Are banks affecting their risk with derivatives? Evidence from the U.S. banking sector

Are banks affecting their risk with derivatives?

Evidence from the U.S. banking sector

Leander Konings* Program: MSc. Finance Supervisor: Prof. dr. W. Bessler

July 19th 2019

Keywords: Finance, Derivatives, Banking, Risk, Foreign Exchange Risk, Credit Risk, Real Estate Risk

Word count: 12,077 (excluding Appendix)

*Faculty of Economics and Business, University of Groningen, PO Box 800, 9700AV Groningen, the Netherlands. Student number: S3264297; e-mail: leander_konings@live.nl

2

1.

Introduction

The use of financial derivatives by banks in the U.S. has increased significantly over the last decades. Up until the economic crisis from 2007-2009, the derivatives market had a notional value outstanding of about four times the notional value of the derivatives market in 2000. After this economic crisis the market has been in decline, but it seems like the bottom has been reached, judging by the graph in Figure 1. Also, the International Swaps and Derivatives Association (ISDA) mention that the US derivatives is growing again. From January 2018 until July 2018, the notional value of derivatives increased by 10% relative to the same period in 2017. A trend that has been quite steady of the past years (ISDA, 2018).

Figure 1: Notional values of financial derivatives held by U.S. BHCs in $. Interest rate

derivatives on the right axis. Other derivatives on the right axis.

Note: IRD stands for interest rate derivatives. FXD stands for foreign exchange derivatives. CD stands for Credit derivatives. Source: yearly data from FR Y-9C report, sample period 2000q4:-2018q4

Within this market interest rate derivatives are most commonly used with a notional value outstanding of almost 250 trillion U.S. dollar in 2009, roughly 84% of the total derivatives market. To put the size of the derivatives market in perspective, the total value of the U.S. equity markets at the last trading day of 2018 was roughly 34 trillion U.S. dollar (WFE, 2019).

That the derivatives market has been in growing rapidly, until the financial crisis, is not surprising. Prominent economists found that derivatives perform a vital role in financial markets, for creating value (Greenspan, 1999) and for providing a proxy for correctly pricing other financial instruments (Trichet, 2007). The crisis has changed the perspective on the

3

derivatives market. A study into the U.S. derivatives market showed that derivatives can have negative and contagious effects on financial markets (FSB, 2010). Especially the over-the-counter market was criticized for being opaque and its interconnectedness of over-the-counterparties (FSB, 2010). Credit derivatives have been pointed at as the cause of the financial crisis (Nijskens and Wagner, 2011)

Derivatives are financial products that derive their value from the performance of another asset. Derivative’s main purpose is to reduce risk exposures of market participants by hedging (Moles, Shin and Shiu, 2010). Banks are dominant participants in the derivatives market as they take long and short positions and often act as market makers (Infante, Piermattei, Santioni and Sorvillo, 2018).

4

Figure 2: Financial derivatives held for trading and hedging in U.S. Dollars

Note: Source: yearly data from FR Y-9C report, sample period 2000q4:-2018q4

This article follows the following structure. Part 2 reviews the literature on prior research and develops the hypotheses. Part 3 explains the data and the methodology. Part 4 discusses the empirical results followed by the conclusion in part 5.

2.

Literature review

The first section covers the particular risk banks face. The second section goes into the details of the derivatives market, the different derivatives that banks can employ, and the basics of how these derivatives work. The third section focuses on why banks may use derivatives. In the fourth and last section earlier studies on the subject are discussed and the hypotheses for this study are stated.

2.1 Bank risks

Banks, like other companies, are exposed to a number of risks. The foremost risks banks face are: credit risk, interest rate risk, real estate risk, and foreign exchange risk (Bessler, Kurmann and Nohel, 2015). The interest rate risk for banks originates from their core business: transforming short-term deposits into long-term loans. This means that there will be a maturity mismatch between a bank’s funding and their loan operations (Hull, 2015b). A common measure of interest rate risk is the duration gap, the difference between the assets’ duration and liabilities’ duration. For banks, this gap is usually positive, meaning

0 2T 4T 6T 8T 10T 12T 14T 16T 18T 0 50T 100T 150T 200T 250T 300T 2000 2003 2006 2009 2012 2015 2018

5

the assets have a longer duration than the liabilities. This is just one part of interest rate risk. A more important aspect of interest rate risk is the repricing risk. Repricing risk appears from the different sensitivities to changes in interest rates between assets and liabilities of a bank (Genay and Podjasek, 2014). There is a negative correlation between market values of interest rate based financial products (bonds, loans, deposits, etc.) and the interest rate in the market. This means that an increase of the interest rate will lead to a decrease of the market value of interest rate based financial products, which will affect the equity value of a bank (Bessler and Kurmann, 2014). Figure 3 shows a normal upward sloping yield curve. It shows that longer maturities have higher yields than shorter maturities. However, yields of different maturities do not necessarily shift parallel. It is common for yields to have non-parallel shifts, which changes the shape of the yield curve. These changes in the yield curve directly affect the prices of the loan portfolio and deposits of banks.

Figure 3: Maturity spread and credit spread

Note: This figure shows how yield spreads exist between bonds due to their credit rating, called the

bond yield spread. And the difference of maturity within the same credit rating, in this study called the maturity spread.

Credit risk is another important risk banks face. Banks provide loans and that makes them vulnerable to credit risk, because lending activities are inherently risky. Credit risk consists of two types of risk. One is counterparty risk and the other is loan default risk (Hull, 2015a). Loan default risk is when a borrower fails to pay the debt they owe.

