Financ3

Journal of Banking & Finance 59 (2015) 486–504

Contents lists available at ScienceDirect

Journal of Banking & Finance

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / j b f

The management of interest rate risk during the crisis: Evidence from Italian banks

http://dx.doi.org/10.1016/j.jbankfin.2015.04.031 0378-4266/� 2015 Elsevier B.V. All rights reserved.

⇑ Corresponding author. 1 We thank Eugenio Gaiotti, Paolo Del Giovane, Stefano Neri, Martina Bignami,

Giovanni Pepe, Giuseppe Della Corte, Piergiorgio Alessandri and two anonymous referees for their useful remarks and suggestions. We also benefited from the comments by the participants at the lunch seminar at the Bank of Italy. Ginette Eramo provided invaluable help in constructing the dataset used in the empirical analysis.

2 See, for example, European Central Bank (2011) and Cecioni et al. (2011) for a discussion of the effectiveness of unconventional monetary policy measures adopted by the ECB and the U.S. Federal Reserve.

3 The slope of the yield curve is measured by the difference between the 10-year government bond yield and 3-month Euribor rate.

4 The most dramatic example in U.S. history is the financial turmoil of th and loan industry between 1980 and 1988, which saw more than 1000 ass fail. The crisis was triggered by a change in monetary policy, when the Reserve began targeting monetary aggregates and let the federal funds ra considerably. The adverse impact on the US banking system originated preci the asset-liability mismatch of savings and loan associations.

Lucia Esposito, Andrea Nobili ⇑, Tiziano Ropele Bank of Italy, Economic Research and International Relations, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Received 11 October 2013 Accepted 9 April 2015 Available online 9 July 2015

JEL classification: E43 G21

Keywords: Interest rate risk Banks Financial crisis

We use a unique dataset to analyze how Italian banking groups managed their exposure to interest rate risk during the recent financial crisis. First of all, we document that on average the interest rate risk expo- sure – measured by duration gap approach – has been limited and well below the alert level enforced by regulators. Second, our econometric results indicate a relation of substitutability between banks’ on-balance-sheet interest rate risk and their use of interest rate derivatives suggesting that banks used these two instruments to curb their overall interest rate risk exposure in case of an increase in interest rates. Furthermore, we also find robust evidence of a negative correlation between banks’ interest rate risk and liquidity risk.

� 2015 Elsevier B.V. All rights reserved.

1. Introduction1

Following the collapse of Lehman Brothers in September 2008 and subsequently with the eruption of the sovereign debt crisis in the euro area, central banks across the world have reacted promptly with bold measures. In the euro area, the European Central Bank (ECB) reduced the official policy rates to historically low levels and introduced unconventional measures to restore the monetary policy transmission mechanism.2 Throughout this period, heightened financial markets volatility, persistent macroeco- nomic uncertainty and investors’ increased risk aversion had strong and heterogeneous impacts on the yield curve across euro-area countries. Fig. 1 shows the ECB rate on main refinancing operations and the slope of the yield curve for Italy from September 2008 to June 2012.3 Quite remarkably, the slope increased rapidly from September 2008 to mid-2009, remained virtually unchanged until

the end of 2010 and then rose again – though more erratically – with the intensification of the euro-area sovereign debt crisis.

Financial intermediaries, mostly because of their maturity transformation role, are exposed to the interest rate risk, i.e. the potential negative impact on their economic value or profitability that arises from unexpected changes in interest rates. Especially when interest rates are volatile and hard to predict, an inadequate management of interest rate risk can erode banks’ capital and lead to macroeconomic and financial instability.4 It is thus crucial also for policymakers to carefully assess the banking sector’s exposure to interest rate risk when designing and implementing policies that might have intended or unintended effects on market interest rates.

For instance one can consider the situation when the ECB will start pulling back on its extraordinary monetary stimuli – either by a rise in interest rates and/or an exit strategy from unconven- tional measures – to return to a less accommodative monetary stance. In this case unintended effects could emerge due to the fact that a rise in the official interest rates could harm euro-area countries whose economic recovery is still underway and in turn renew sovereign debt tensions. Likewise, the termination of

e savings ociations

Federal te move

sely from

-1

0

1

2

3

4

5

6

7

2008H1 2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1

Monetary policy rate

Slope of the yield curve

Slope of the yield curve (net of sovereign risk)

Fig. 1. Slope of the yield curve and the monetary policy rate. End-of-month values in percentage points. The slope of the yield curve is calculated as the difference between the yield on 10-year Italian government bonds and the 3-month money market rate. The slope of the yield curve (net of sovereign risk) is calculated as the difference between the 10-year euro swap rate and the 3-month money market rate. The policy rate is the ECB rate on main refinancing operations. Source: Bloomberg and ECB.

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 487

unconventional measures could be associated with an abrupt steepening of the yield curve as expectations of low short-term interest rates are reversed. Furthermore, from a financial stability perspective banks’ exposure to interest rate risk as well as to other financial risks (e.g. liquidity and credit risk) need to be jointly eval- uated in order to assess the direct effect on capital and the indirect effect on it via other risks of an unexpected change in interest rate. In light of these considerations, it is particularly relevant to know to what extent banks rely on interest rate derivatives and thus evaluate their resilience to adjust their interest rate risk exposure to sudden shifts in interest rates.

The main goal of this study is to examine the exposure to inter- est rate risk – measured by the duration gap approach as proposed by the Basel Committee on Banking Supervision (2004, 2006)5 – for a representative sample of 68 Italian intermediaries using high-quality and confidential information from the second half of 2008 through the first half of 2012. More specifically, we are inter- ested in: (1) quantifying the interest rate risk and examining how it changed over time and across banks; (2) assessing how financial intermediaries managed their interest rate risk exposure either mod- ifying the duration mismatch between on-balance-sheet assets and liabilities (on-balance-sheet restructuring) or relying on interest rate derivatives (off-balance-sheet adjustment); (3) evaluating the interaction between interest rate risk and other financial risks, such as liquidity risk and credit risk.

From a general theoretical perspective there is no consensus regarding the optimal degree of exposure to interest rate risk. On the one hand, Diamond (1984) advocates that banks should not assume any interest rate risk and thus they should fully hedge it. By doing so, furthermore, banks would lower the cost of monitor- ing and gain in terms of intermediation efficiency. On the other hand, when some financial risks are not tradable in the capital markets (Froot and Stein, 1998) or when financial risks (e.g.

5 As more thoroughly discussed in Section 2, according to the duration gap approach the interest rate risk is measured as the potential effect of a 200 basis points parallel shift of the yield curve on the bank’s economic value (present value of future cash flows) as a percentage of regulatory capital.

interest rate risk, credit risk and liquidity risk) are interrelated then a non-zero exposure to interest rate risk might be optimal (Saunders et al., 1990; Brewer et al., 1996). Banks might also be willing to assume interest rate risk to increase their returns by exploiting shifts in the yield curve (Deshmukh et al., 1983; Sartoris, 1993; Memmel, 2011).

Interestingly, the diffusion of interest rate derivatives (e.g. interest rate swap, financial futures, options, etc.) has largely facil- itated banks’ active management of interest rate risk and increased the potential for banks to achieve their desired levels of exposure. In fact, financial intermediaries are usually able to optimally adjust their on-balance-sheet interest rate risk exposure only partially because of various constraints, such as customers’ preferences, competition in the banking industry, difficulties in accessing wholesale funding markets, etc. Regarding the purpose of using interest rate derivatives the existing empirical studies have found conflicting results. Gorton and Rosen (1995), Brewer et al. (1996), Schrand (1997), Purnanandam (2007), Zhao and Moser (2009) find that banks have relied on interest rate derivatives mainly to hedge against the on-balance-sheet exposure to interest rate risk. Sinkey and Carter (1994), Esty et al. (1994), Gunther and Siems (1995) and Hirtle (1997) provide instead evidence consistent with the view that financial intermediaries have used interest rate derivatives as a tool to enhance interest rate risk exposure. Simons (1995) and Angbazo (1997) find no significant relationship between derivatives usage and on-balance-sheet exposure to interest rate risk.6 More recently, Begenau et al. (2013) proposed a strategy to estimate the exposure due to interest rate derivatives from regula- tory data on notional and fair values together with the history of interest rates and document that U.S. banks’ net interest rate deriva- tive position enhances their on-balance-sheet exposure to interest rate risk.

As for the interaction among risks, during the financial crisis the supervisory authorities were particularly concerned about finan- cial stability conditions and, in this regard, about risk management

6 Reichert and Shyu (2003) and Choi and Elyasiani (1997) investigate how banks’ use of derivative instruments is associated with interest rate and exchange risks.

8 Another important source of interest rate risk, not discussed in this paper, is the basis risk (for a discussion see English, 2002), which occurs when the adjustment of the rates earned and paid on different instruments is imperfectly correlated with otherwise similar repricing characteristics (for example, a three-month Treasury bill versus a three-month EURIBOR). When interest rates change, these differences can change the cash flows and earnings spread between assets, liabilities, and off-balance-sheet instruments of similar maturities or repricing frequencies. During

488 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

decisions adopted by banks. While the best practices of risk man- agement should be effectively based on an integrated approach (see Bank of Italy, 2012; European Banking Authority, 2011), moral hazard considerations suggest that banks, especially large ones that are considered ‘‘too-big-to-fail’’, may have the incentive to take on a higher level of risk via multiple sources, which would imply a positive correlation among risks of different nature. Drehmann et al. (2008) and Alessandri and Drehmann (2010) developed a model where credit and interest rate risk in the bank- ing book are analyzed jointly and showed that ignoring interac- tions among these risks leads to an overall risk overstatement.

We contribute to the literature on banks’ exposure to interest rate risk in a number of aspects. First, to the best of our knowledge, we are the first to empirically examine the management strategy of interest rate risk exposure for Italian banks.7 On this regard, the Italian banking sector is an interesting case to look at because a traditional business model typically characterizes Italian financial intermediaries and, thus, banks’ exposure to the interest rate risk represents an important factor affecting both their profitability and financial conditions. Second, unlike most of the existing empirical studies on the interest rate risk that rely on either a maturity gap approach or an estimated indicator derived à la Flannery and James (1984), we use as a measure of interest rate risk based on the duration gap methodology as suggested by the Basel Committee on Banking Supervision (2004, 2006). Third, our high-quality dataset, entirely based on Supervisory reports, allows us to examine the interest rate risk management strategies using the same duration gap approach to evaluate the on-balance-sheet and the off-balance-sheet interest rate risk exposure. Previous stud- ies have often used the notional value of interest rate derivatives to assess banks’ hedging strategies, which may be questionable since the notional value disregards the maturity or duration of these instruments.

Our main results can be summarized as follows. First, from the second half of 2008 through the first half of 2012, the Italian bank- ing system as a whole had a limited exposure to interest rate risk, well below the alert level enforced by regulators. In the event of a parallel upward (downward) shift of the yield curve by 200 basis points the economic value of the Italian banking industry would have risen (declined) by about 3.1% of regulatory capital.

Second, our econometric results suggest the existence of a rela- tion of substitutability between banks’ on-balance-sheet exposure and their use of interest rate derivatives. In other terms, we find that banks have manoeuvred their on-balance-sheet interest rate risk and their off-balance-sheet exposure to partially offset each other rather than to pursue an enhancing strategy and boost the potential gains that might have occurred had interest rates increased.

Third, we find that one third of the banks, namely the smaller ones and those characterized by a traditional business activity, followed an integrated risk-management approach aiming at offsetting the interest rate risk and the credit risk. The majority of banks, which comprises the largest banking groups, tended to enhance the gains from an increase in interest rates also in the face of a widening of the funding gap.

The rest of the paper is organized as follows. Section 2 discusses the sources of interest rate risk and the measurement of banks’ exposure to this risk. Section 3 presents the data used in the empir- ical investigation while Section 4 presents some descriptive statis- tics. In Section 5 we describe the methodology used in the analysis while in Section 6 we discuss the estimation results. In Section 7

7 Fiori and Iannotti (2006) examine Italian banks’ interest rate risk exposure but their contribution is more methodological. In particular, they are interested in developing a framework for measuring banks’ interest rate risk exposure by taking into account the fact that financial data exhibit skewness and fat tails.

we test the robustness of our main results. In Section 8 we provide some concluding remarks and discuss the main policy implications.

2. Sources and measurement of banks’ exposure to interest rate risk

The interest rate risk can be defined as the potential impact on a bank’s economic value or profitability from a change in interest rates. The sources of interest rate risk can be various (see, e.g., English, 2002 or Fraser et al., 2002).

The most important source of interest rate risk is the repricing risk, which for banks naturally arises because of the mismatch in the time to maturity (for fixed-rate instruments) or time to repric- ing (for floating-rate instruments) between balance sheet assets and liabilities. For example, a bank that finances fixed-rate long-term loans with short-term deposits will suffer a decline in economic value if interest rates increase as the cost of funds will rise while the earnings from loans remain fixed. The bank will instead gain if interest rates decrease.

