5Y (4) We let the matrix α, with elements [αij ], i,j=1,2, denote the loadings of the vectors in the cointegrated VAR. Next we proceed to impose identifying and over-identifying restrictions on the two vectors and the loadings. The results in table 2 show that if we identify the equation by assuming that lnRNET and lnFEE do not influence each other in the long run (that is β10= β20=0), we find that net interest income (RNET) in the long run depends positively on GDP and the short interest rate, whereas the fee income (FEE) is determined by GDP and the difference between the long and short interest rates. These qualitative conclusions are robust 14 with respect alternative identifying restrictions and we also find the same end result if we redo the analysis with a higher order VAR (of order 2 or 4). Recursive graphs of the estimated long run coefficients from Panel 5 in table 2 areshown in figure 1 below. 1995 2000 2005 -1.25 -1.00 -0.75 -0.50 -0.25 β11 × +/- 2 st.errors 1995 2000 2005 -5.0 -2.5 0.0 β14 × +/- 2 st.errors 1995 2000 2005 -2.0 -1.5 -1.0 β21 × +/-2 st.errors 1995 2000 2005 5 10 15 20 β23 (= – β24) × +/-2 st.errors Figure 1: Recursive graphs the estimated long run coefficients from Panel 5 of Table 2. 15 Tests of over-identifying restrictions on a VAR(1) in lnRNET, lnFEE conditional on lnGDP, lnHW, R3M, R5Y (restricted) and ΔFORB (unrestricted). Identifying restriction: β10= β20=0. Data is from 1990Q2 – 2007Q2. Standard errors are given in parentheses, and p-values in square brackets. a) Panel 1 = Weak exogeneity of lnRNET and lnFEE across equations: α12 = α21 = 0. lnRNET – 0.79 lnGDP – 0.02 lnHW + 2.07 R5Y – 3.41 R3M (0.35) (0.11) (1.58) (0.99) lnFEE – 1.94 lnGDP + 0.13 lnHW– 4.70 R5Y + 5.21 R3M (0.55) (0.17) (2.52) (1.58) α11 = 0.54 (0.09), α22 = 0.33 (0.08) χ2(2) = 5.65 [0.06] b) Panel 2 = a) and only the interest rate difference (SLOPE) matters in A2 β23= – β24 lnRNET – 0.80 lnGDP – 0.02 lnHW + 2.01 R5Y – 3.39 R3M (0.34) (0.11) (1.55) (0.98) lnFEE – 2.03 lnGDP + 0.15 lnHW – 5.35 R5Y + 5.35 R3M (0.50) (0.17) (1.57) ( - ) α11 = 0.54 (0.09), α22 = 0.34 (0.08) χ2(3) = 5.74 [0.12] χ2(1) = 0.10 [0.75] c) Panel 3 = b) and no effect of lnHW in A1 β12= 0 lnRNET – 0.87 lnGDP + 1.97 R5Y – 3.41 R3M (0.11) (1.56) (0.99) lnFEE – 2.05 lnGDP + 0.15 lnHW – 5.34 R5Y + 5.34 R3M (0.49) (0.17) (1.57) ( - ) α11 = 0.53 (0.09), α22 = 0.33 (0.08) χ2(4) = 5.79 [0.22] χ2(1) = 0.04 [0.84] d) Panel 4 = c) and no effect of R5Y in A1 β14= 0 lnRNET – 0.95 lnGDP– 2.50 R3M (0.11) (0.68) lnFEE – 2.01 lnGDP + 0.14 lnHW – 5.54 R5Y + 5.54 R3M (0.48) (0.17) (1.53) ( - ) α11 = 0.50 (0.09), α22 = 0.33 (0.08) χ2(5) = 7.49 [0.19] χ2(1) = 1.70 [0.19] e) Panel 5 = d) and no effect of lnHW in A2 β22= 0 lnRNET – 0.95 lnGDP– 2.52 R3M (0.11) (0.69) lnFEE – 1.62 lnGDP – 5.59 R5Y + 5.59 R3M (0.11) (0.17) ( - ) α11 = 0.50 (0.09), α22 = 0.33 (0.08) χ2(6) = 8.25 [0.22] χ2(1) = 0.76 [0.38] Table 2: Tests of restrictions on the long term relationships 16 7. The short term dynamics. The co-integration tests above indicate the presence of two co-integrating vectors; with one long term relationship between real net interest income, real GDP and the 3-month interest rate; and the second long term relationship between real fee income, real GDP and the slope of the yield curve (defined as SLOPE = R5Y - R3M). The lagged values of these level variables are then retained in the equations (1-2) from section 5, together with the change in the market share of foreign branches and the quarterly dummies. Notice, however, that the coefficients are being re-estimated and will in general deviate from those reported in table 2. The difference variables in equations 1-2 enter with a maximum of four quarter lags. As explained in section 5, we use the automated search procedure of the PC-Give statistical package to eliminate insignificant terms in a search process until we arrive at a preferred parsimonious representation. The coefficient estimates for both relationships are listed in tables 3-4 below. Dependent variable: ΔlnRNET Coefficient t-value lnRNET-1 -0.622 -5.98 lnGDP -1 0.593 5.61 R3M-1 1.835 4.37 ΔFORB -0.018 -2.61 Δ R3M-1 -1.332 -2.27 Constant -2.129 -2.92 Q2 0.033 2.03 Q3 0.064 3.60 Q4 0.049 2.71 R2 = 0.497 1990Q2 – 2007Q2 Sigma = 0.045 RSS = 0.116 F(8,56) = 6.39 [p-value= 0.00] AR 1-5 test F(2,54) = 0.72 [0.49] Normality test Chi^2(2) = 2.16 [0.34] Hetero test F(2,52) = 0.05 [0.95] Table 3: Estimated relationship for the determination of real net interest income 17 In table 3 the estimated long term coefficients are very similar to those reported in table 2. Net interest income will according to these estimates tend to grow nearly in step with GDP, with a ratio of 0.95 (0.593/0.622). Divergence from this long term relationship has an implied quarterly adjustment coefficient of 0.62, which is relatively high and close to the value we found in the co-integration analysis. The volume effect thus appears to outweigh the interest margin effect identified in previous studies. The level of interest rates as represented by the three month money market rate also has a positive impact on net interest income. This is in line with previous research (Dueker and Thornton, 1997; Aliga-Diaz and Olivero, 2005) and can be interpreted as saying that a higher (real) interest rate makes it easier for banks to obtain a higher interest rate margin. Alternatively a high real interest rate could be interpreted as a second business cycle indicator, with the central bank pushing up the short rate when the economy’s production capacity is fully utilised. The short term dynamics indicate that changes in the degree of competition as represented by the market share of foreign branches has the expected negative impact. Consistent with the findings of Mandelman (2006), we have assumed no long term effect of foreign entry. The other change variable in the final specification is the change in the money market interest rate, which has a negative impact. This could reflect the regulation that banks can only raise loan rates with six weeks notice. But it could also be seen as a correction on the long term effect; the net effect of a higher money market rate is slightly positive in the first quarter and then gradually builds up to the long term effect. As explained in section 4, net interest income should ideally have been corrected for the effect of interest rate swaps. That information is not available, but swap revenue are part the total derivatives revenue. We have re-estimated equation (1) with all derivatives revenues added to net interest income. The coefficient estimates and the test statistics remained 18 practically unchanged. This indicates to us that correcting for interest rate swaps would not be important. Dependent variable: ΔlnFEE Coefficient t-value lnFEE-1 -0.371 -3.98 lnGDP -1 0.597 3.69 SLOPE-1 1.507 2.65 ΔFORB -0.003 -0.48 ΔlnGDP 0.944 3.27 ΔlnFEE-4 0.234 2.08 Constant -4.817 -3.47 Q2 0.024 1.46 Q3 0.007 0.37 Q4 -0.001 -0.03 R2 = 0.502 1990Q2 – 2007Q2 Sigma = 0.046 RSS = 0.115 F(9,55) = 6.15 [p-value = 0.00] AR 1-2 test F(2,53) = 0.10 [0.91] Normality test Chi^2(2) = 2.83 [0.24] Hetero test F(4,50) = 0.23 [0.89] Table 4: Estimated relationship for the determination of real fee income In table 4 the estimated long term coefficients are again similar to those reported in table 2. Fee income will according to these estimates tend to grow significantly faster than GDP, with a ratio of 1.6 (0.597/0.371). Divergence from this long term relationship has an implied quarterly adjustment coefficient of 0.37, which is slightly higher than the value found in the co-integration analysis. The slope of the yield curve as represented by the difference between 5 year and 3 month interest rates also appears to have a positive impact on fee income. Steeply rising yield curves may be associated with situations with slow growth and where the short term interest rate has been reduced to stimulate economic activity. This result could then be interpreted as a correction of the very strong effect found from GDP: Periods with GDP close to capacity have during our observation period to a large