6

Counterparty risk is the risk that the counterparty fails to meet its contractual obligation. The default of a counterparty can be isolated, but there is a chance that the default affects other borrowers, the credit contagion (Jorion and Zhang, 2009). Suppose that bank A has loans outstanding with firm B and C. Firm B and C also interact with each other, and suppose that B delivers products to C. Firm C defaults and besides not being able to the debt owed to bank A, it also can no longer pay firm B for delivered products. Firm B is highly dependent on firm C and defaults as well, leaving bank A with two non-performing loans. In Figure 3 the difference in yield is shown between a AAA-bond and BBB-bond. This bond yield spread is based on the perceived risk by investors. Because of this credit risk banks charge a risk premium on their credit to be compensated for the risk they take (Bessler and Kurmann, 2014). Yield spreads are often examined to derive the default probabilities as they change over the business-cycle (Fama and French, 1989). One particular kind of credit risk is the risk on sovereign debt. Credit risk does not only apply to firms but also to countries so that sovereign debt has also contains credit risk. Sovereign risk has become a much more significant factor during the financial crisis of 2007-2009. When countries defaulted on their bonds it became clear that banks had great exposure to sovereign credit risk. During the financial crisis of 2007-2009, this risk became increasingly more important as the bond yield spreads between German debt and Euro peripheral area sovereign debt diverged to pre-euro levels (Bessler, Leonhardt and Wolff, 2016). These sovereign credit risks also increased due to bank bailouts (Acharya, Drechsler and Schnabl, 2014). According to Bessler, et. al (2016) sovereign risk was mainly a risk for European banks and of lesser importance for U.S. banks. Therefore, sovereign risk is not covered in the analytical part of this study.

Thirdly we examine foreign exchange risk. This is not a unique risk for banks, as many companies are exposed to changes in the value of currencies. There is no evidence that foreign exchange risks are any different for financial institutions than for other types of industries (Bessler et al, 2015). On top of that, there is a strong correlation between the liquidity of foreign exchange markets and equity markets, which makes isolating foreign exchange risks difficult. Both financial institutions and other companies are usually able to hedge this risk effectively. Therefore, foreign exchange risk can be considered to be a market wide risk, rather than a banking risk.

7

the real estate, which is used as collateral (Bessler et al, 2015). The Financial crises have shown that changes in real estate values directly affect the profitability on the mortgage portfolio (Hendershott and Kane, 1992; Huizinga and Laeven, 2012; Bessler et al., 2015).

2.2 Derivatives and their uses

Derivatives are financial products that derive their value from the value of another asset, called the underlying asset. (Hull, 2015b). The underlying asset can be many things, for example: Stocks, an interest rate, the weather, et cetera. Derivatives are contracts between two parties in which they agree to pay each other, depending on what position the parties take, and how the underlying asset performed. The reason why derivatives have been ‘invented’ was mainly for risk reduction purposes, called hedging. However, they are also used for speculation and arbitrage (Hull, 2015b). Derivatives play an important role in our financial markets and are used widely by many firms, including banks. The vast majority of derivatives is traded OTC (Over-The-Counter) rather than on exchanges. The most common forms of derivatives are options, forwards/futures, and swaps. In Figure 4 shows a simplified representation is given on the these derivatives and whether they are traded OTC or on exchanges.

The most straight forward derivatives are futures and forwards. With these kind of derivatives, you agree with a counterparty to either buy or sell an asset at some point in the future for an agreed upon price. This way someone can lock in today’s price and hedge against the price difference of the underlying asset between now and delivery. The main difference between a forward and a future is that forwards a solely traded over-the-counter, whereas futures are only traded on exchanges. Therefore, forwards can be modified extensively on the type of underlying asset, contract size, delivery date, and delivery month. For futures contracts this is standardized. Which is necessary, because they are traded on exchanges (Hull, 2015b).

8

(buy) the asset when the option is exercised. The seller receives are premium for selling a right to buy (sell) (Hull, 2015b).

The third major derivative category are swaps. A swap is a contract in which two parties agree to exchange cash flows from certain assets. The main three swaps are interest rate swaps (IRS), currency swaps, and credit default swaps (CDS). With an IRS one party agrees to pay a fixed predetermined interest rate on a notional amount to another party. In return the other party pays a floating interest rate to the first party. Firms may do this if there is an advantage of ‘swapping’ their interest rates. With a currency swap two parties exchange both the principal and the interest rate in one currency for another currency. The CDS is like an insurance policy that pays out in case of a credit event. The ISDA considers three types of credit events. The first one is failing to repay the principal .or payout coupon. The second is restructuring of the debt underlying the CDS contract. And thirdly the Moratorium or Repudiation, where a debtor temporarily seizes to make any payments due to financial distress (Berndt, Jarrow and Kang, 2007). The buyer of the CDS will pay a premium to the seller, which is known as the CDS spread. Swaps are only traded over-the-counter (Hull, 2015b).

Figure 4: Schematic overview of the types of derivatives and where they are traded.

9

Prior to the financial crisis of 2007-2009 over-the-counter market were largely unregulated and parties were trading bilateral without any centralized administration of all the contracts that are outstanding. After the financial crisis of 2007-2009 regulatory changes have been made. Most over-the-counter trades are now registered at a Central Clearing Party (CCP). These CCPs provide the same service as clearing houses on exchanges. Clearing houses provide a vital role in financial markets, as they make sure the counterparty risk is virtually zero. Anyone that trades derivatives on exchanges needs to clear their trades through a clearing house. This means that the trader has to post collateral, which is called a margin. This process ensures that when one of the trading parties in a contract is losing money, the losses are covered. If a loss is becoming too large the clearing house will ask for additional margin. If the trader cannot provide additional margin the clearing house will automatically close the position to make sure the loss is not greater than the margin posted. Prior to the financial crisis over-the-counter trades were cleared bilateral, using an ISDA master agreement (Hull, 2015b). A visual representation of the bilateral clearing and central clearing is shown in Figure 5.

Figure 5: Bilateral clearing in OTC markets (on the left) and central clearing (on the right)

Figure taken from Options, Futures, and other Derivatives (Hull, 2015b, Figure 18.1, p.388), © Pearson Education Limited, 2018.

2.3 Why firms use derivatives

According to Modigliani and Miller (1958) hedging should not make a difference in the firm’s value in the absence of imperfections in financial markets, because in their theory financial decisions do not affect the value of a firm. However, there is plenty of evidence that whether or not to engage in hedging makes a difference as markets are not perfect. Hedging

10

can result in making future cash flows less volatile by reducing foreign exchange risks, interest rate risks and changes in commodity prices, which in turn makes a corporation more valuable (Clifford, et al., 1993). However, hedging can also reduce profitability, since hedging is costly, both from engaging in derivatives trading and by reducing the risk exposure, which also means reducing the upside potential (Allyanis and Ofek, 2001). Another reason for companies to use derivatives is to lower the costs of financial distress (Smith and Stultz, 1985). There is also evidence that corporations that use derivatives avoid underinvestment problems, increasing the growth potential of a firm (Gay and Nam, 1998). Empirical evidence for value creation through hedging is not conclusive. However, a meta-analysis by Bessler et al. (2019) showed that the use of derivatives for hedging increases a non-financial firm’s Tobin’s Q, a measure of value creation.