Another relevant source of interest rate risk is the yield curve risk that occurs when the term structure of interest rates steepens, flattens or becomes negatively sloped. In this case non-parallel shifts in the yield curve can accentuate the interest rate risk by amplifying the effect of maturity mismatches. To see this, suppose that a bank goes long on a 10-year government bond and short on the 5-year government bond. Next, assume the yield curve steep- ens because of a larger rise in the long-term rate than in the short-term rate. Under these circumstances, the bank’s economic value will diminish as the value of the 10-year bond declines more than the value of the 5-year bond.

Finally, the interest rate risk can also originate from the so-called option risk.8 This happens when changes in interest rates lead a bank’s costumers to influence the timing and the magnitude of assets and liabilities cash flows. Typically, the option risk is linked to the presence of explicit and/or implicit optionality clauses embed- ded in some financial products. On the asset side of the balance sheet, prepayment options – most typically in residential mortgage and consumer loans – are the most prevalent embedded option. For instance, an optionality clause may assign the borrower the right (not the obligation) to prepayment without incurring in any penalty. A decline in interest rates could induce borrowers to accelerate pre- payments of loans and this would determine a decline in cash flows, as the bank would reinvest the proceeds at a lower interest rate. On the liability side of the balance sheet, the most prevalent option given to customers is the right of early withdrawal especially for sight deposits. If interest rates increase, the opportunity cost of hold- ing deposits will rise and then depositors will withdraw their funds from the bank.9

Banks’ exposure to interest rate risk can be measured using various methodologies. Roughly speaking, one can identify two broad approaches: the earnings approach and the economic value approach.

the financial crisis one has observed a decoupling between EURIBOR-indexed swap yields and the yield of Italian sovereign and corporate bonds, however, given that our measurement of interest rate risk based on the duration gap approach disregards such a source of risk we decided not to investigate the basis risk.

9 Bank financial products that embed interest rate ‘‘caps’’ or ‘‘floors’’ represent other sources of option risk.

12 Unfortunately, we cannot take an earlier starting date because data from

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 489

The earnings approach seeks to quantify the impact of a change in interest rates ðDiÞ on the bank’s net interest income ðDNIIÞ, i.e. the difference between interest earned on loans and other assets and interest paid on funding and other liabilities. This measurement is based on the so-called maturity gap analysis, which considers the mismatch between interest-sensitive assets ðRSAÞ and liabilities ðRSLÞ usually over a horizon of up to two years. Hence, the effect of Di on DNII is given by DNII ¼ðRSA � RSLÞDi. When RSA > RSL there is a positive maturity gap, implying that if interest rates decrease (or increase) the net interest income diminishes (or rises).10

The economic value approach focuses instead on the potential impact of a change in interest rates on the market value of bank’s capital.11 This approach heavily builds on the concept of Macaulay (1938) duration (DUR), which approximates how sensitive a secu- rity’s price is to a change in interest rates, i.e.

DP P ��

DUR 1 þ i

� Di: ð1Þ

In (1) the term DUR=ð1 þ iÞ is commonly referred to as the mod- ified duration. Similarly, to measure the sensitivity of the bank’s economic value to a change in interest rates one calculates the duration of on-balance-sheet and off-balance-sheet assets and lia- bilities. For this purpose, all balance sheet items are typically ordered in n time bands according to their remaining time to matu- rity or re-pricing schedule. Then, for each time band a net position

is computed as on-balance-sheet assets ðAONÞ minus on-balance-sheet liabilities ðLONÞ plus the net position in off-balance-sheet instruments ðDOFFÞ: Provided a modified duration is assigned to each time band, then the duration gap (GAP) can be written as follows

GAP ¼ Xn j¼1

DURj 1 þ i

AONj � L ON j þ D

OFF j

Z

! ð2Þ

where Z is a normalizing variable, such as total assets or regulatory capital. Note that in (2) the modified durations are calculated assuming that all positions in each time band have the same yield 1 + i. Finally, bank’s exposure to interest rate risk (IRR), i.e. the change in the value of capital (K) as a percentage of Z to a parallel shift in the yield curve by Di can be approximated by

IRR � DK Z ��GAP � Di ð3Þ

From (3) one can note that a bank with a positive (negative) duration gap will lose (gain) from an increase (decrease) in interest rates. A zero gap, instead, implies that the bank has immunized the market value of its equity against interest rate changes.

As one of the goals of our paper is to examine how Italian banking groups managed their exposure to interest rate risk, it is useful to rewrite (3) as the sum of the on-balance-sheet and the off-

balance-sheet duration gaps (respectively, GAPON and GAPOFF ), that is

IRR �� Xn j¼1

DURj 1 þ i

AONj � L ON j

Z

! þ Xn j¼1

DURj 1 þ i

DOFFi Z

!" # Di

��½GAPON þ GAPOFF�Di ð4Þ

Hence, expression (4) clearly distinguishes between the interest rate risk related with the decisions taken by the bank regarding on-balance-sheet restructuring and that stemming from

10 Flannery and James (1984) is an early empirical study on the relevance of banks’ exposure to interest rate risk, measured by 12-month maturity gap, in explaining the interest rate sensitivity of banks’ stock returns.

11 See Armeanu et al. (2008) for an overview of the duration gap analysis.

off-balance-sheet interest rate derivatives. We define banks with a positive on-balance-sheet duration as asset-sensitive and those with a negative on-balance-sheet duration gap as liability-sensitive.

The measure of interest rate risk in expression (4) disregards the yield curve risk as it only considers parallel shifts of the term structure of interest rates, which is furthermore assumed to be flat. We overcome the above limitations by carrying out a number of relevant exercises. Furthermore, the duration gap analysis does only fully take into account the option risk, as effectively it is a sta- tic method that shows a snapshot of the interest rate risk exposure based upon the composition of balance sheet at a given time. Given these limitations, banks – in particular those relatively more exposed to sources of interest rate risk other than repricing risk – measure their exposure also using a variety of more sophisticated techniques. For examples, techniques based on dynamic simula- tions usually allow to make assumptions about future paths of interest rates (such as changes in the shape and slope of the yield curve) and about expected changes in balance sheet compositions (such as prepayment forecasts, product pricing assumptions, etc.).

3. Data

For the purposes of our analysis we use a unique panel dataset of semi-annual observations for 68 Italian financial intermediaries from the second half of 2008 through the first half of 2012.12 More specifically, our panel dataset comprises only Italian banking groups – i.e. listed companies, cooperative banks, subsidiaries and branches of foreign banks operating in Italy – while it excludes individual Italian banks that do not belong to groups (mostly mutual banks). Overall, our panel of intermediaries accounts for nearly 70% of the total assets of the banking system. Data are confidential and come from banks’ supervisory reports to the Bank of Italy; in compliance with the regulatory framework, they are based on consolidated balance sheet items for the banking groups and on unconsolidated balance sheet items for foreign intermediaries.

3.1. Measurement of banks’ exposure to interest rate risk

We measure banks’ interest rate risk using the simplified methodology established by the Bank of Italy (2006), which relies on the duration gap approach and is consistent with the principles stated by the Basel Committee on Banking Supervision (2004, 2006).13 According to the regulatory framework, banks’ exposure to interest rate risk is measured with reference to the on-balance-sheet and off-balance-sheet assets and liabilities in the banking book and is quantified by the potential effect on banks’ economic value of a standardized interest rate shock, defined as a parallel shift of the yield curve by 200 basis points.

The standardized method for interest rate risk requires that all assets, liabilities and off-balance-sheet items be allocated in 14 maturity buckets according to their remaining time to maturity or, in the case of variable rate items, according to their re-pricing schedule (see Table 1). Assets and liabilities that do not have an explicit maturity have a different treatment. The reserve require- ment is classified in the ‘‘up to one month’’ time band, reflecting the frequency of the Eurosystem’s main refinancing operations, the yield of which is used as a benchmark in determining the inter- est rate on the reserve requirement. Bad debts (net of value

supervisory reports display a break between November 2008 and December 2008, owing to a change in data collection methods.

13 Fiori and Iannotti (2006) develop a value-at-risk model for measuring the interest rate risk on both the banking and the trading book for the 18 largest Italian banks. Their results in terms of interest rate risk evaluation are consistent with what predicted by the ‘‘simplified methodology’’.

Table 1 Weighting factors to compute banks’ interest rate risk exposures. As explained in Section 2, in order to calculate the interest rate risk exposures for each time band the offset of assets against liabilities (i.e. the net position) is multiplied by a weighting factor. The factors are based on a hypothetical shift in the term structure of interest rates and a proxy of the modified duration for each time band. As specified by the Basel Committee, the modified duration is calculated assuming that all net positions in each time band have a yield of 5%. In the table we report the weighting factors associated with three hypothetical shifts of the terms structure of interest rates: a parallel shift by + 200 basis points and two non-parallel shifts (either downward-sloping or upward-sloping).

Time band Approximate modified

duration (years) (A)

Parallel shift in yield curve (basis

points) (B)

Weighting factor

(A) � (B)

Non-parallel shift in yield curve (basis

points) (C)

Weighting factor

(A) � (C)

Non-parallel shiftin yield curve (basis

points) (D)

Weighting factor

(A) � (D)

Overnight 0.00 200 0.00 200 0.00 200 0.00 (0, 1 month] 0.04 200 0.08 200 0.08 200 0.08 (1 month, 3 months] 0.16 200 0.32 200 0.32 200 0.32 (3 months, 6 months] 0.36 200 0.72 200 0.72 200 0.72 (6 months, 1 year] 0.71 200 1.42 200 1.42 200 1.42 (1 year, 2 years] 1.38 200 2.76 200 2.76 200 2.76 (2 years, 3 years] 2.25 200 4.50 188 4.22 213 4.78 (3 years, 4 years] 3.07 200 6.14 175 5.37 225 6.91 (4 years, 5 years] 3.85 200 7.70 163 6.26 238 9.14 (5 years, 7 years] 5.08 200 10.16 150 7.62 250 12.70 (7 years, 10 years] 6.63 200 13.26 138 9.12 263 17.40 (10 years, 15 years] 8.92 200 17.84 125 11.15 275 24.53 (15 years, 20 years] 11.21 200 22.42 113 12.61 288 32.23 (20 years, oo] 13.01 200 26.02 100 13.01 300 39.03

Source: Bank of Italy and authors’ calculations.

490 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

adjustments) are classified in the ‘‘5 to 7 years’’ band, in line with the estimated residual life of the loans based on their turnover rate. Cash is classified in the ‘‘demand and revocable’’ time band. The sum of overnight deposits and demand deposits are allocated in the following way: a fraction of 25% (the so-called non-core compo- nent) goes into the ‘‘demand and revocable’’ time band while the remaining fraction of 75% (the so-called core component) goes in the next four time bands (from ‘‘up to one month’’ to ‘‘6 months to 1 year’’) in proportion to the number of months contained in them.14 Derivatives are allocated to the time bands in accordance with the criteria for capital requirements in respect of market risks.

As in Eq. (2), for each time band, assets are offset against liabilities to produce a net position. The net position of every time band is then multiplied by a weighting factor based on a proxy of the modified duration for each time band and on a hypothetical 200 basis points parallel shift of the yield curve. As specified by the Basel Committee, the modified duration is calculated assuming that all positions in each time band have a yield of 5%. Finally, the weighted exposures of the different bands are summed and then divided by supervisory capital to determine the interest rate risk indicator. Although banks are not subject to a specific capital requirement for interest rate risk exposure the regulatory provisions recommend an alert threshold at 20% for the interest rate risk indicator.

Some Italian banks, typically the large ones, also follow the indi- cations supplied by their own internal models for risk evaluation. In this regard, the Basel Committee has recently provided an over- view of the best practices used by the largest banking groups to compute the various risks that characterize their business activity, including the interest rate risk in the banking book (see Basel Committee on Banking Supervision, 2008). The supervisory author- ity validates banks’ internal models and their outcomes are usually collected in regular reports. However, in our case, an in-depth panel analysis of the consistency between the information deriving from the internal models and that from the standardized approach is very difficult because only a few large banks rely on internal models for risk evaluation15 and even these indicators cover too short a sample period for us to conduct a reliable econometric study.

14 For example, the band ‘‘up to one month’’ includes 1/12 of the residual amount, and the band ‘‘6 months–1 year’’ includes 6/12.

15 As of November 2012 only 13 Italian banking groups were using internal models for interest rate risk evaluation (see the Bank of Italy’s Financial Stability Report, No. 4, November 2012).

3.2. Other bank-specific characteristics

We match the information on banks’ interest rate risk with a number of bank-specific characteristics. Following previous stud- ies on the determinants of interest rate risk (e.g. Stein, 1998; Akella and Chen, 1990; Fraser et al., 2002; Saporoschenko, 2002; Reichert and Shyu, 2003; Au Yong et al., 2009; Ballester et al., 2009), we consider the following variables: size (logarithm of total assets), capitalization (ratio of core equity capital to total risk-weighted assets, i.e. the core tier 1 ratio), profitability (ratio of net income to shareholders’ equity, i.e. the return on equity), credit risk (ratio of non-performing loans to total assets) and fund- ing liquidity risk (ratio of the difference between loans and retail funds to loans, i.e. the funding gap).