extent been characterised by negatively sloping yield curves. On the other hand a steep yield curve is also known to predict higher GDP growth in the next year (see e.g. Chen, 1991). These 19 expectations may stimulate the demand for banking services. This is also consistent with Schuerman and Stiroh (2006), who finds that a steeper yield curve has a positive impact on bank stock returns. The short term dynamics depends on the current GDP growth and on the lagged change in fee income itself. The positive impact of lagged fee income indicates a significant momentum effect on that income component. The positive impact of GDP growth is readily explained as a demand side factor in the market for bank service volumes. This is in line with the conclusions of Calmès and Liu (2009), but in contrast to the findings of Albertazzi and Gambacorta (2009). The difference may be explained by the fact Albertazzi and Gambacorta looked at non-interest income in total, whereas this paper follows Calmès and Liu by looking specifically at fee income. The change in the market share of foreign branches is retained because it has been used as a control variable in the co-integration analysis, but it does not appear as statistically significant in this final equation. Foreign entries do not seem to have an immediate effect on other bank services than lending. The total GDP effect may seem too high, however, with fee income increasing much faster than GDP even when taking the modifying yield curve effect into account. A possible explanation could be that there is a spill-over effect in the data from the gradual build-up of transaction volumes that started in the early 1990’s. In this period high prices on paper-based payment transactions made for a substantial shift of transactions into cheaper electronic payment systems, which again generated a rapid increase in the number of transactions handled by banks. This structural change in the payment system was gradual and it would therefore be hard to represent it properly in the model specification. We have checked that the upward trending time series of prices on electronic and manual transactions are not statistically significant explanatory variables. Consistent data on transaction volumes are only available from 1997. 20 The standard test statistics generated by PC-Give on the residuals indicate that the relationships in tables 3 and 4 are both well behaved. The plot of recursive coefficient estimates in figures 2 and 3 indicates reasonable stability of the coefficients. Figure 4 shows the actual and fitted values of quarter-on-quarter growth in net interest income, and the corresponding residuals. We notice that the fit is relatively poor in the mid 1990’s, but is becoming much better after that. Correspondingly, figure 5 shows the actual and fitted values of quarter-on-quarter growth in fee income. Again, the fit appears to be relatively good. Figure 2: Recursive parameter estimates for the net interest income relationship. 21 Figure 3: Recursive parameter estimates for the fee income relationship Figure 4: Actual and fitted quarterly changes in real net interest income in the upper panel and residuals in the lower panel. 22 Data appendix 6,5 7 7,5 8 8,5 9 1988Q1 1991Q1 1994Q1 1997Q1 2000Q1 2003Q1 2006Q1 ln(RNET) ln(FEE) Figure A1: Logs of net interest and fee income. Base variables are in millions of NOK. 15 15,5 16 16,5 17 17,5 18 1988Q1 1991Q1 1994Q1 1997Q1 2000Q1 2003Q1 2006Q1 11,8 12 12,2 12,4 12,6 12,8 13 ln(HW) ln(GDP) Figure A2: Logs of real housing wealth (left axis) and real GDP (right axis). Base variables are in millions of NOK. 0 0,02 0,04 0,06 0,08 0,1 0,12 1988Q1 1991Q1 1994Q1 1997Q1 2000Q1 2003Q1 2006Q1 -0,06 -0,04 -0,02 0 0,02 0,04 0,06 3 month 5 years Slope Figure A3: Three month and five year real interest rates (left axis) and the yield difference or slope (right axis). 29 0 3 6 9 12 15 18 1988Q1 1991Q1 1994Q1 1997Q1 2000Q1 2003Q1 2006Q1 Market share of foreign branches Figure A4: The market share (per cent of total assets) of branches owned by foreign banks. 30