Within the non-financial sector the use of derivatives tends to only be done by large and mature corporations (Bartram, Brown and Fehle, 2009; Paligorova and Staskow, 2014). This is because of the complexity of derivatives, as it requires skilled personnel and enough resources to use derivatives to the firms’ advantage.

2.4 Banks and derivatives

Any bank, like a non-financial firm, faces risks. Using derivatives for hedging purposes can decrease risks and it can increase value. Using interest rate derivatives can help a bank manage changes in the yield curve. It can employ interest rate swaps to change the fixed rates on their long term assets to adjustable rates, so a bank is hedged against increasing interest rates. A bank can remove the risk that a borrower defaults on their debt by buying a credit default swap. However, it looks like banks are not using derivatives for just hedging. The vast majority of derivative position do not have hedging as a purpose, looking at Figure 2. So what do banks use derivatives for, other than hedging?

2007-11

2009 central clearing has become more predominant in the over-the-counter market, reducing counterparty risk greatly (Hull, 2015b). Table 1 provides a summary of the use of derivatives by U.S. BHCs. As known from earlier studies, the derivatives market is highly concentrated towards the largest BHCs (Sinkey et al., 2000; Booth, Smith and Stolz, 1984). The 100 largest BHCs are responsible for about 87% of the trade in derivatives. We also see a high concentration towards the largest BHCs; more than 99% of the derivatives trade within the sample falls to the BHCs with total assets of more than 50 billion USD. The largest pool of derivatives are the swap contracts, which are predominantly used for interest rate risk and credit risk. Second are options, which are mainly used for interest rate risk, but also in foreign exchange (Stock market). Third are forwards and fourth come futures. Given that most trade seems to happen with swaps and a substantial amount in forwards, it is no surprise that BHCs mostly trade over-the-counter rather than through exchanges. When we look at the different risks underlying the derivatives we see that the largest part of the derivatives are on interest rates. The vast majority of interest rate derivatives are swap contracts. Then we can see that the same goes for credit derivatives and foreign exchange derivatives, but they are in value a smaller part of the whole market. Some extra attention is drawn to bottom part of the table. We can see here that most of the notional value is allocated towards trading purposes, rather than hedging.

2.5 Earlier research

12

derivatives were indeed the cause for the financial crisis and increase the risk, is ambiguous. Stulz (2010) finds evidence in favor of this conclusion by stating that the use of credit derivatives has been influenced by wrong incentives leading to excessive risk-taking, market manipulations, and creating larger independencies between banks, all contributing to more risk. Rule (2001) has come to similar conclusions by stating that the use of credit derivatives makes capital allocation more efficient, however, transferring this risk to the market and thus separating it directly from the borrower may lead to the incentive of less creditworthiness monitoring and could make restructuring more difficult. Statistical evidence supports the conclusion that banks increase their risk using derivatives (Li and Marinč, 2014; Choi and Elyasiani, 1997; Mayordomo, Rodiguez-Moreno, and Peña, 2014; Huan and Parbonetti, 2017).

Some studies do not support the claim that derivatives increase risk. A study by Calistru (2012) says that it was not the credit derivatives, but the actions by the U.S. Government and Federal Reserve that were responsible for the panic on financial markets. Minton, Stulz, and Williamson (2009) mention that even though derivatives can lead to higher risk, the proportion of credit derivatives is too small to really have an impact on the risk profile of a bank. Cyree, Huang and Lindley (2011) have also not found clear evidence that suggests that the use of derivatives is linked to more risk within banks or in the system. Since there is much ambiguity on whether using derivatives increases or decreases risk of BHCs, this study tries to shed more light on the subject. This leads to following the hypotheses of this study:

H1: The use of interest rate derivatives by BHCs impacts the interest rate risk of BHCs. H2: The use of foreign exchange derivatives by BHCs impacts the foreign exchange risk of BCHs.

13

Table 1: Overview of the use of derivatives by U.S. BHCs (notional amounts in millions U.S. Dollar)

Source: The financial data is for the 4th _{quarter 2006 of the FR Y-9C report}

U.S. BHCs in FR Y-9C _{BHCs in the Sample Largest BHCs (Total assets >25b)}

Value _{Value % of total Value % of total % of sample}

1 _{2 3 4 5 6} Types of contracts Futures _{16,899,077 4,120,756 24.38 4,118,367 24.37} 99.94 Forwards _{12,648,989 11,616,169 91.83 11,594,513 91.66} 99.81 Options _{29,414,123 28,151,248 95.71 28,124,102 95.61} 99.90 Swaps _{91,270,885 87,133,639 95.47 86,595,840 94.88} 99.38 Exchange Traded/OTC Exchange traded _{25,139,342 12,755,374 50.74 12,751,606 50.72} 99.97 Over-the-counter _{119,169,463 118,280,259 99.25 118,194,712 99.18} 99.93

Interest Rate Derivatives

Futures _{4,419,431 3,904,206 88.34 3,901,974 88.29} 99.94 Forwards _{6,328,569 5,724,316 90.45 5,713,245 90.28} 99.81 Options _{23,274,748 21,853,853 93.90 21,845,882 93.86} 99.96 Swaps _{78,382,615 75,845,813 96.76 75,811,948 96.72} 99.96 Forex Derivatives Futures _{45,715 28,730 62.85 28,730 62.85} 100.00 Forwards _{6,118,751 5,836,341 95.38 5,825,765 95.21} 99.82 Options _{3,253,033 3,121,851 95.97 3,121,687 95.96} 99.99 Swaps _{2,568,774 2,407,174 93.71 2,406,649 93.69} 99.98 Credit Derivatives Swaps _{9,239,135 8,377,250 90.67 8,377,243 90.67} 100.00 Options _{15,251 14,963 98.11 14,963 98.11} 100.00 Financial derivatives

Financial derivatives for hedging _{3,749,970 2,637,633 70.34 2,637,633 70.34} 100.00

14 3. Data and Methodology

3.1 Sample data

To investigate and answer the main question I have collected data from bank holding companies (BHC). Most of the U.S. banks are a BHC and depending on the size of their assets, they have to report every quarter to the Federal Reserve detailed information about their balance sheet and income statements, the so-called FR Y-9C reports1_{. These reports are}

publicly available at the Federal Reserve and date back to as far as 1978. This study covers a time period from 2000 until 2018, to be able to cover a solid pre-crisis period and post-crisis period. Next to the financial accounting data, stock market data is obtained. The stock market data is recovered from the Center for Research in Security Prices (CRSP).