4. Developments of interest rate risk exposure over time and across banks

Table 2 presents descriptive statistics regarding the Italian banks’ exposure to interest rate risk. As reported in Panel A, from the second half of 2008 through the first half of 2012 Italian banking groups’ exposure to interest rate risk was on average equal to �3.3% of regulatory capital, well below the 20% threshold. In the event of a parallel upward shift of the yield curve by 200 basis points, the economic value of the Italian banking system would have risen, on average, by 3.3% of regula- tory capital.

As illustrated in Fig. 2, banks modified their interest rate expo- sure over time. In the second half of 2008 and first half of 2009 the average interest rate risk was slightly positive (on average 3.1% of regulatory capital), as a result of a positive on-balance-sheet exposure that was only partially offset by the use of interest rate derivatives. Recall that in this period the ECB quickly reduced the official rate on the main refinancing operations by a cumulative 425 basis points to counteract the downward risks to price stability in the euro area connected to the worsening of macroeconomic outlook. Subsequently, from the second half of 2009 onwards, as the official rate stayed virtually unchanged at back then unprece- dented low levels and the slope of the yield curve increased, banks switched the sign of their on-balance-sheet duration gap from pos- itive to negative and kept their position in hedging derivatives vir- tually unchanged.

Table 2 Descriptive statistics on Italian banking groups’ interest rate risk exposures and other characteristics. Average values are computed from 2008H2 through 2012H1. Interest rate risk exposures are calculated according to the duration gap approach and are expressed as percentage of regulatory capital. Overall, on-balance-sheet and off-balance-sheet duration gaps are computed as in Eqs. (3) and (4) in Section 2. In particular, overall duration gap is calculated considering banks’ on- and off-balance- sheet items. On-balance-sheet duration gap is calculated only on banks’ on-balance-sheet items. Off-balance-sheet duration gap is calculated only on banks’ off-balance- sheet items. Regarding the categories of banks, liability-sensitive banks are those that exhibited a negative on-balance-sheet duration gap over the entire sample period. Asset-sensitive banks are those that instead exhibited a positive on-balance-sheet duration gap. Finally, the category other banks includes banks whose on-balance-sheet duration gap changed sign at least once over the sample period. As for the other bank-specific characteristics we report: total assets (millions of euro), the ratio of non- performing loans to total loans, the funding gap (ratio of the difference between loans and retail deposits to loans) the tier 1 ratio (ratio of core equity capital to total risk-weighted assets) and the return on equity (ratio of net income to shareholders’ equity). Interest rate exposures and the other bank-specific characteristics (excluding total assets) are expressed in percentage values.

All banks Liability-sensitive banks Asset-sensitive banks Other banks

Interest rate risk exposures: On-balance-sheet duration gap 0.0 �14.4 32.7 �0.3 Off-balance-sheet duration gap �3.1 1.6 �20.5 �1.6 Overall duration gap �3.1 �12.8 12.3 �1.8

Other bank-specific characteristics: Total assets 46,258 13,109 59,491 54,127 Non-performing loans/total loans 4.0 3.1 3.9 4.3 Funding gap �8.8 �39.4 43.1 �5.7 Funding gap (net of foreign banks) �7.6 �16.2 25.0 �7.2 Tier 1 ratio 13.0 13.2 13.1 12.9 Return on equity 3.7 4.5 3.3 1.2

N. banking groups 67 19 9 39 of which: derivatives users 56.0 15.0 7.0 34.0 of which: foreign banks 6.0 1.0 4.0 1.0 of which: top 5 banking groups 5.0 0.0 1.0 4.0

N. observations 498 135 62 301

Source: authors’ calculations on data from banks’ supervisory reports to the Bank of Italy.

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 491

This preliminary descriptive evidence for the entire banking system disregards a substantial degree of heterogeneity among intermediaries concerning their exposure to the interest rate risk, as well as the different strategies pursued to manage it. In this regard, we split the panel of banks into three distinct groups according to the sign of the on-balance-sheet duration gap: (1) banks that exhibited a negative on-balance-sheet duration gap over the entire sample period (hereafter, liability-sensitive banks); (2) banks that exhibited a positive on-balance-sheet duration gap over the entire sample period (hereafter, asset-sensitive banks); and (3) banks that varied the sign of their on-balance-sheet expo- sure over time (hereafter, other banks). As reported in Table 2, our panel comprises 20 asset-sensitive banks, 9 liability-sensitive banks and 39 other banks.

Fig. 3 shows the interest rate risk indicators for each of the three groups of banks. In the second half of 2008 liability-sensitive banks (see plot A) had an average duration gap of about �10%. As the yield curve quickly steepened from the second half of 2009 onwards, liability-sensitive banks decreased their on- balance-sheet exposure thus enhancing the gain from a potential increase in interest rates and barely relied on interest rate deriva- tives. For most of the sample period, asset-sensitive banks (see plot B) diminished their exposure to interest rate risk from 20% to 11% implementing a risk management strategy very different from that pursued by liability-sensitive peers, namely they relied to a large extent on interest rate derivatives to partially offset their on-balance-sheet interest rate risk exposure, which would have implied a loss if interest rates had increased.

Finally, the developments of interest rate risk for the other banks (see plot C), which account for about half of the interme- diaries in our sample, closely resembles what is shown in Fig. 2 for the entire banking system. Roughly speaking, these banks rapidly modified their overall interest rate risk by changing the sign of their on-balance-sheet duration gap, from being asset-sensitive in the first two semesters of the sample to liability-sensitive thereafter. Interestingly, while in the first two semesters these banks used interest rate derivatives to offset their on-balance-sheet exposure, in the following periods these

instruments were used to enhance the gains from a potential increase in interest rates.

This preliminary look at the data indicates that during the financial crisis most Italian banks actively used financial deriva- tives to manage their interest rate risk although with different goals: either as a hedging tool to offset the on-balance-sheet inter- est rate risk exposure or as an enhancing tool to maximize the potential gain arising from a rise in interest rates. Clearly, this descriptive evidence is only suggestive and an in-depth assessment of the correlation among variables requires a formal econometric analysis (this is what we do in the next sections).

Table 2 provides some insights into the relationship between interest rate risk exposure and the other bank-specific characteris- tics discussed in the previous section. Liability-sensitive intermedi- aries have been on average small-sized, more capitalized and less profitable. They have also exhibited a negative funding gap – meaning that they did not rely on wholesale funding to support their lending activity – and a little exposure to credit risk – possi- bly reflecting more stable lending relationships.

The group of ‘‘other banks’’, which represents the largest frac- tion of the Italian banking system, includes most of the largest banking groups. These intermediaries have been, on average, rela- tively less capitalized and largely exposed to credit risk. As for liability-sensitive banks, they do not appear to have been exposed to a liquidity risk.

Finally, the group of asset-sensitive banks also comprises large banks. They have been more profitable, on average, but they have been characterized by a much higher liquidity risk. Note that these intermediaries include branches and subsidiaries of foreign banks that usually get funding from the holding group. Excluding these banks, the average funding gap decreases from 40% to 25%.

5. Managing the exposures to interest rate risk: methodological issues

In this Section we describe the econometric approach used in the empirical analysis. Practical considerations may suggest that

-9

-6

-3

0

3

6

9

12

2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1

Off-balance-sheet interest rate risk

On-balance-sheet interest rate risk

Overall interest rate risk

Fig. 2. Developments of interest rate risk exposure of Italian banking groups in the period from 2008H2 through 2012H1. Interest rate risk exposures are calculated according to the duration gap approach and are expressed as percentage of regulatory capital. Overall, on-balance-sheet and off-balance-sheet duration gaps are computed as in Eqs. (3) and (4) in Section 2. For each semester we report average values across banks. Data are expressed in percentage points. Source: authors’ calculations on data from banks’ supervisory reports to the Bank of Italy.

Panel A: liability-sensitive banks Panel B: asset-sensitive banks

-30

-15

0

15

30

45

60

2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1 -20

-15

-10

-5

0

5

10

2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1

Panel C:other banks

-12

-8

-4

0

4

8

12

2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1

Off-balance-sheet interest rate risk

On-balance-sheet interest rate risk

Overall interest rate risk

Fig. 3. Developments of interest rate risk exposure for categories of Italian banking groups from 2008H2 through 2012H1. Interest rate risk exposures are calculated according to the duration gap approach and are expressed as percentage of regulatory capital. Overall, on-balance-sheet and off-balance-sheet duration gaps are computed as in Eqs. (3) and (4) in Section 2. The figure distinguishes among liability-sensitive banks (Panel A), asset-sensitive banks (Panel B) and other banks (Panel C). Liability-sensitive banks are those with a negative on-balance-sheet duration gap over the entire sample period. Asset-sensitive banks are those with a positive on-balance-sheet duration gap over the entire sample period. Other banks are those for which the on-balance-sheet duration gap changed sign at least once over the sample period. For each semester we report average values across banks. Data are expressed in percentage points. Source: authors’ calculations on data from banks’ supervisory reports to the Bank of Italy.

492 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 493

banks first evaluate their on-balance-sheet interest rate risk and then decide the strategy to achieve their desired level of overall interest rate risk exposure relying on interest rate derivatives. This sequential ordering in banks’ decisions is consistent with the view that frequent variations in the on-balance-sheet exposure are costly because of various factors that are out of banks’ control, such as customers’ preferences for loans and deposits, competition in the banking industry and access conditions to wholesale funding markets. This hypothesis can be tested empirically specifying the following panel regression in which the off-balance-sheet duration

gap (GAPOFFit ) is the dependent variable and the on-balance-sheet

duration gap (GAPONit ) is among the explanatory variables:

GAPOFFit ¼ c þ ai þ k0GAP ON it þ k1 SIZEit þ k2 CAPit þ k3ROEit

þ k4NPLit þ k5LIQ it þ k6 SLOPEt þ eit ð5Þ

In Eq. (5), c is the constant term, ai is a bank-specific fixed effect that is meant to capture across-bank unobserved heterogeneity that is not explained by the bank-specific characteristics included in the specification, i.e.: SIZEit (the logarithm of total assets), CAPit (the core tier 1 ratio), ROEit (the return on equity), NPLit (the indicator of credit risk), LIQ it (the indicator of funding liquidity risk). The variable SLOPEt is the slope of the yield curve (defined as in the difference between the 10-year Government bond yield and the 3-month Euribor rate) and eit is an idiosyncratic error term.

We are particularly interested in the sign of the coefficient k0 . A negative (positive) value for k0 would imply a hedging (enhancing) strategy, whereby suggesting that banks have used interest rate derivatives to offset (amplify) the on-balance-sheet exposure to interest rate risk. It is important to note that a negative (positive) value for k0 implies a hedging (enhancing) strategy irrespectively of whether the signs of the on-balance-sheet and the off-balance-sheet duration gap are concordant or not. Indeed, when the on-balance-sheet and the off-balance-sheet duration gap exhibit opposite signs, banks are offsetting a potential gain with a potential loss; when instead the on-balance-sheet duration gap and the off-balance-sheet duration gap have the same sign, banks are exposed to an increase of interest rates (when they are both positive) or from a decrease of interest rates (when they are both negative).

However, since we use a low-frequency (half-yearly) dataset and it is very difficult to directly track the timing of banks’ interest rate risk decisions that likely occur throughout each semester, we also estimate an alternative regression in which the on-balance-sheet duration gap is the dependent variable and the off-balance-sheet duration is the explanatory variable, i.e.

GAPONit ¼ c þ ai þ b0GAP OFF it þ b1 SIZEit þ b2 CAPit þ b3 ROEit

þ b4NPLit þ b5 LIQ it þ b6 SLOPEt þ eit ð6Þ

In Eq. (6) the sign of the coefficient b0 has the same economic interpretation as k0 in the previous equation specification.

Another consideration is in order. Using a different indicator of on-balance-sheet interest rate risk (i.e. the maturity gap) and the notional value of financial derivatives, Purnanandam (2007) points out that the two risk-management strategies should be modeled simultaneously in a system of equations and proposes the use of two-stage least squares instrumental variables (2SLS-IV) regres- sions to alleviate part of the endogeneity problem and the resulting simultaneity bias that could characterize the estimated OLS coeffi- cients. As hinted above, the endogeneity problem may be particu- larly relevant when estimating regressions based on low-frequency data.

Hence, for each of the above two specifications we perform a 2SLS-IV estimation. As for specification (5), we follow Purnanandam (2007) and instrument the on-balance-sheet

duration gap with its first-period lag. Here, the underlying idea is that on-balance-sheet restructuring is persistent in the short- run. As a matter of fact, making adjustments to their on- balance-sheet asset and liability management policies can be costly if banks must offer different terms to their relationship bor- rowers or depositors to achieve such adjustments. The first-stage regression of the 2SLS-IV estimation reads as:

GAPONit ¼ p0 þ p1GAP ON it�1 þ P

0X þ git ð7Þ

where P0 is a vector of coefficients and X is a vector including all remaining variables specified in (5). The second-stage regression is instead given by the following specification:

GAPOFFit ¼ c þ ai þ k0GAP ON it þ k1SIZEit þ k2 CAPit þ k3 ROEit

þ k4NPLit þ k5 LIQ it þ k6 SLOPEt þ eit ð8Þ

where GAPONit is the predicted value of GAP ON it , i.e.