For this study, the top 100 largest publicly listed banks (BHC) have been selected, based on total assets, at 31-12-2006. Their main listing must be on one of the US stock exchanges. This period has been chosen because it is prior to the financial crisis of the years 2007-2009. Table 2 shows how many BHCs drop out between 2000-2018 and the reason why it happens. An overview of all the BHCs in the sample is shown in Appendix A.

Table 2: Status of BHCs in sample

Status Amount

Active 61

Mergers 30

Distress and insolvency 9

Total 100

Note: determined by using the delisting codes from the Center for Research in Security Prices (CRSP)

Stock data on the largest 100 BHCs have been obtained from CRSP. On a monthly basis the last price of the month and the number of shares outstanding are collected. The stock data is adjusted for (reversed) stock splits. As shown in Table 1, the derivatives market is highly concentrated towards the largest banks with total assets of at least 25 billion U.S. dollar. Because of this a subsample is drawn consisting of the banks with at least 25 billion U.S. dollar in total assets at 31-12-2006. This way it is possible to see how the largest banks are affected by certain risks and how the use of derivatives impact these risks.

15 3.2 First-stage regression

To analyze whether or not the use of derivatives increases bank risk I use a two-step regression as suggested by Li and Marinč (2014). In the first regressions, we estimate how specific risks affect the stock returns of BHCs. The risk factors used are interest rate risk (IR), foreign exchange risk (FX), credit risk (CR) and real estate risk (REIT).

𝑟𝑖𝑡 = 𝛼𝑖 + 𝛽𝑖𝑟,𝑖𝑡(𝐼𝑅) + 𝛽𝑓𝑥,𝑖𝑡(𝐹𝑋) + 𝛽𝑐𝑟,𝑖𝑡(𝐶𝑅) + 𝛽𝑟𝑒𝑖𝑡,𝑖𝑡(𝑅𝐸𝐼𝑇) + 𝜀𝑖𝑡, (1)

The 𝑟_𝑖𝑡 is the monthly return of the a BHC in the sample. 𝐼𝑅 is the proxy for interest rate risk. It is the percentage change in the difference in yield between a 3-month U.S. treasury bill and a 10-year U.S. government bond. In this study the difference between these yields is called the maturity spread. For credit risk the percentage change of the difference in yield is used between AAA U.S. corporate bonds and BBB U.S. corporate bonds. This spread is called the bond yield spread. For the AAA U.S. corporate bonds yield the ICE Bank of America U.S. Corporate AAA Effective Yield index is used, and for the BBB U.S. corporate bonds yield the ICE Bank of America U.S. Corporate BBB Effective Yield index is used.

FX denotes foreign exchange risk. The rate of change of the U.S. Dollar Nominal Broad index is used to model foreign exchange risk. This is an index that compares the value of a U.S. Dollar to a broad basket of other currencies such as the Japanese yen, euro, and pound sterling. The fourth factor is the real estate risk proxy, based on the return of the North American REIT Index, in line with Bessler and Kurmann (2014) . The 𝛼 is the constant error term and 𝜀 is the random error. All the data is collected on a monthly basis. In Figure 6 the different variables are graphed over time. For the risk factors interest rate, credit, and foreign exchange the necessary data is obtained through Datastream. The North American REIT Index data is obtained from the organization’s publicly available price data. In Appendix B you can find a detailed description of all the variables.

16

be the most static of the three,. This window has 60 observations per window. Model (1) is adjusted for heteroskedasticity by using White standard errors.

Figure 6: Plots of the explanatory variables over time

Other issues that may arise with the data are heteroskedasticity and autocorrelation. A number of tests are run to capture any of these issues and proper measures are taken to alleviate or 0.00 2.00 4.00 6.00 8.00 1999 2003 2007 2011 2015 An n u aliz ed Ra te

3-Month US T-Bill rate vs. 10-year US Government Bond Rate

0.00 2.00 4.00 6.00 8.00 10.00 12.00 1999 2003 2007 2011 2015 Ann u alzie d Rat e

US AAA Corporate Bond Yield vs. US BBB Corporate Bond Yield

60 70 80 90 100 110 120 1999 2003 2007 2011 2015 In d ex US FX 0 2,000 4,000 6,000 8,000 1999 2003 2007 2011 2015 In d ex

NAREIT Total Real Estate Index

-1.00 0.00 1.00 2.00 3.00 4.00 5.00 1999 2003 2007 2011 2015 Sp re ad Maturity Spread 0.00 1.00 2.00 3.00 4.00 5.00 1999 2003 2007 2011 2015 Spre ad Credit Spread

Note: Timeseries of the explanory variables of the first-stage regression model between December

1999 and December 2018. The upper two plots show the 3-month U.S. treasury bill rate versus the U.S. 10-year government bond rate on the left, and on the right the yields on U.S. Corporate AAA bonds and U.S. Corporate BBB bonds are shown. The middle plots show the spread between the 3-month U.S. treasury bill and the 10-year U.S. government bond on the left, and on the right the spread between U.S. Corporate AAA bonds and U.S. Corporate BBB bonds is shown. In the lower two graphs the USD nominal broad index and the FTSE/NAREIT index are shown.

10-year maturity

3-month maturity

BBB

17 3.3 Second-stage regression

In the second stage, we regress the estimated coefficients of different risk factors on different accounting variables from the FR Y-9C reports with a panel data regression. The work of Li and Marinč (2014) is the foundation of this part of the analysis. They base their regressions on the work of Hutson and Stevenson (2009), Choi and Jiang (2009), and Bredin and Hyde (2011), who all contributed in the research to what affects risk factors based on on-balance and off-balance variables.

One important aspect to note with respect to the second-stage regression is the time indication. In the first stage rolling windows have been applied. This means that the first data point for a three-year window will be in the first quarter of 2003. For the four-year window the first data point is in the first quarter of 2004, and of course the first data point for the five-year window is in the first quarter of 2005. However, following this time indication for the second-stage regression could result in reversed causality. Hence, when regressing a risk exposure from the three-year window of the first quarter of 2003, the financial accounting data 3 years prior to that moment is used in an attempt to explain the exposure. Figure 7 visualizes this principle.

The data used from the FR Y-9C reports are all scaled by the total assets of the corresponding BHC, except for size. The size variable takes the natural logarithm from the total assets of a BHC. The reason to scale the variables from the FR Y-9C report is to make these variables comparable from BHC to BHC.

Figure 7: Visualized time indicator for the second-stage regression

Note: This figure visualizes how the time indication works for the second-stage regression. A window of length T generates a data point at the end of the window period (point A). To be able to measure any explanatory value from the financial data, the financial accounting data from the beginning of the window is used (point B).

18

The risk factors that are regressed are the interest rate risk, forex risk, and credit risk. The interest risk beta is regressed as follows:

𝛽_{𝑖𝑟,𝑡} = 𝜙_𝑖+ ∑ 𝜉_𝑗𝐴_𝑗𝑖𝑡 + ∑ 𝜍_𝑗𝐵_𝑗𝑖𝑡+ ∑ 𝐷(𝜍_𝑗𝐵_𝑗𝑖𝑡) 𝑗 + 𝜌_𝑗𝑍_𝑗𝑖𝑡 𝑗 𝑗 , ₍₂₎

where 𝐴𝑗𝑖𝑡 denotes certain on-balance sheet variable 𝑗, from BHC 𝑖 at time 𝑡. The on-balance

sheet variables are commercial and industrial loans (C&I loans), Mortgages (which are all loans with real estate as collateral), other loans, deposits, and the Gap ratio (the difference between the assets and liabilities that are repriced within a year). These variables are scaled by the total assets. The reason for including these on-balance sheet variables is that they are likely to impact a bank’s exposure to interest rate risk (Li and Marinč, 2014). 𝐵_𝑗𝑖𝑡 denotes the notional value of both the interest rate derivatives used for trading and hedging, scaled by the total assets. Also dummy variable is added that takes on a one when the observation is during the economic crisis in the years: 2007, 2008, and 2009, denoted by 𝐷. The crisis dummy interacts with the variables on derivatives usage to measure if the crisis amplifies possible effects. Then also three control variables are added: size, capital ratio, and GDP-growth, which are denoted by 𝑍𝑗𝑖𝑡.

For the foreign exchange risk the equation looks as follows: 𝛽_{𝑓𝑥,𝑡} = 𝜈_𝑖 + ∑ 𝜃_𝑗𝑂_𝑗𝑖𝑡+ ∑ 𝜇_𝑗𝑃_𝑗𝑖𝑡+ ∑ 𝐷(𝜇_𝑗𝑃_𝑗𝑖𝑡) + 𝑗 𝜌_𝑗𝑍_𝑗𝑖𝑡 𝑗 𝑗 , ₍₃₎

where 𝑂𝑗𝑖𝑡 are on-balance variables (𝑗) market liquidity and funding liquidity for BHC 𝑖 at

time 𝑡, scaled by the total assets. 𝑃_𝑗𝑖𝑡 are the notional values of foreign exchange derivatives for trading and hedging, scaled by the total assets. Then a dummy variable is added that takes on a one when the observation is during the economic crisis in the years: 2007, 2008, and 2009, denoted by 𝐷. The crisis dummy interacts with the variables on derivatives usage to measure if the crisis amplifies possible effects. Lastly three control variables are added: size, capital ratio, and GDP-growth, which are denoted by 𝑍_𝑗𝑖𝑡. Thirdly, to measure the effect of credit derivatives on credit risk the following model is used:

𝛽_{𝑐𝑟,𝑡} = 𝜓_𝑖+ ∑ 𝜏_𝑗𝑈_𝑗𝑖𝑡+ ∑ 𝜅_𝑗𝑃_𝑗𝑖𝑡 + ∑ 𝐷(𝜅_𝑗𝑃_𝑗𝑖𝑡) + 𝑗 𝜌_𝑗𝑍_𝑗𝑖𝑡 𝑗 𝑗 , (4)

19

loss provision of BHC 𝑖 at time 𝑡, scaled by total assets. 𝑍_𝑗𝑖𝑡 denotes the notional value credit protection bought and credit protection sold, scaled by total assets. The FR Y-9C reports do not make the same distinction with credit derivatives as they do with interest rate derivatives and foreign exchange derivatives. I follow Nijskens and Wagner (2011) in that a bank is hedging its credit risk when it is buying credit protection and trading when the bank is selling the credit protection. Then a dummy variable is added that takes on a one when the observation is during the economic crisis in the years: 2007, 2008, and 2009, denoted by 𝐷. The crisis dummy interacts with the variables on derivatives usage to measure if the crisis amplifies possible effects. And the three control variables size, capital ratio, and GDP growth are added as control variables.

Lastly, to measure the effect of derivatives on real estate the following model is used:

𝛽_{𝑟𝑒𝑖𝑡,𝑡} = 𝜆_𝑖 + ∑ 𝜒_𝑗𝑅_𝑗𝑖𝑡+ ∑ 𝜔_𝑗𝑄_𝑗𝑖𝑡+ ∑ 𝐷(𝜔_𝑗𝑄_𝑗𝑖𝑡) + 𝑗 𝜌_𝑗𝑍_𝑗𝑖𝑡 𝑗 𝑗 , (5)

Model (5) is almost the same as used to measure the effect of derivatives on credit risk in model (4), since loans to real estate are not much different from regular loans, except they have real estate as collateral. 𝑅_𝑗𝑖𝑡 denotes the following on-balance variables (𝑗): Mortgages, market liquidity, funding liquidity, non-performing loans, loans charge-offs, and loan loss provision, scaled by the total assets. 𝑄_𝑗𝑖𝑡 are the notional values of credit protection bought and credit protection sold, scaled by the total assets. Then a dummy variable is added that takes on a one when the observation is during the economic crisis in the years: 2007, 2008, and 2009, denoted by 𝐷. The crisis dummy interacts with the variables on derivatives usage to measure if the crisis amplifies possible effects. And also here the three control variables size, capital ratio and GDP growth are included.

In Appendix B you can find detailed information on all the variables, their definitions and source.