GAPONit ¼ p0 þ p1 GAP ON it�1 þ P

0X . The relevance of the instrument is tested in the first-stage

regression. Staiger and Stock (1997) and other authors have docu- mented that the weak instrument problem can arise even when the first stage tests are significant at conventional levels, say 5% or 1%. In case of a single endogenous regressor and a single instru- ment, as it is in our case, the F-statistic associated with the signif- icance test for the excluded instrument should be larger than 10 (or similarly that the corresponding p-value should be smaller than 0.0016.

For the 2SLS-IV estimation of specification (6), in which the on-balance-sheet duration is treated as the dependent variable, we proceed in the same fashion. In this case, however, we rely on two different instruments for the off-balance-sheet duration gap. The first is its first-period lag (as done for the on-balance-sheet duration gap), which implies the following first-stage regression:

GAPOFFit ¼ p0 þ p1GAP OFF it�1 þ P

0X þ git ð9Þ

The second and alternative instrument is a ‘‘derivatives skill’’ dummy, as in Purnanandam (2007). Accordingly, the first-stage regression becomes the following:

GAPOFFit ¼ p0 þ p1SKILLi þ P 0X þ git ð10Þ

In particular, the dummy variable SKILL takes a value of one if a bank uses interest rates, commodities or exchange rates deriva- tives for trading purposes over the entire sample period. The ratio- nale for this instrument is that a bank with derivatives position for trading purposes has the necessary skills to engage in interest rate risk management using derivative instruments. Notice that by con- struction this ‘‘derivatives skill’’ dummy is a cross-sectional dummy, thus preventing from performing a bank-specific fixed-effect estimation.

In either case the second-stage regression reads as:

GAPONit ¼ c þ ai þ k0GAP OFF it þ k1SIZEit þ k2 CAPit þ k3 ROEit

þ k4NPLit þ k5 LIQ it þ k6 SLOPEt þ eit ð11Þ

where GAPOFFit is the predicted value of GAP OFF it calculated according

either to (9) or (10). Finally, to further address the endogeneity problem in testing

the empirical relation between the on-balance-sheet and the off-balance-sheet interest rate risks we also replace in (8) and (11) the SLOPE variable with a set of time fixed effects. Time dum- mies are important for two reasons. First, their inclusion may help limiting the endogeneity problem between the interest rate risk indicators as far as these two indicators might be driven by com- mon macroeconomic factors. Second, time dummies may capture

494 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

macroeconomic conditions different from the slope of the yield curve such as business cycle fluctuations that may affect banks’ interest rate risk decisions.

6. Estimation results

6.1. Results for the panel of all banks

Table 3 reports the estimation results for the whole sample of all banks. The table is divided in two panels. Panels A and B show the results obtained when the dependent variable is, respectively, the off-balance-sheet duration gap and the on-balance-sheet duration gap. Hypothesis testing is based on Newey–West heteroskedasticity and autocorrelation (HAC) consistent standard errors with lag trun- cation of two semesters. We first discuss the empirical relation between the on-balance- and the off-balance-sheet duration gaps and then assess the role of the other explanatory variables.

In both Panels A and B simple OLS regressions show that the estimated coefficients linking the off-balance-sheet and on-balance-sheet duration gaps are negative and statistically sig- nificant. These results indicate that Italian banking groups carried out a hedging strategy whereby using the on-balance-sheet restructuring and the interest rate derivatives as substitute instru- ments to achieve their desired level of interest rate risk exposure. On average, banks hedged about 20% of their on-balance-sheet interest rate risk exposure using derivatives (see Panel A) and about 75% of their off-balance-sheet interest rate risk exposure using the on-balance-sheet restructuring (see Panel B).

In order to account for the possible endogeneity between the on-balance-sheet and the off-balance-sheet duration gaps and the simultaneity bias in single-equation regressions, we also report the estimated coefficients from 2SLS-IV regressions.16 In particular, we present all the estimated coefficients for the second-stage regres- sion whereas from the first-stage regression we only report the coef- ficient of the variable used as instrument, in order to save space. We also show the F-statistic to test the null hypothesis that the used instrument is weakly correlated with the dependent variable in the first-stage regression. The results of the first-stage regressions are shown in the bottom part of each panel.

In Panel A we present the results when the off-balance-sheet duration gap is the dependent variable and the on-balance-sheet duration gap is instrumented with its first-period lag. This approach is consistent with the view that frequent changes in the on-balance exposure may be costly for banks. Column (2) suggests that, in the first-stage regression, the instrument enters significantly with a positive coefficient. The F-statistic for the weak identification test exceeds the reference value of 10, thus providing support about the reliability of the instrument. As for the second-stage regression, which is reported in the upper part of the panel, the estimation results confirm the negative and highly statistically significant rela- tionship between the off-balance-sheet and the on-balance-sheet duration gaps. Compared to the OLS estimation, the 2SLS-IV regres- sions provide somewhat larger (in absolute value) estimated coefficients indicating that the simultaneity bias is rather limited in our case. For robustness we also present in column (3) the esti- mated coefficients of a 2SLS-IV regression in which we replace the slope of the yield curve with a complete set of time fixed effects. The results remain virtually the same.

In Panel B we present the results when the on-balance-sheet duration gap is treated as the dependent variable and the off-balance-sheet duration gap is the endogenous regressor. In

16 The endogeneity issue may also regard the other bank-specific characteristics. We run alternative regressions in which all these regressors enter the specification with a 1-period lag. The estimation results (not reported but available upon request) remain virtually unchanged.

the regression presented in column (2) the off-balance-sheet dura- tion gap is instrumented with its first-period lag. Interestingly, the F-test lends support to the validity of our instrument and in the second-stage regression the estimated coefficient for the off-balance-sheet duration gap is again negative, highly significant and somewhat larger (in absolute value) than that obtained with the OLS regression, thus indicating that the simultaneity bias is limited also in this case. Column (3) shows that these results are very robust to the inclusion of time dummies.

We now discuss alternative regressions in which we use the derivative-skill dummy as instrument for the off-balance duration gap. Columns (4) and (5) refer to regressions excluding and com- prising time fixed effects, respectively. An important result is that the F-test is well below 10, thus suggesting a weak instrument problem. A possible explanation is that most of the banks in our sample held derivatives positions in the banking book as well as in the trading book, thus making hard to discriminate between ‘‘skilled’’ and ‘‘not-skilled’’ banks. The weak instrument problem naturally leads to a non-statistically significant estimated coeffi- cient for the off-balance-sheet duration gap in the second-stage regressions.

The estimated coefficients for the other bank-specific variables included in the various regressions suggest some other important findings. First, the funding gap turns out to be negatively correlated with the off-balance-sheet duration gap while not significantly related to the on-balance-sheet duration gap. These results would suggest that, on average, when banks faced a higher liquidity risk they used interest rate derivatives to enhance the potential gain from an increase in interest rates.17 This outcome is broadly robust to the various specifications. Interestingly, Froot et al. (1993) and Purnanandam (2007) found that US banks with a relatively low liq- uidity ratio, as captured by the ratio between cash and securities to total assets, made more extensive use of financial derivatives.

Second, the correlation between the credit quality indicator and the interest rate risk measures is not statistically significant. Nevertheless in the next sections we show that this link becomes significant if we look at specific groups of banks. Purnanandam (2007) found that US banks with a higher probability of default, as captured by the ratio of non-performing to total assets, main- tained a lower maturity mismatch between their assets and liabilities.

Third, we find a negative coefficient for bank size in the regres- sions where the off-balance-sheet duration gap is the dependent variable (see Panel A). This outcome suggests that large banks might have used interest rate derivatives to increase the potential gain stemming from an upward shift in interest rates. Since larger banks have typically better access to capital markets and benefit from a greater diversification, they prefer to pursue riskier activi- ties and therefore a higher exposure to interest rate risk may be expected. Moreover, because of moral hazard considerations, too-big-to-fail banks may also have the incentive to take on more risk (Ballester et al., 2009).

Fourth, the positive and significant coefficient between the ROE and the off-balance-sheet duration gap suggests that more efficient banks were more risk-averse and decreased the potential gain stemming from an upward shift of interest rates.

Finally, we do not find a significant relationship between inter- est rate risk and the Tier 1 ratio. Banks with higher capital ratios are usually considered as more risk adverse (Gennotte and Pyle, 1991; Dewatripont and Tirole, 1994) and one could expect an inverse relation between capitalisation and interest rate risk. Furthermore, banks with high capital ratios are less prone to

17 Looking at foreign banks the impact is nil. In this case the variable is not a reliable measure of liquidity risk since foreign banks belonging to groups usually get funds from the holding bank.

Table 3 Management of interest rate risk exposure of Italian banking groups. Estimation is based on half-yearly data from 2008H2 through 2012H1. Panels A and B report the estimation results obtained with ordinary least square (OLS) (in column (i)) and with two-stage least square instrumental variable (2SLS-IV) (in columns (ii) and (iii)). For a discussion of the 2SLS-IV estimation see Section 5. For 2SLS-IV estimations we report all the estimated coefficients of the second-stage regression and the estimated coefficient of the solely instrument computed in the first-stage regression. The F-statistic for weak identification test is also reported in the lower part of each panel. All estimations use Newey–West heteroskedasticity- and autocorrelation-consistent standard errors.

Panel A

Dependent variable: OLS 2SLS-IV 2SLS-IV Off-balance-sheet duration gap (1) (2) (3)

Explanatory variables: Results from second-stage regression

On-balance-sheet duration gap �0.20⁄⁄ On-balance-sheet duration gap (predicted) – �0.28⁄⁄ �0.32⁄ Slope of the yield curve 0.01 0.02 – Size �0.12⁄⁄⁄ �0.11⁄ �0.10⁄⁄

Non-performing loans/total loans 0.15 �0.09 �0.03 Funding gap �0.02⁄⁄ �0.04⁄⁄ �0.03⁄ Funding gap⁄ Foreign dummy 0.03⁄⁄ 0.04⁄⁄ 0.04⁄⁄⁄

Foreign dummy – – – Tier 1 ratio 0.11 �0.07 �0.04 ROE 0.06 0.09⁄ 0.08 Bank-specific fixed effects Yes Yes Yes Time-specific fixed effects No No Yes

Results from first-stage regression

Endogenous variable: on-balance-sheet duration gap Instrument: on-balance-sheet duration gap(-1) – 0.366⁄⁄⁄ 0.351⁄⁄⁄

F-statistic for weak identification test – 13.41 11.35 N. of observations 496 429 429 N. of banks 67 66 66 Uncentred R-squared 0.28 0.28 0.29

Panel B

Dependent variable: OLS 2SLS-IV 2SLS-IV 2SLS-IV 2SLS-IV On-balance-sheet duration gap (1) (2) (3) (4) (5)

Explanatory variables: Results from second-stage regression

Off-balance-sheet duration gap �0.75⁄⁄⁄ Off-balance-sheet duration gap (predicted) �0.89⁄⁄⁄ �0.81⁄⁄⁄ �1.00 �1.10 Slope of the yield curve �0.01⁄⁄ 0.00 – �0.02⁄⁄⁄ – Size 0.03 0.00 0.02 0.00 0.00 Non-performing loans/total loans �0.80 �0.90 �0.68 0.98 1.13 Funding gap 0.00 �0.01 0.01 0.04⁄⁄⁄ 0.04⁄⁄⁄ Funding gap⁄ Foreign dummy 0.05⁄⁄ 0.05⁄⁄ 0.04⁄ 0.05 0.04 Foreign dummy – – – 0.10 0.08 Tier 1 ratio �0.31 �0.32 �0.22 0.24 0.22 ROE 0.00 0.00 �0.03 0.00 �0.02 Bank-specific fixed effects Yes Yes Yes No No Time-specific fixed effects No No Yes No Yes

Results from first-stage regression

Endogenous variable: off-balance-sheet duration gap Instrument: off-balance-sheet duration gap(-1) – 0.50⁄⁄⁄ 0.51⁄⁄⁄ – – Instrument: skill derivatives dummy – – – 0.02 0.02 F-statistic for weak identification test – 43.2 44.3 1.49 1.46 N. of observations 496 429 429 496 496 N. of banks 67 66 66 67 67 Uncentred R-squared 0.24 0.24 0.32 0.89 0.66

⁄, ⁄⁄, ⁄⁄⁄ denote the statistical significance of the coefficient at the 10%, 5% and 1% level, respectively.

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 495

financial distress and bankruptcy and thus might be willing to use less intensively financial derivatives for hedging purposes.

6.2. Results for different groups of banks

In this section we exploit the heterogeneity that characterizes the sample of Italian banks in order to assess how risk management decisions differed among different groups of banks. To this end, we estimate separately the 2SLS-IV bank-specific fixed-effects regres- sions for the three groups of intermediaries discussed above, namely the liability-sensitive banks, the asset-sensitive banks and the other banks. The estimated coefficients are reported in Table 4. As in the previous section, Panels A and B show the results obtained when

the dependent variable is, respectively, the off-balance-sheet dura- tion gap and the on-balance-sheet duration gap.

As for the relationship between the on-balance-sheet and the off-balance-sheet duration gap, the evidence of simultaneity of hedging is confirmed only for asset-sensitive banks, for which the estimated coefficients double in magnitude with respect to the whole panel of banks. The estimated coefficients are not statis- tically significant for liability-sensitive banks. Since these interme- diaries would have gained from an increase in interest rates, they did not need to offset their on-balance-sheet exposure. As regard the ‘‘other banks’’, the correlation between off-balance-sheet dura- tion gap and on-balance-sheet duration gap is negative in Panel B. Overall, the heterogeneity in banks’ hedging strategies against

Table 4 Management of interest rate risk exposure of Italian banking groups: heterogeneity across banks. Estimation is based on half-yearly data from 2008H2 through 2012H1. Panels A and B report the estimation results obtained with two-stage least square instrumental variable (2SLS-IV). Column (i) reports the estimates for the entire sample of banks whereas columns (ii), (iii) and (iv) report the results for, respectively, liability-sensitive, asset-sensitive and other banks. For the definition of these three categories of bank see note to Table 2. For 2SLS-IV estimations we report all the estimated coefficients of the second-stage regression and the estimated coefficient of the solely instrument computed in the first-stage regression. The F-statistic for weak identification test is also reported in the lower part of each panel. All estimations use Newey–West heteroskedasticity- and autocorrelation-consistent standard errors.

Panel A

Dependent variable: Liability-sensitive banks Asset-sensitive banks Other banks off-balance-sheet duration gap

Results from second-stage of 2SLS-IV estimation

Explanatory variables: (1) (2) (3) (4) (5) (6) On-balance-sheet duration gap (predicted) �0.06 �0.10 �0.56⁄⁄⁄ �0.53⁄⁄⁄ �0.21 �0.24 Slope of the yield curve 0.00 – 0.05⁄⁄ – 0.01 – Size 0.01 �0.01 �0.25⁄ �0.19 �0.13⁄ �0.12⁄ Non-performing loans/total loans 0.93⁄ 0.90⁄ �3.68⁄ �3.22 0.01 0.05 Funding gap 0.01⁄⁄ 0.01 �0.44⁄⁄⁄ �0.37⁄⁄ �0.08⁄⁄ �0.08⁄⁄ Funding gap⁄ Foreign dummy 0.00 0.00 0.15 0.02 0.82⁄⁄⁄ 0.76⁄⁄⁄

Foreign dummy – – – – – – Tier 1 ratio �0.11 �0.15 �0.37 �0.10 �0.18 �0.14 ROE �0.02 �0.02 0.08 0.03 0.04 0.03 Bank-specific fixed effects Yes Yes Yes Yes Yes Yes Time-specific fixed effects No Yes No Yes No Yes

Results from first-stage of 2SLS-IV estimation

Endogenous variable: on-balance-sheet duration gap Instrument: on-balance-sheet duration gap(�1) 0.33⁄⁄⁄ 0.21⁄⁄ 0.51⁄⁄ 0.51⁄⁄⁄ 0.35⁄⁄ 0.33⁄⁄ F-statistic for weak identification test 13.80 5.48 7.90 9.11 6.38 5.54

N. of observations 116 116 53 53 259 259 N. of banks 18 18 9 9 38 38 Uncentred R-squared 0.28 0.17 0.21 0.67 0.34 0.34

Panel B

Dependent variable: Liability-sensitive banks Asset-sensitive banks Other banks On-balance-sheet duration gap

Results from second-stage of 2SLS-IV estimation

Explanatory variables: (1) (2) (3) (4) (5) (6) Off-balance-sheet duration gap (predicted) �0.86 �0.73⁄⁄ �1.52⁄⁄⁄ �1.51⁄⁄⁄ �0.65⁄⁄⁄ �0.57⁄⁄⁄ Slope of the yield curve 0.01 – 0.07⁄⁄ – �0.02 – Size �0.11 �0.11 �0.39 �0.28 0.05 0.08 Non-performing loans/total loans 0.91 1.22 �5.86⁄ �6.08⁄ �1.16 �0.93 Funding gap 0.01 0.03 �0.76⁄⁄⁄ �0.79⁄⁄⁄ �0.02 0.01 Funding gap⁄ Foreign dummy 0.03 0.01 0.28 0.21 0.17 �0.02 Foreign dummy – – – – – – Tier 1 ratio 0.26 0.00 �1.03 �0.95 �0.37 �0.19 ROE 0.06 0.04 0.06 �0.10 �0.09 �0.11 Bank-specific fixed effects Yes Yes Yes Yes Yes Yes Time-specific fixed effects no Yes No Yes No Yes

Results from first-stage of 2SLS-IV estimation

Endogenous variable: off-balance-sheet duration gap Instrument: off-balance-sheet duration gap(�1) 0.61⁄⁄⁄ 0.65⁄⁄⁄ 0.70⁄⁄⁄ 0.69⁄⁄⁄ 0.50⁄⁄⁄ 0.50⁄⁄⁄ F-statistic for weak identification test 14.41 22.75 34.59 32.95 18.29 24.28

N. of observations 116 116 53 53 259 259 N. of banks 18 18 9 9 38 38 Uncentred R-squared 0.325 0.59 0.61 0.66 0.19 0.26

Panel C

Dependent variable: Liability-sensitive banks Asset-sensitive banks Other banks On-balance-sheet duration gap

Results from second-stage of 2SLS-IV estimation

Explanatory variables: (1) (2) (3) (4) (5) (6) Off-balance-sheet duration gap (predicted) 0.32 0.28 �1.48⁄⁄ �1.42⁄⁄⁄ 1.33 1.48 Slope of the yield curve �0.02⁄⁄⁄ �0.03 �0.02 Size 0.01⁄⁄ 0.01⁄⁄ �0.09 �0.08 �0.01 �0.01 Non-performing loans/total loans 0.76⁄⁄ 0.78⁄⁄⁄ 4.44 3.88 �0.52 �0.67 Funding gap 0.04⁄⁄⁄ 0.04⁄⁄⁄ �0.18 �0.16 0.03⁄⁄ 0.03⁄⁄ Funding gap⁄ Foreign dummy 0.01 0.00 �0.76 �0.76 �0.96 �0.91 Foreign dummy �0.05 �0.05 0.58 0.58 0.38 0.35 Tier 1 ratio 0.99⁄⁄⁄ 0.96⁄⁄⁄ �2.07 �1.83⁄ �0.61 �0.63 ROE 0.16⁄ 0.19⁄⁄ �0.37 �0.57 0.34 0.36 Bank-specific fixed effects No No No No No No Time-specific fixed effects No Yes No Yes No Yes

496 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

Table 4 (continued)

Panel C

Dependent variable: Liability-sensitive banks Asset-sensitive banks Other banks On-balance-sheet duration gap

Results from first-stage of 2SLS-IV estimation Endogenous variable: off-balance-sheet duration gap Instrument: derivatives skill dummy 0.04⁄⁄⁄ 0.04⁄⁄⁄ �0.35 �0.25 �0.01 �0.01 F-statistic for weak identification test 15.38 15.40 1.96 2.28 1.17 1.00

N. of observations 135 135 61 61 300 300 N. of banks 18 18 Uncentred R-squared 0.85 0.89 0.90 0.92 – –

⁄, ⁄⁄, ⁄⁄⁄ denote the statistical significance of the coefficient at the 10%, 5% and 1% level, respectively.

-15

-10

-5

0

5

10

2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1

shortening maturity of overnight deposits

benchmark case

lengthening maturity of overnight deposits

Fig. 4. Effects of lengthening or shortening the maturity of overnight deposits for the computation of the on-balance-sheet interest rate risk of Italian banking groups. As discussed in Section 3, the Italian regulatory framework considers 25% of total overnight deposits as a ‘‘non-core’’ component and includes it in the ‘‘overnight’’ time-band; the remaining ‘‘core’’ component is allocated to the time-bands from ‘‘up to one month’’ to ‘‘over four years up to five years’’ in proportion to the number of months assigned to each band. In the case of lengthening (shortening) the maturity of overnight deposits we set to 5% (50%) the non-core’’ component. On-balance-sheet interest rate risk is calculated according to the duration gap approach and is expressed as percentage of regulatory capital (see Section 2). For each semester we report average values across banks. Data are expressed in percentage points. Source: authors’ calculations on data from banks’ supervisory reports to the Bank of Italy.

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 497

interest rate risk simply reflects the different on-balance-sheet position but does not pose significant threat to financial stability.

We now turn to the discussion of the interaction among risks of different nature. As far as liquidity risk is considered, some results are noteworthy. In the case of the asset-sensitive banks we find a negative and significant coefficient for the funding gap indicator both in Panels A and B. When the liquidity risk increased, these banks limited their exposure to a potential increase in interest rates by reducing both the on-balance-sheet and the off-balance-sheet duration gap. For liability-sensitive banks, which have essentially no liquidity risk (see again Table 2), we find a sig- nificant positive correlation with the off-balance sheet exposure. For the ‘‘other banks’’, the evidence is in line with that of the entire sample: in the face of a higher liquidity risk, these banks imple- mented a strategy that would have entailed an off-balance-sheet capital gain in case of an upward shift in the yield curve.

As for credit risk, we find a negative and significant coefficient both in Panels A and B for asset-sensitive banks, in presence of a higher liquidity risk, limited their exposure to a potential increase

in interest rates by reducing both the on-balance-sheet and the off-balance-sheet duration gap. A positive relationship with the off-balance-sheet duration gap is recorded for liability-sensitive banks in Panel A. In this case, a higher credit risk is associated to an increase in the off-balance-sheet duration gap, meaning that these banks used derivatives to increase the potential gain stem- ming from a decrease in interest rates. Since this group of banks is more risk averse (they have more capital and liquidity) and a dis- tinctive feature of the crisis was the sudden and dramatic deterio- ration of loan quality, they relied on interest rate derivatives in order to compensate the strong increase in the credit risk. Interestingly, over the sample period the ratio of non-performing loans to total loans more than doubled for liability-sensitive banks, as opposed to a 40% and 80% increase for asset-sensitive banks and other banks, respectively.

Overall, our main findings suggest a significant heterogeneity across intermediaries in the estimated correlation among risks. On the one hand, we find that asset-sensitive banks followed an integrated risk-management approach aiming at counterbalancing

Table 5 Measuring Italian banking groups’ interest rate risk exposure: changing the maturity for overnight deposits and considering non-parallel shifts in the yield curve. As discussed in Section 3, the Italian regulatory framework considers 25% of total overnight deposits as a ‘‘non-core’’ component and includes it in the ‘‘overnight’’ time-band; the remaining ‘‘core’’ component is allocated to the time-bands from ‘‘up to one month’’ to ‘‘over four years up to five years’’ in proportion to the number of months assigned to each band. Furthermore, the computation of the interest rate risk exposure is based on a parallel shift in the yield curve. In the case of changing the maturity of overnight deposits we consider two scenarios: lengthening (the non-core component is decreased to 5%) and shortening (the non-core component is increased to 50%). Also in the case of non-parallel shift in the yield curve we consider two scenarios: downward-sloping shift (short-term interest rates rise by 200 basis point at maturities up to one year while longer-term interest rates by 100 basis points) and upward-sloping shift (short-term interest rates rise by 200 basis point while longer-term interest rates by 300 basis points). In these cases the weighting factors are reported in Table 1. Interest rate risk exposures are calculated according to the duration gap approach and are expressed as percentage points of regulatory capital (see Section 2).

Benchmark Changing maturity of overnight deposits

Non-parallel shift in yield curve

Shortening Lenghtening Downward Upward

Interest rate risk exposures: On-balance-sheet duration gap 0.0 �4.6 4.8 �4.3 4.0 Off-balance-sheet duration gap �3.1 �3.1 �3.1 �1.6 �9.9 Overall duration gap �3.1 �7.7 1.7 �5.9 �5.9

Number of: Liability-sensitive banks 19 28 11 28 11 Asset-sensitive banks 9 6 18 3 9 Other banks 39 33 38 35 46

Source: authors’ calculations on data from banks’ supervisory reports to the Bank of Italy.

-10

-8

-6

-4

-2

0

2

4

6

8

10

2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1

downward-sloping shift in yield c urve

benchmark case

upward-sloping shift in yield curve

Fig. 5. Effects of assuming a non-parallel shift of the term structure of interest rates for the computation of the on-balance-sheet interest rate risk of Italian banking groups. As discussed in Section 3, according to the Italian regulatory framework the computation of the interest rate risk exposure is based on a parallel shift in the yield curve. In this figure we consider the effects of either a downward sloping or a downward sloping shift of the term structure of interest rates on the overall interest rate risk exposure. In the former case we assume that short-term interest rates rise by 200 basis point at maturities up to one year while longer-term interest rates by 100 basis points whereas in the latter case we assume that the short-term interest rates rise by 200 basis point while longer-term interest rates by 300 basis points). The weighting schemes and modified durations used for these exercises are reported in Table 1. Overall interest rate risk is calculated according to the duration gap approach and is expressed as percentage of regulatory capital (see Section 2). For each semester we report average values across banks. Data are expressed in percentage points. Source: authors’ calculations on data from banks’ supervisory reports to the Bank of Italy.