20

absence of heterogeneity bias, which is often not the case (Bell and Jones, 2015). The Hausman test (Hausman, 1978) is often used to determine which type to use. However, choosing one model or the other based on a statistical test is myopic, and it should not be used as a conclusive argument on whether to use fixed effects or random effects (Bell et al., 2017). A good reason to use random effects over fixed effects is when we can assume that the different entities (BHCs) are different from each other. This assumption looks valid, as we have seen that the use of derivatives greatly differs from the largest banks to smaller banks. For this reason this study diverges from (Li and Marinč, 2014) and uses random effects instead of fixed effects. The standard errors are heteroskedasticity consistent and yearly dummies are employed to control for unobserved factors that change over time.

4. Empirical results

4.1 First stage: Estimation of the risk factors

In the first stage of regression we have estimated how U.S. BHCs are exposed to the risk factors interest rate risk, credit risk, foreign exchange risk, and real estate risk. These risks are discussed separately. A general observation is that all the risks are far from static, they are rather dynamic.

21

interest rate dropped sharply to as low as zero. The short-term interest stayed at low levels until 2015. As monetary policy has been so prominent since 2007, the change of the interest rate risk exposure is likely a consequence of this.

Figure 8: Interest rate exposures (coefficients on the left axis) on the 3, 4, and 5-year windows for

the full sample and large BHC sample

Note: The shaded areas show when the estimated interest rate risk exposure has significance of at

least 10%. The solid line shows an unweighted arithmic average beta of the BCHs in the (sub)sample. The confidence interval is based on the standard deviation of the individual estimated interest rate risk exposure at a certain point in time. The upper row shows the results for the entire sample. The lower row shows the results for the subsample consisting of the largest BHCs.

22

Research into the effects of sudden interest rate changes in Europe has found that generally speaking it hard to predict what the effect on the profitability of banks will when interest rates changes due to monetary policy (Ampudia, M. and Heuvel, S. van den, 2018). They mention that a steepening yield curve tends to improve interest margins, but the capital gains or losses that happen due to changes in the yield curve eliminate the gain from larger margins. Ampudia and Van den Heuvel (2018) have found that pre-crisis, an unexpected increase of the short-term interest rate, equity values of banks decreased. However, post-crisis this effect was reversed. Meaning that the relationship between the interest rate risk went from negative, to positive. Fiordelisi, Galappo, and Ricci (2014) have found evidence that when the Federal Funds rate was left unchanged or was increased, stock returns of banks increase as well. A reason for this might be that after the financial crisis an increase in interest rates was viewed as a sign of a recovering economy, which in turn could increase lending activities at banks. And that when the interest rates are too low, it diminishes the This is a conclusion Aït-Sahalia, Andritzky, Jobst, Nowak, and Tamirisa (2012) come to as well, mentioning the positive response to unchanged or increasing interest rates could have been observed by the market as a strengthening economy and increasing confidence in the economy. There is no real difference between the estimated interest rate risk for the full sample and the large BHC subsample. They follow the same pattern.

23

riskier asset (Gambacorta, 2009). Investors moved from lower-yielding government bonds and investment grade corporate bonds towards yield bonds. The demand for more high-yield bonds from asset managers would mean that banks could securitize risky non-investment grade bonds rather easily. Low interest rates also affect valuations as it increases collateral values, which reduces the loss-given-default of a loan (Gambacorta, 2009). These arguments are shared by Adrian and Shin (2009) and Altunbas, Gambacorta, and Marqués-Ibáñez (2010). Another important argument is the behavior of the Federal Reserve. During the crisis the Federal Reserve and the U.S. Government relieved stress on the financial sector through a number of ways: Tax cuts, capital injections, and refinancing schemes. In other words, if a bank was in distress during the crisis, the government and the Federal Reserve stepped in to alleviate (Kincaid, 2010). This could mean that banks could take on risks through lending they would normally not provide, because there was a safety net in place. The increased risk premium during the crisis then provided banks with more profitable loans.

24 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 2000q1 2006q1 2012q1

Lar

ges

t

B

HCs

samp

le

-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 2000q1 2006q1 2012q1 Credit Risk 95% confidence interval -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 2000q1 2006q1 2012q1 -0.4 -0.2 0 0.2 0.4 0.6 2000q1 2006q1 2012q1

Full

Samp

le

3-year window

-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 2000q1 2006q1 2012q1

4-year window

-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 2000q1 2006q1 2012q1

5-year window

Note: The shaded areas show when the estimated credit risk exposure has significance of at least 10%.

The solid line shows an unweighted arithmic average beta of the BCHs in the (sub)sample. The confidence interval is based on the standard deviation of the individual estimated interest rate risk exposure at a certain point in time. The upper row shows the results for the entire sample. The lower row shows the results for the subsample consisting of the largest BHCs.

Figure 9: Credit risk exposures (coefficients on the left axis) on the 3, 4, and 5-year windows on the full

25

The last risk examined is real estate risk. This is in line with the theory: Positive exposure and significant (Bessler and Kurmann, 2014; Bessler et al., 2015). In Figure 11 we can see that for both the full index as the reduced index real estate risk is significant for large part of the period. Until the third quarter of 2015 real estate risk has a significant effect on Figure 10: Foreign exchange risk exposures (coefficients on the left axis) on the 3, 4, and 5-year

windows on the full sample and large BHC sample

-2 -1 0 1 2 3 4 2000q1 2006q1 2012q1

Lar

ges

t

B

HCs

samp

le

-2 -1 0 1 2 3 4 2000q1 2006q1 2012q1 Foreign Exchange Risk 95% confidence interval -2 -1 0 1 2 3 4 2000q1 2006q1 2012q1 -2 -1 0 1 2 3 4 2000q1 2006q1 2012q1

Full

Samp

le

3-year window

-2 -1 0 1 2 3 4 2000q1 2006q1 2012q1

4-year window

-2 -1 0 1 2 3 4 2000q1 2006q1 2012q1

5-year window

Note: The shaded areas show when the estimated foreigen exchange risk exposure has significance of at

26

both the total BHC index as the large BHC index. This makes sense since real estate loans and mortgages make up a sizeable portion of a bank’s loan portfolio. The percentage of loans backed by real estate as part of total assets grew sharply after the Glass-Steagall act was repealed in 1999 (Carmichael and Coën, 2015). When this act was repealed banks could both provide investment banking services and commercial banking services, opening up the mortgage market to investment banks. After 2010 the percentage of real estate loans dropped to levels we have seen under the Glass-Steagall act (Carmichael and Coën, 2015). This is likely due to the Dodd-Frank act that was instated in 2010 which attempted to change the incentive for banks to engage in real estate loans. Under Dodd-Frank banks cannot completely securitize real estate loans and they also have to make a fair assessment on whether a borrower is likely to repay. Adams-Kane (2018) shows that the exposure from banks to mainly commercial real estate has been transferred for large part from geographically diversified to geographically concentrated banks since 2007-2009 financial crisis. Adams-Kane (2018) mentions that these geographically concentrated banks are usually specialized in commercial real estate. As an effect local banks are less affected by cyclical and volatile lending for construction. The concentration of commercial real estate has shifted mostly banks that are not systematically important (Adams-Kane, 2018). As an effect we can see that real estate risk has lost its statistical importance in bank stock returns.