498 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

different risks. On the other, liability-sensitive banks reacted to a rise of liquidity and credit risk by enhancing the gains on their off-balance-sheet exposure from a potential reduction of interest rate. The majority of Italian banks (i.e. ‘‘other banks’’) instead, in the face of a higher liquidity risk, enhances the gains on their off-balance sheet exposure stemming from a potential increase in interest rates. While some of these outcomes (i.e. the enhancing strategy followed by some banks) may rise some concerns for pol- icymakers and regulators, it should be noted that the limited

exposure of Italian banks to the interest rate risk may make such arguments of a second-order importance.

7. Robustness checks

In this section we present several robustness checks of our pre- vious findings regarding: (i) the duration of overnight deposits; (ii) the possibility of non-parallel shifts of the yield curve; (iii) the potential non-stationarity of the data.

Table 6 Management of interest rate risk exposure of Italian banking groups: the effects of changing the maturity of overnight deposits. The estimation is based on half-yearly observations spanning the period 2008H2–2012H1. Panels A and B report the estimation results obtained with two-stage least square instrumental variable (2SLS-IV) method. In each panel we report the estimation results in the two cases of shortening or lengthening the duration of overnight deposits. AB, LSB, ASB and OB stand respectively for: all banks, liability-sensitive banks, asset-sensitive banks and other banks. For the definition of these three categories of bank see note to Table 2. For 2SLS-IV estimations we report all the estimated coefficients of the second-stage regression and the estimated coefficient of the solely instrument computed in the first-stage regression. The F-statistic for weak identification test is also reported in the lower part of each panel. All estimations use Newey–West heteroskedasticity- and autocorrelation-consistent standard errors.

Panel A

Shortening the duration of overnight deposits Lengthening the duration of overnight deposits

Dependent variable: All banks L–S banks A–S banks Other banks All banks L–S banks A–S banks Other banks Off-balance-sheet duration gap (1) (2) (3) (4) (5) (6) (7) (8)

Results from second-stage of 2SLS-IV estimation Results from second-stage of 2SLS-IV estimation Explanatory variables: On-balance-sheet duration gap �0.27⁄⁄⁄ �0.11 �0.73⁄⁄⁄ �0.21⁄ �0.29 �0.18⁄ �0.73⁄⁄⁄ �0.19 Slope of the yield curve 0.02 0.00 0.07⁄⁄⁄ 0.01 0.02 0.00 0.07⁄⁄⁄ 0.02 Size �0.11⁄ �0.01 �0.73⁄⁄⁄ �0.13⁄ �0.11⁄ �0.01 �0.79⁄⁄⁄ �0.14⁄⁄ Non-performing loans/total loans �0.04 0.56 �3.99⁄⁄ 0.09 �0.15 0.46 �4.05⁄⁄ 0.06 Funding gap �0.03⁄ 0.01 0.32 �0.08⁄⁄ 0.04⁄⁄ 0.00 0.20 �0.08⁄⁄⁄ Funding gap⁄ Foreign dummy 0.04⁄⁄ 0.00 �0.75 0.47 0.04⁄⁄⁄ 0.00 �0.66 0.48 Foreign dummy – – – – – – – – Tier 1 ratio �0.03 0.11 �2.86⁄⁄ �0.20 �0.11 0.11 �3.26⁄⁄ �0.24 ROE 0.09⁄ �0.02 0.08 0.05 0.09⁄ �0.01 0.06 0.05 Bank-specific fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Time-specific fixed effects No No No No No No No No

Results from first-stage of 2SLS-IV estimation Results from first-stage of 2SLS-IV estimation Endog. var.: on-balance-sheet duration gap Instrument: on-balance-sheet duration gap(�1) 0.35⁄⁄⁄ 0.26⁄⁄⁄ 0.42⁄ 0.35⁄⁄ 0.38⁄⁄⁄ 0.34⁄⁄⁄ 0.42⁄⁄ 0.36⁄ F-statistic for weak identification test 14.76 12.41 4.26 7.83 12.06 23.93 4.55 6.37

N. of observations 427 170 35 222 427 170 35 222 N. of banks 64 26 5 33 64 26 5 33 Uncentred R-squared 0.28 0.02 0.73 0.36 0.28 0.02 0.73 0.36

Panel B

Shortening the duration of overnight deposits Lengthening the duration of overnight deposits

Dependent variable: All banks L–S banks A–S banks Other banks All banks L–S banks A–S banks Other banks On-balance-sheet duration gap (1) (2) (3) (4) (5) (6) (7) (8)

Results from second-stage of 2SLS-IV estimation Results from second-stage of 2SLS-IV estimation

Explanatory variables: Off-balance-sheet duration gap �0.91⁄⁄⁄ �0.61 �1.55⁄⁄⁄ �0.76⁄⁄⁄ �0.86⁄⁄⁄ �0.57 �1.56⁄⁄⁄ �0.65⁄⁄⁄ Slope of the yield curve �0.01 �0.01 0.09⁄⁄⁄ �0.02 0.00 �0.01 0.09⁄⁄⁄ �0.02 Size 0.00 �0.01 �0.98⁄⁄⁄ 0.04 �0.01 0.03 �1.07⁄⁄⁄ 0.04 Non-performing loans/total loans �0.76 0.38 �6.53⁄⁄⁄ �0.87 �1.09⁄ �0.22 �6.65⁄⁄⁄ �1.20⁄ Funding gap 0.00 0.03 0.54 �0.03 �0.02 0.00 0.38 �0.03 Funding gap⁄ Foreign dummy 0.06⁄⁄ 0.03 �1.11 �0.01 0.05⁄⁄ 0.03 �0.99 �0.06 Foreign dummy – – – – – – – – Tier 1 ratio �0.21 0.77⁄ �3.05⁄ �0.41 �0.47 0.49 �3.56⁄⁄ �0.62 ROE 0.01 0.02 0.18 �0.07 �0.01 0.04 0.15 �0.09 Bank-specific fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Time-specific fixed effects No No No No No No No No

Results from first-stage of 2SLS-IV estimation Results from first-stage of 2SLS-IV estimation

Endog. var.: off-balance-sheet duration gap Instrument: off-balance-sheet duration gap(�1) 0.50⁄⁄⁄ 0.46⁄⁄⁄ 0.67⁄⁄⁄ 0.52⁄⁄⁄ 0.50⁄⁄⁄ 0.46⁄⁄⁄ 0.67⁄⁄⁄ 0.52⁄⁄⁄ F-statistic for weak identification test 43.2 28.08 27.58 20.08 43.2 28.08 27.58 20.08

N. of observations 427 170 35 222 427 170 35 222 N. of banks 64 26 5 33 64 26 5 33 Uncentred R-squared 0.25 0.35 0.77 0.20 0.25 0.35 0.77 0.20

⁄, ⁄⁄, ⁄⁄⁄ denote the statistical significance of the coefficient at the 10%, 5% and 1% level, respectively.

18 A similar exercise could have been done with regards to the option risk embedded in mortgages. However, we decided not to do so because, although the 2007 mortgage market reform in Italy (which, among other things, eliminated prepayment penalties) effectively increased the mobility of banks’ customers, mortgage substitutions and renegotiations became less attractive with worsening financing terms on new mortgages. Also the various initiatives promoted by the banks and the government to protect low-income and distressed borrowers during the crisis, while succeeding in improving the creditworthiness of the participating households (see Bartiloro et al., 2012) had a modest aggregate impact due to the strict requirements for accessing these initiatives and the limited amount of funds allocated for these purposes.

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 499

7.1. Changing the duration of overnight deposits

Overnight deposits do not have an explicit maturity. As dis- cussed in Section 3, the Italian regulatory framework considers 25% of total overnight deposits as a ‘‘non-core’’ component and includes it in the ‘‘overnight’’ time-band; the remaining ‘‘core’’ component is allocated to the following eight time-bands (from ‘‘up to one month’’ to ‘‘over four years up to five years’’) in propor- tion to the number of months assigned to each band. Clearly, dif- ferent distributions of overnight deposits by time-band may imply significant changes in banks’ exposure to the interest rate risk. And this may be particularly relevant for Italy where the ratio

of overnight deposits to total assets is large in the international comparison.18

Table 7 Management of interest rate risk exposure of Italian banking groups: the effects of assuming a non-parallel shift the yield curve. The estimation is based on half-yearly observations spanning the period 2008H2–2012H1. Panels A and B report the estimation results obtained with two-stage least square instrumental variable (2SLS-IV) method. In each panel we report the estimation results in the two cases of an upward-sloping shift and of a downward-sloping shift. In the former case we assume that the short-term interest rates rise by 200 basis point at maturities up to one year while longer-term interest rates by 100 basis points. In the latter case instead we assume that the short-term interest rates rise by 200 basis point while longer-term interest rates by 300 basis points. The weighting factors used to compute the interest rate risk exposures are reported in Table 1. AB, LSB, ASB and OB stand respectively for: all banks, liability-sensitive banks, asset-sensitive banks and other banks. For the definition of these three categories of bank see note to Table 2 For 2SLS-IV estimations we report all the estimated coefficients of the second-stage regression and the estimated coefficient of the solely instrument computed in the first-stage regression. The F-statistic for weak identification test is also reported in the lower part of each panel. All estimations use Newey–West heteroskedasticity- and autocorrelation-consistent standard errors.

Panel A

Downward-sloping shift in the yield curve Upward-sloping shift in the yield curve

Dependent variable: All banks L–S banks A–S banks Other banks All banks L–S banks A–S banks Other banks Off-balance-sheet duration gap (1) (2) (3) (4) (5) (6) (7) (8)

Results from second-stage of 2SLS-IV estimation Results from second-stage of 2SLS-IV estimation Explanatory variables: On-balance-sheet duration gap �0.34⁄⁄⁄ �0.76⁄⁄⁄ �0.14⁄ �0.25⁄ �0.74⁄⁄⁄ �0.81⁄⁄⁄ �0.35⁄⁄⁄ �0.73⁄⁄⁄ Slope of the yield curve 0.01 0.07 0.00 0.01 0.01 0.03⁄⁄ �0.01 0.01 Size �0.07 �0.70⁄⁄ 0.00 �0.10⁄ �0.02 �0.14 0.07 �0.02 Non-performing loans/total loans 0.00 �4.40⁄⁄⁄ 0.47⁄⁄ �0.02 0.12 �0.03 2.05⁄⁄⁄ 0.17 Funding gap �0.02 �0.47⁄⁄ 0.01 �0.06⁄⁄ �0.01 �0.17 0.04⁄⁄⁄ �0.03 Funding gap⁄ Foreign dummy 0.03⁄⁄ – 0.00 0.22 0.03⁄⁄ �0.12 �0.02⁄⁄⁄ 0.12 Foreign dummy – – – – – – – – Tier 1 ratio �0.01 �3.22⁄ 0.12 �0.12 0.16 0.19 0.08 0.12 ROE 0.07⁄ 0.32 �0.03 0.05 0.06 0.18 �0.02 0.03 Bank-specific fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Time-specific fixed effects No No No No No No No No

Results from first-stage of 2SLS-IV estimation Results from first-stage of 2SLS-IV estimation Endog. var.: on-balance-sheet duration gap Instrument: on-balance-sheet duration gap(�1) 0.36⁄⁄⁄ 0.47⁄⁄ 0.24⁄⁄⁄ 0.35⁄⁄ 0.37⁄⁄ 0.44⁄⁄⁄ 0.24⁄⁄⁄ 0.36⁄ F-statistic for weak identification test 17.36 6.99 10.47 9.78 11.86 7.83 13.64 6.51

N. of observations 427 21 196 210 427 63 77 287 N. of banks 64 3 28 33 64 9 11 44 Uncentred R-squared 0.35 0.89 0.04 0.43 0.87 0.92 0.67 0.89

Panel B

Downward-sloping shift in the yield curve Upward-sloping shift in the yield curve

Dependent variable: All banks L–S banks A–S banks Other banks All banks L–S banks A–S banks Other banks On-balance-sheet duration gap (1) (2) (3) (4) (5) (6) (7) (8)

Results from second-stage of 2SLS-IV estimation Results from second-stage of 2SLS-IV estimation Explanatory variables: Off-balance-sheet duration gap �0.95⁄⁄⁄ �1.36⁄⁄⁄⁄ �0.86⁄⁄⁄ �0.78⁄⁄⁄ �1.19⁄⁄⁄ �1.14⁄⁄⁄ �2.23⁄⁄⁄ �1.24 Slope of the yield curve 0.00 0.10⁄⁄ 0.00 �0.01 0.01 0.03 �0.02 0.00 Size 0.02 �0.88⁄⁄ �0.02 0.05 �0.01 �0.20 0.11 �0.01 Non-performing loans/total loans �0.33 �5.87⁄⁄⁄ 0.35 �0.51 �0.06 0.14 4.51 0.05 Funding gap 0.00 �0.61⁄⁄ 0.03 �0.02 0.00 �0.18 0.09⁄⁄⁄ �0.04 Funding gap⁄ Foreign dummy 0.04⁄⁄ – 0.02 �0.27 0.04⁄⁄ �0.21 �0.05⁄⁄⁄ 0.12 Foreign dummy – – – – – – – – Tier 1 ratio �0.02 �3.91 0.50⁄ �0.20 0.11 �0.53 0.27 0.10 ROE 0.01 0.45 0.00 �0.02 0.05 0.19 �0.05 0.02 Bank-specific fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Time-specific fixed effects No No No No No No No No

Results from first-stage of 2SLS-IV estimation Results from first-stage of 2SLS-IV estimation Endog. var.: off-balance-sheet duration gap Instrument: off-balance-sheet duration gap(�1) 0.52⁄⁄⁄ 0.71⁄⁄⁄ 0.45⁄⁄⁄ 0.53⁄⁄⁄ 0.43⁄⁄⁄ 0.54⁄⁄⁄ 0.28⁄ 0.41⁄⁄ F-statistic for weak identification test 42.23 17.10 49.92 19.00 20.76 14.80 5.33 10.74

N. of observations 427 21 196 210 427 63 77 287 N. of banks 64 3 28 33 64 9 11 44 Uncentred R-squared 0.32 0.91 0.33 0.26 0.87 0.92 0.76 0.89

⁄, ⁄⁄, ⁄⁄⁄ denote the statistical significance of the coefficient at the 10%, 5% and 1% level, respectively.