27

Figure 11: Real estate risk exposures (coefficients on the left axis) on the 3, 4, and 5-year windows

-0.5 0 0.5 1 1.5 2 2.5 2000q1 2006q1 2012q1

Full

samp

le

3-year window

-0.5 0 0.5 1 1.5 2 2.5 2000q1 2006q1 2012q1

4-year window

-0.5 0 0.5 1 1.5 2 2.5 2000q1 2006q1 2012q1

5-year window

-0.5 0 0.5 1 1.5 2 2.5 2000q1 2006q1 2012q1

Lar

ges

t

B

HC

samp

le

-0.5 0 0.5 1 1.5 2 2.5 2000q1 2006q1 2012q1 Real Estate Risk 95% Confidence interval -0.5 0 0.5 1 1.5 2 2.5 2000q1 2006q1 2012q1

Note: The shaded areas show when the estimated real estate risk exposure has significance of at least 10%.

28

Figure 12: Adjusted R-squared of the full model as shown in (1)

Note: This figure shows the average adjusted r-squared of model (1) as shown in 3.2. This shows how

much explanatory power model (1) has for all the different windows and both the full sample as the sub sample. The top row shows the results from the full sample and the bottom row shows the results from the subsample.

29

4.2 Second stage: Determinants of the risk factors

Based on the systematic risk exposures obtained in the first stage regression, we now look at if derivatives usage by BHCs affect the exposure of interest rate risk, credit risk, foreign exchange risk, and real estate risk. To measure the effect of derivatives on bank risks a random effects model is used, as discussed in part 3.3.

30

Table 3: Results random effects model on interest rate risk exposures on the 3, 4, and 5-year windows.

3-year window 4-year window 5-year window

VARIABLES Full sample Large BHC

sample Full sample Large BHC sample Full sample Large BHC sample

IRD for trading -0.00222 0.000189 -0.00210 -8.53e-05 -0.00227 0.000470

(0.00451) (0.00338) (0.00488) (0.00363) (0.00396) (0.00360)

IRD for trading in crisis -0.00280 -0.00132 -0.00183 -0.000551 -0.000956 0.00134

(0.00498) (0.00417) (0.00500) (0.00405) (0.00439) (0.00433)

IRD for hedging -0.0381 -0.100*** -0.0441 -0.101*** -0.0474 -0.0788**

(0.0405) (0.0302) (0.0382) (0.0346) (0.0346) (0.0374)

IRD for hedging in crisis 0.0936 0.0320 0.0336 -0.0243 0.0496 0.0141

(0.0579) (0.0412) (0.0412) (0.0333) (0.0430) (0.0285) C&I Loans 0.490*** 0.116 0.402** 0.0654 0.337* 0.0266 (0.170) (0.263) (0.173) (0.263) (0.173) (0.250) Mortgages -0.159** -0.248** -0.159** -0.261** -0.0987 -0.235* (0.0774) (0.126) (0.0710) (0.121) (0.0676) (0.126) Other loans -0.828*** -0.637*** -0.831*** -0.696*** -0.847*** -0.787*** (0.132) (0.163) (0.131) (0.164) (0.131) (0.131) Deposits 0.495*** 0.494*** 0.538*** 0.533*** 0.470*** 0.509*** (0.100) (0.123) (0.0914) (0.119) (0.0865) (0.151)

Gap ratio 6.31e-06 8.78e-05 8.45e-05* -7.52e-06 6.85e-05* -5.41e-05

(1.09e-05) (0.000102) (5.03e-05) (0.000102) (4.15e-05) (4.85e-05)

Size 0.109*** 0.0711*** 0.114*** 0.0761*** 0.106*** 0.0785*** (0.0106) (0.0180) (0.0106) (0.0191) (0.0107) (0.0199) Capital ratio 2.708*** 2.749*** 2.958*** 3.082*** 3.699*** 5.060*** (0.443) (0.777) (0.460) (0.724) (0.608) (1.142) GDP Growth -7.368*** -10.41*** -7.012*** -9.442*** -7.310*** -8.935*** (0.634) (1.090) (0.523) (0.900) (0.479) (0.859) Constant -1.962*** -1.269*** -2.088*** -1.398*** -1.985*** -1.589*** (0.202) (0.344) (0.190) (0.362) (0.183) (0.360) Observations 4,543 1,328 4,398 1,304 4,043 1,202 Number of BHCs 98 27 98 27 97 26

Note: IRD stands for interest rate derivatives. The regressions are made using a random effects model. For readability purposes the yearly dummy variables are

31

32

Table 4: Results random effects model on foreign exchange risk exposures on the 3, 4, and 5-year windows on the full sample and the

large BHC sample.