500 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

We check the robustness of our results with respect to two changes in the ‘‘non-core’’ fraction of total overnight deposits: (1) decreasing it to 5%, which implies a substantial lengthening of the on-balance-sheet duration gap; (2) increasing it to 50% which implies a severe shortening of the on-balance-sheet dura- tion gap. The first case is in line with the view that overnight deposits represent a core source of banks’ funding and should be treated as long-term liabilities. Conversely, the second case is consistent with the view that these deposits are subject to withdrawal at any time and should be regarded as

shorter-term liabilities. Increased competition in the deposit market and the spectre of bank runs in the most acute phases of the recent financial crisis could have made the latter scenario more compelling.

Fig. 4 and Table 5 show that the Italian banking system’s overall interest rate risk exposure remains limited. At the same time, however, changing the assumption on overnight deposits has sig- nificant effects on the panel composition. In particular, the shorter the duration of overnight deposits, the larger the number of liability-sensitive banks. On the contrary, the longer the duration

Panel BPanel A

6.5

6.7

6.9

7.1

7.3

7.5

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1

7-year government bond yield spread (rhs)

Residual life of Italian government debt

-8.0

-6.0

-4.0

-2.0

0.0

2.0

4.0

6.0

2008H2 2009H1 2009H2 2010H1 2010H2 2011H1 2011H2 2012H1

benchmark case

adjusted for decoupling in 7-year interest rates

Fig. 6. Panel A shows the development of the residual life of government bonds held by Italian banks (solid line) expressed in years (l.h.s.) and the spread between the 7-year Italian government bond yield and the euro interest rate swap rate of the same maturity (histograms) expressed in percentage points (r.h.s.). Panel B compares the on- balance-sheet duration gap calculated in the standard case of a parallel shift of the yield curve by 200 basis points with the on-balance-sheet duration gap that would arise assuming an in interest rates shift by 200 basis points at all maturity but the 7-year maturity, for which the shock also includes the sovereign spread.

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 501

of overnight deposits, the larger the number of the group of ‘‘other banks’’.

In Table 6 we present the estimated coefficients for panel regres- sions based on these alternative measures for the on-balance-sheet duration gap computed under the assumption of, respectively, a shorter and a longer duration of overnight deposits. The result of a negative correlation between the on-balance-sheet and the off-balance-sheet duration gap is very robust and the estimated coefficients are broadly similar in magnitude to those obtained under the standard assumption recommended by the Basel Committee for all bank categories. All the evidence, as far as asset-sensitive banks and ‘‘other banks’’ are considered, is also very robust. Nevertheless, when we shorten the duration of overnight deposits, the coefficients capturing correlation among the different risks for liability-sensitive banks lose statistical significance. These are the only results that seem to be sensitive to the treatment of overnight deposits in the computation of the interest rate risk.

7.2. Alternative scenarios for the term structure of interest rates

According to the regulatory methodology, the computation of the interest rate risk is based on a parallel shift in the yield curve. Nonetheless, variations in market interest rates are often associ- ated to changes in the slope of the yield curve. Typically, in response to a monetary policy tightening the long-term rate rises less than the short-term rate, resulting in a downward sloping yield curve (e.g. Evans and Marshall 1998; Haldane and Read 2000). However, as also noted in Cœuré (2013), exit strategies from accommodative unconventional monetary policies could instead be associated with a steepening of the yield curve as expectations of low short-term rates reverse and central banks reduce their holdings of long-term securities.

In this section we consider alternative scenarios characterized by non-parallel shifts of yield curve. In particular, we examine two cases: (1) an increase in interest rates that results in a down- ward sloping yield curve (the short-term rates rise by 200 basis point at maturities up to one year while longer-term rate by 100 basis points); (2) an increase in interest rates that steepens the yield curve at longer maturities (the short-term rates rise by 200 basis point while longer-term rate by 300 basis points). The weighting schemes and modified durations used for these exer- cises are reported in column (C) of Table 1.

Fig. 5 and Table 5 suggest that the interest rate shock that flat- tens the yield curve does not have substantial effect on the overall interest rate risk exposure, which remains, on average, limited. As for the consequences in terms of the bank panel composition, we find an increase in the number of liability-sensitive banks when assuming a downward sloping term structure of interest rates. The estimated regressions based on the scenario of a downward sloping yield curve and under the assumption of an upward sloping yield curve are shown in Table 7. Overall, our main results on the relation between on-balance and off-balance interest rate risk exposure appear to be robust. However, we find some differences for liability-sensitive banks: it seems that they also carried out a hedging strategy.

Regarding the integrated management of different risks, the picture is somewhat different compared to our benchmark results. In the upward sloping scenario, asset sensitive banks, when facing higher credit and/or liquidity risk, appear to adjust their exposures in a way that amplifies the gain they would hold on balance if interest rate decrease and that reduces the loss they would get off balance in this same scenario. In the downward sloping scenar- io, liability sensitive banks, when facing higher credit and/or liquidity risk, would adjust their exposures in a way that instead amplifies the gain they would hold on balance if interest rate increase and that reduces the loss they would get off balance in this same scenario.

Another important issue that may affect our results is that we observed a decoupling between long-term risk-free rates and the government bond yields, due to the sudden increase in the sover- eign risk during the financial crisis. Since some balance-sheet items are based on the latter while others are more closely indexed to the former, we have a violation of the standard assumption of a paral- lel shift in the term structure of interest rates, thus affecting the interpretation of the duration gap as a measure of interest rate risk. Fig. 6 shows that, during the crisis, the average residual life of government securities held by Italian banks has been about 7 years. In the same vein, the sovereign spread at this maturity, as captured by the difference between the 7-year Italian govern- ment bond yields and the swap rates of the corresponding matu- rity, which may present a good measure of the disconnect, increased considerably.

We, therefore, compute an alternative measure of the on-balance-sheet duration gap by assuming that a shock in interest rates by 200 basis points is transmitted to a larger extent at the

Table 8 Management of interest rate risk exposure of Italian banking groups: the effects of allowing disconnect between risk-free rates and government bond yields. The estimation is based on half-yearly observations spanning the period 2008H2–2012H1. L–S banks and A–S banks stand respectively for liability-sensitive banks and asset-sensitive banks. For the definition of these three categories of bank see note to Table 2. For 2SLS-IV estimations we report all the estimated coefficients of the second-stage regression and the estimated coefficient of the solely instrument computed in the first-stage regression. The F-statistic for weak identification test is also reported in the lower part of each panel. All estimations use Newey–West heteroskedasticity- and autocorrelation-consistent standard errors.

Panel A

Dependent variable: All banks L–S banks A–S banks Other banks Off-balance-sheet duration gap (1) (2) (3) (4)

Results from second-stage of 2SLS-IV estimation

Explanatory variables: On-balance-sheet duration gap (predicted) �0.65⁄⁄⁄ �0.83⁄⁄⁄ �0.34⁄⁄⁄ �0.64⁄⁄⁄ Slope of the yield curve 0.01 0.02 0.00 0.01 Size �0.06 �0.15 0.00 �0.06 Non-performing loans/total loans 0.13 �0.68 1.07⁄⁄ 0.08 Funding gap �0.02 �0.53⁄⁄ 0.03⁄⁄⁄ �0.06⁄⁄ Funding gap⁄ Foreign dummy 0.04⁄⁄⁄ 0.28 �0.01⁄⁄⁄ 0.23⁄ Foreign dummy – – – – Tier 1 ratio 0.10 �0.63 �0.07 0.05 ROE 0.11⁄⁄ 0.01 �0.03 0.07 Bank-specific fixed effects Yes Yes Yes Yes Time-specific fixed effects No No No No

Results from first-stage of 2SLS_IV estimation

Endogenous variable: on-balance-sheet duration gap Instrument: on-balance-sheet duration gap(�1) 0.41⁄⁄⁄ 0.49⁄⁄⁄ 0.32⁄⁄⁄ 0.39⁄⁄⁄ F-statistic for weak identification test 19.71 8.81 13.56 9.93

N. of observations 427 56 105 266 N. of banks 64 8 15 41 Uncentred R-squared 0.78 0.85 0.54 0.82

Panel B

Dependent variable: All banks L–S banks A–S banks Other banks On-balance-sheet duration gap (1) (2) (3) (4)

Results from second-stage of 2SLS-IV estimation

Explanatory variables: Off-balance-sheet duration gap (predicted) �1.22⁄⁄⁄ �1.16⁄⁄⁄ �1.48⁄⁄⁄ �1.28⁄⁄⁄ Slope of the yield curve 0.01 0.03⁄⁄ 0.00 0.00 Size �0.03 �0.20 �0.05 �0.04 Non-performing loans/total loans �0.11 �0.72 1.57⁄ �0.17 Funding gap �0.01 �0.64⁄⁄ 0.04⁄⁄ �0.06 Funding gap⁄ Foreign dummy 0.05⁄⁄ 0.32 0.00 0.21 Foreign dummy – – – – Tier 1 ratio 0.05 �1.04 0.13 0.04 ROE 0.08 0.00 0.00 0.03 Bank-specific fixed effects Yes Yes Yes Yes Time-specific fixed effects No No No No

Results from first-stage of 2SLS_IV estimation

Endogenous variable: off-balance-sheet duration gap Instrument: off-balance-sheet duration gap(�1) 0.52⁄⁄⁄ 0.65⁄⁄⁄ 0.63⁄⁄⁄ 0.49⁄⁄⁄ F-statistic for weak identification test 42.2 23.61⁄⁄⁄ 25.88⁄⁄ 25.74⁄⁄⁄

N. of observations 427 56 105 266 N. of banks 64 8 15 41 Uncentred R-squared 0.78 0.88 0.64 0.79

⁄, ⁄⁄, ⁄⁄⁄ denote the statistical significance of the coefficient at the 10%, 5% and 1% level, respectively.

502 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

7-year maturity (i.e. by an amount of 200 basis points plus the sovereign spread occurred at each point in time). As shown in Table 8, under this scenario the picture is somewhat different com- pared to the benchmark case. In particular, liability-sensitive banks appear to manage actively both liquidity and credit risk, while asset-sensitive banks seem to respond only to variations in liquid- ity risk. The former group, increases the gain it would get off-balance if interest rate decrease and reduce the loss on balance in the same scenario. The latter group, amplifies the gain they would get off-balance if interest rate increase.

7.3. Non-stationarity of the variables

An important concern may be that some variables used in the regressions may be non-stationary in the short sample period we

considered, thus generating spurious results. We did not perform standard unit root tests, given their low power in very short sam- ple period. However, a visual inspection of Fig. 2 suggests that this does not seem to be the case for the interest rate risk indicators. The only variables that exhibit some systematic trends are the credit quality indicator, due to the significant increase in non-performing loans during the crisis, and the core tier 1 ratio, due to banks’ recapitalisation needs imposed by the changes in the regulatory framework.

In the light of these considerations, we perform a robustness check by repeating all the above analyses with these two variables transformed in first-differences to rule out potential problems of spurious regressions. As shown in Table 9, the esti- mated coefficients remain very similar for all bank categories. Also in this case, the coefficients capturing correlation among

Table 9 Management of interest rate risk: allowing for non-stationarity in the variables. The estimation of the coefficients is based on half-yearly observations spanning the period 2008H2–2012H1.All variables are expressed in first-differences with the exception of bank size. L–S banks and A–S banks stand respectively for liability-sensitive banks and asset- sensitive banks. For the definition of these three categories of bank see note to Table 2. For 2SLS-IV estimations we report all the estimated coefficients of the second-stage regression and the estimated coefficient of the solely instrument computed in the first-stage regression. The F-statistic for weak identification test is also reported in the lower part of each panel. All estimations use Newey–West heteroskedasticity- and autocorrelation-consistent standard errors.

Panel A

Dependent variable: All banks L–S banks A–S banks Other banks Off-balance-sheet duration gap (1) (2) (3) (4)

Results from second-stage of 2SLS-IV estimation

Explanatory variables: On-balance-sheet duration gap (predicted) �0.27⁄ �0.09 �0.51⁄⁄⁄ �0.21 Slope of the yield curve 0.01⁄ 0.01 0.02 0.01 Size �0.11⁄⁄ 0.00 �0.33⁄⁄ �0.12⁄⁄

Non-performing loans/total loans (First difference) 0.10 1.20⁄ �2.18⁄ 0.32 Funding gap �0.03⁄⁄ 0.00 �0.27 �0.08⁄⁄ Funding gap⁄ Foreign dummy 0.04⁄⁄⁄ 0.00 �0.09 0.78⁄⁄⁄ Foreign dummy – – – – Tier 1 ratio (First difference) �0.14 �0.04 �0.26 �0.09 ROE 0.10⁄ 0.00 0.24 0.04 Bank-specific fixed effects Yes Yes Yes Yes Time-specific fixed effects No No No No

Results from first-stage of 2SLS-IV estimation

Endogenous variable: On-balance-sheet duration gap Instrument: on-balance-sheet duration gap(�1) 0.37⁄⁄⁄ 0.32⁄⁄⁄ 0.62⁄⁄⁄ 0.35⁄ F-statistic for weak identification test 12.80 11.33 18.46 6.25

N. of observations 429 116 53 260 N. of banks 66 18 9 39 Uncentred R-squared 0.28 0.28 0.21 0.34

Panel B

Dependent variable: All banks L–S banks A–S banks Other banks On-balance-sheet duration gap (1) (2) (3) (4)

Results from second-stage of 2SLS-IV estimation

Explanatory variables: Off-balance duration gap (predicted) �0.90⁄⁄⁄ �0.71 �1.56⁄⁄⁄ �0.67⁄⁄⁄ Slope of the yield curve �0.02 0.02 0.03 �0.03⁄⁄ Size 0.01 �0.14 �0.49⁄ 0.06 Non-performing loans/total loans (First difference) �1.47 �0.56 �4.48⁄ �1.88⁄ Funding gap 0.00 0.00 �0.59 �0.01 Funding gap⁄ Foreign dummy 0.05⁄⁄ 0.04 �0.03 0.34 Foreign dummy – – – – Tier 1 ratio (First difference) 0.00 0.13 �0.20 0.08 ROE �0.01 0.06 0.35 �0.10 Bank-specific fixed effects Yes Yes Yes Yes Time-specific fixed effects No No No No

Results from first-stage of 2SLS-IV estimation

Endogenous variable: off-balance duration gap Instrument: one-period lag of off-balance duration gap 0.50⁄⁄⁄ 0.64⁄⁄⁄ 0.72⁄⁄⁄ 0.50⁄⁄⁄

F-statistic for weak identification test 27.71 15.76 41.22 17.81

N. of observations 429 116 53 260 N. of banks 66 18 9 39 Uncentred R-squared 0.24 0.33 0.61 0.19

⁄, ⁄⁄, ⁄⁄⁄ denote the statistical significance of the coefficient at the 10%, 5% and 1% level, respectively.

L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504 503

the different risks for liability-sensitive banks loose statistical significance.

8. Conclusions and policy implications

We have used a unique panel dataset to document the exposure to interest rate risk of Italian banking groups during the recent financial crisis period. Unlike what is typically done in previous studies, we have measured banks’ exposure to interest rate risk using the duration gap approach proposed by the Basel Committee on Banking Supervision rather than a maturity gap indicator.

The Italian banking system exhibited a limited exposure to interest rate risk during the period under review, well below the

20% regulatory alert threshold. We have shed light on how Italian financial intermediaries managed such a risk by changing their on-balance-sheet exposure as well as by relying on interest rate derivatives. Our econometric results indicate that on average banks have used these two instruments as substitute meaning that the on-balance-sheet interest rate risk and the off-balance-sheet exposure have been manoeuvred to partially offset each other rather than to pursue an enhancing strategy and boost the poten- tial gains that might have occurred had interest rates increased. Overall, these findings are reassuring from a monetary policy view- point. Future changes in the ECB official policy rates should not represent a major concern for the Italian banking system.

Regarding the correlation among different financial risks, our results suggest a substantial heterogeneity in banks’ risk

504 L. Esposito et al. / Journal of Banking & Finance 59 (2015) 486–504

management practices. On the one side, we find that one third of the banks, namely the smaller ones and those characterized by a traditional business activity, followed an integrated risk- management approach aiming at counterbalance the interest rate risk and the credit risk. On the other side, the majority of Italian banks, which comprises the largest banking groups, tended to enhance the gains from an increase in interest rates also in the face of a widening of the funding gap.

One possible interpretation is that the extraordinary liquidity injections implemented by the ECB during the crisis acted as sub- stitutes for banks’ wholesale funding, thus, de facto offsetting banks’ liquidity risk and inducing largest intermediaries to take on more interest rate risk, especially by means of large purchases of government bond yields. In this regard, moral hazard consider- ations related to the ‘‘too-big-to-fail’’ argument remain a matter of concern for policymakers when they evaluate the design of new unconventional measures and their implications for financial sta- bility. Given the high correlation between the liquidity risk and the interest rate risk, the new prudential rules on long-term liquid- ity position (e.g. the standards measured by the Net Stable Funding Ratio) endorsed by the Basel Committee on Banking Supervision are particularly welcome since they will be particularly effective in containing the duration mismatch between assets and liabilities. Our result overall also provide empirical support to the develop- ment of theories aiming at modeling the various financial risks jointly. The recent contributions in Drehmann et al. (2008) and Alessandri and Drehmann (2010) where credit and interest rate risk in the banking book are modeled jointly represent a step forward.

References

Akella, S.R., Chen, S., 1990. Interest rate sensitivity of bank stock returns: specification effects and structural changes. J. Financial Res. 13, 147–154.

Alessandri, P., Drehmann, M., 2010. An economic capital model integrating credit and interest rate risk in the banking book, Bank of England Working Paper no. 388.

Angbazo, L., 1997. Commercial bank net interest margins, default risk, interest-rate risk, and off- balance-sheet banking. J. Bank. Finance 21, 55–87.

Armeanu, D., Balu, F., Obreja, C., 2008. Interest rate risk management using duration gap methodology. Theor. Appl. Econ. 1, 3–10.

Au Yong, H., Faff, R., Chalmers, K., 2009. Derivative activities and Asia-Pacific banks’ interest rate and exchange rate exposures. J. Int. Financial Markets Inst. Money 19, 16–32.

Ballester, L., Ferrer, R., Gonzales, C., Soto, G.M., 2009. Determinants of interest rate exposure of the Spanish banking industry. Working Paper of University of Castilla-La Mancha.

Bank of Italy, 2006. New Regulations for the Prudential Supervision of banks, n. 263. Bank of Italy, 2012. Applicazione delle disposizioni di vigilanza in materia di

organizzazione e governo societario delle banche, January. Bartiloro, L., Carpinelli, L., Finaldi Russo, P., Pastorelli, S., 2012. Access to credit in

times of crisis: measures to support firms and households. Bank of Italy, Occasional Papers 111.

Basel Committee on Banking Supervision, 2004. Principles for the management and supervision of interest rate risk. Bank for International Settlements.

Basel Committee on Banking Supervision, 2006. International convergence of capital measurement and capital standards: a revised framework. Bank for International Settlements.

Basel Committee on Banking Supervision (2008). Range of practices and issues in economic capital modeling, Bank for International Settlements.

Begenau, J., Piazzesi, M., Schneider, M., 2013. Banks’ risk exposures. Stanford University, Mimeo.

Brewer, E., Jackson, W.E., Moser, J.T., 1996. Alligators in the swamp: the impact of derivatives on the financial performance of depository institutions. J. Money Credit Bank. 28, 482–497.

Cecioni, M., Ferrero, G., Secchi, A., 2011. Unconventional monetary policy in theory and in practice? Bank of Italy, Occasional Paper Series, No. 102.

Choi, J., Elyasiani, E., 1997. Derivative exposure and the interest rate and exchange rate risks of U.S. banks. J. Financial Serv. Res. 12, 267–286.

Cœuré, B., 2013. Where to exit to? Monetary policy implementation after the crisis. In: Speech at the 15th Geneva Conference on the World Economy: ‘‘Exit strategies: time to think about them’’, Geneva.

Dewatripont, M., Tirole, J., 1994. The Prudential Regulation of Banks. Massachusetts, MIT Press, Cambridge.

Deshmukh, S.D., Greenbaum, S.I., Kanatas, G., 1983. Lending policies of financial intermediaries facing credit and funding risks. J. Financ. 38 (3), 873–886.

Diamond, D.W., 1984. Financial intermediation and delegated monitoring. Rev. Econ. Stud. 51 (3), 393–414.

Drehmann, M., Sorensen, S., Stringa, M., 2008. The integrated impact of credit and interest rate risk on banks: an economic value and capital adequacy perspective, Bank of England Working Paper no. 339.

English, W.B., 2002. Interest rate risk and bank net interest margins. BIS Q. Rev., 67– 82, December.

Esty, B.C., Tufano, P., Headley, J., 1994. Banc one corporation: asset and liability management. Harvard Business School Case, 294–079.

European Banking Authority, 2011. Guidelines on Internal Governance (GL44). September.

European Central Bank, 2011. Financial Markets in Early August 2011 and the ECB’s Monetary Policy Measures. ECB Monthly Bulletin, September.

Evans, C., Marshall, P., 1998. Monetary policy and the term structure of nominal interest rates: Evidence and theory. Carnagie-Rochester Conf. Public Policy 49, 53–112.

Fiori, R., Iannotti, S., 2006. Scenario based principal component value-at-risk: an application to Italian bank’s interest rate risk exposure. J. Risk 3, 63–99.

Flannery, M.J., James, C.M., 1984. The effect of interest rate changes on the common stock returns of financial institutions. J. Finance 39, 1141–1153.

Fraser, D.R., Madura, J., Weigand, R.A., 2002. Sources of bank interest rate risk. Financial Rev. 37, 351–368.

Froot, K.A., Scharfstein, D.S., Stein, J.C., 1993. Risk management: coordinating corporate investment and financing policies. J. Finance 48, 1629–1658.

Froot, K.A., Stein, J.C., 1998. Risk management, capital budgeting, and capital structure policy for financial institutions: an integrated approach. J. Financ. Econ. 47 (1), 55–82.

Gennotte, G., Pyle, D., 1991. Capital controls and bank risk. J. Bank. Finance 15, 805– 824.

Gorton, G.B., Rosen, R.J., 1995. Banks and derivatives. In: Bernanke, B.S., Rotemberg, J.J. (Eds.), NBER Macroeconomics Annual. The MIT Press, Cambridge, pp. 299– 349.

Gunther, J.W., Siems, T.F., 1995. The likelihood and extent of banks’ involvement with interest-rate derivatives as end-users. Working paper, Federal Reserve Bank of Dallas.

Haldane, A., Read, V., 2000. Monetary policy surprises and the yield curve, Bank of England, Working Paper, No. 106.

Hirtle, B.J., 1997. Derivatives, portfolio composition, and bank holding company interest rate risk exposure. J. Financial Serv. Res. 12, 243–266.

Sinkey J.F., Jr., Carter D., 1994. The derivatives activities of U.S. commercial banks, Proceedings from Federal Reserve Bank of Chicago 29.

Macaulay, F.R., 1938. The Movements of Interest Rates, Bond Yields and Stock Prices in the United States Since 1856. National Bureau of Economic Research, New York.

Memmel, C., 2011. Banks’ exposure to interest rate risk, their earnings from term trans-formation, and the dynamics of the term structure. J. Bank. Finance 35, 282–289.

Purnanandam, A., 2007. Interest rate derivatives at commercial banks: an empirical investigation. J. Monetary Econ. 54, 1769–1808.

Reichert, A., Shyu, Y., 2003. Derivative activities and the risk of international banks: a market index and var approach. Int. Rev. Financial Anal. 12, 489–511.

Saporoschenko, A., 2002. The sensitivity of Japanese bank stock returns to economic factors. An examination of asset/liability differences and main bank status. Global Finance J. 13, 253–270.

Sartoris, W.L., 1993. The term structure of interest rates and the asset and liability decisions of a financial intermediary. J. Econ. Bus. 45, 129–142.

Saunders, A., Strock, E., Travlos, N.G., 1990. Ownership structure, deregulation, and bank risk taking. J. Finance 45, 643–654.

Schrand, C.M., 1997. The association between stock-price interest rate sensitivity and disclosure about derivative instruments. Account. Rev. 72, 87–109.

Simons, K., 1995. Interest Rate Derivatives and Asset-Liability Management by Commercial Banks. New England Economic Review, Federal Reserve Bank of Boston, January/February, 17–28.

Staiger, D., Stock, J.H., 1997. Instrumental variables regression with weak instruments. Econometrica 65, 557–586.

Stein, J.C., 1998. An adverse-selection model of bank asset and liability management with implications for the transmission of monetary policy. RAND J. Econ. 29, 466–486.

Zhao, F., Moser, J., 2009. Use of derivatives and bank holding companies interest- rate risk. Bank. Finance Rev. 1, 51–62.