3-year window 4-year window 5-year window

VARIABLES Full sample Large BHC

sample Full sample Large BHC sample Full sample Large BHC sample

FXD for trading 0.000373 0.0911 -0.194* -0.253** -0.165 -0.279**

(0.0986) (0.109) (0.117) (0.121) (0.130) (0.139)

FXD for trading in crisis 0.0290 0.0614 -0.265*** -0.276** -0.199* -0.261**

(0.0704) (0.0953) (0.101) (0.112) (0.116) (0.126)

FXD for hedging -2.229*** 3.776* -3.949*** -2.741*** -2.684** -0.704

(0.614) (2.210) (1.232) (1.004) (1.325) (1.035)

FXD for hedging in crisis -0.789* 7.485** 2.778 2.525 1.990 1.312

(0.452) (3.744) (4.738) (4.558) (4.474) (4.163)

Foreign exchange assets 0.103 0.558 -1.386 -0.202 -1.475 -0.678

(0.850) (0.722) (2.035) (1.881) (1.475) (1.654)

Foreign exchange deposits -2.801*** -2.292*** -0.729 1.247 -0.729 1.420

(0.530) (0.885) (0.661) (1.262) (0.703) (1.173) Size 0.257*** 0.0392 0.204*** 0.220*** 0.208*** 0.258*** (0.0457) (0.0640) (0.0521) (0.0555) (0.0564) (0.0670) Capital ratio 8.513*** 5.567 -7.095*** -2.555 -9.740*** -6.717*** (2.278) (4.118) (1.641) (1.830) (1.619) (2.568) GDP Growth -28.98*** -8.378 -21.34*** -11.80*** -14.24*** -9.306*** (3.347) (5.204) (2.240) (3.238) (1.663) (2.357) Observations 4,289 1,158 4,058 1,199 3,662 1,099 Number of BHCs 98 26 97 26 97 26

Note: FXD stands for foreign exchange derivatives. The regressions are made using a random effects model. For readability purposes the

33

Next, the effect of the use of credit derivatives on credit risk is measured. There some differences with the random effects models (2) and (3). First, the FR 9-YC reports do not distinguish between credit derivatives used for hedging and trading. Instead the report distinguishes only between credit derivatives bought and sold. For the purpose of this study it is considered that sold credit exposure is for trading purpose, whereas buying credit protection is considered hedging. This is in line with Nijskens and Wagner (2011). A second point with regard to credit derivatives is that the FR Y-9C reports are not available before the first quarter of 2006. In Table 5 the results are shown. In the three-year window selling credit protection does not seem to have an effect on credit risk for the average BHC in the sample. However, during the crisis the effect was positive and significant at the 1% level. For large BHCs this is the other way around. Outside of the crisis the effect is positive at significance level of 1%, whereas during the crisis the effect is no longer significant. A similar result is found when a BHC buys credit protection. Outside of the crisis the effect for a BHC in the full sample is on average not significant, but during the crisis the effect is negative and significant at the 1% level. For the large BHC it is again the other way around. Buying credit protection outside of the crisis period has negative and significant (1%) impact on credit risk, whereas during the crisis this effect is no longer significant. In the four-year window only the full sample shows some positive impact on credit risk during the crisis at a 10% level of significance. And when buying credit protection there is a slight negative impact at a 10% level of significance for the full sample during the crisis period. The five-year window shows no significance at all. This is somewhat in line with Li and Marinč (2014), who also find that using credit derivatives positively impact credit risk. However, statistically significant results are only to be found in the three-year window, which has by construction the least accurately estimated credit risk exposures. The statistical significance rapidly disappears when lengthening the windows with more accurate estimations. Because of this the third hypothesis is not fully accepted. There is some evidence that credit derivatives impact credit risk, but it is not persuasive enough.

34

Table 5: Random effects model on credit risk exposures on the 3, 4, and 5-year windows on the full sample and large BHC sample.

3-year window 4-year window 5-year window

VARIABLES Full sample Large BHC

sample Full sample Large BHC sample Full sample Large BHC sample

Credit protection sold 0.364 2.333*** 0.131 0.426 0.360 0.703

(0.317) (0.586) (0.682) (0.711) (0.776) (0.948)

Credit protection sold in crisis 2.995*** 0.0434 0.495* 0.346 0.440 0.380

(0.441) (1.878) (0.260) (0.364) (0.433) (0.503)

Credit protection bought -0.364 -2.738*** -0.0181 -0.348 -0.210 -0.562

(0.299) (0.634) (0.709) (0.722) (0.770) (0.923)

Credit protection bought in crisis -2.917*** -0.479 -0.453* -0.324 -0.385 -0.322

(0.459) (1.914) (0.235) (0.333) (0.389) (0.458) Market liquidity 0.0697 -0.0989 -0.000560 -0.171 -0.0683 -0.144 (0.0684) (0.115) (0.0640) (0.113) (0.0567) (0.112) Funding liquidity 0.460 0.213 0.217 0.0266 -0.0885 -0.305 (0.358) (0.744) (0.305) (0.447) (0.277) (0.396) Non-performing loans 1.563** 0.243 0.390 -0.0556 -0.225 -0.388 (0.678) (0.631) (0.387) (0.527) (0.382) (0.368) Loan charge-offs -8.165*** -1.685 -5.528** -0.750 -6.264*** -0.634 (2.558) (3.731) (2.463) (3.693) (2.186) (3.132)

Loan loss provision 7.417** 3.181 6.778*** 2.461 8.083*** 3.658

(2.948) (3.806) (2.411) (3.462) (2.080) (2.864) Size -0.0345*** 0.00604 -0.0384*** -0.0315** -0.0328*** -0.0333*** (0.00735) (0.0339) (0.00709) (0.0135) (0.00701) (0.0125) Capital ratio -0.416 -0.0363 -0.545* -0.166 -0.639** -0.536 (0.294) (0.427) (0.280) (0.447) (0.299) (0.620) GDP growth 1.152*** 2.094*** 0.917*** 1.610*** 1.400*** 1.799*** (0.336) (0.739) (0.266) (0.495) (0.245) (0.450) Constant 0.641*** -0.0431 0.737*** 0.667** 0.680*** 0.727*** (0.136) (0.631) (0.131) (0.287) (0.124) (0.278) Observations 2,383 548 2,317 712 2,029 629 Number of BHCs 79 17 75 22 73 21

Note: The regressions are made using a random effects model. For readability purposes the yearly dummy variables are not shown. Robust standard errors

35

Lastly, the effect of credit derivatives on real estate risk is assessed. Also, this part of the analysis has the same restrictions as the credit risk analysis does. There is no data available on credit derivatives before the first quarter of 2006. The random effects model (4) has run over the three, four, and five-year windows on both the total BHC sample as the large BHC subsample. The results in Table 6 show in the three-year window there is no statistical evidence that credit derivatives impact the real estate exposure of BHCs. The four-year window has much stronger evidence. Selling credit protection increases real estate exposure for both the average BHC in the full sample as a large BHC at a 1% and 5% significance level respectively. During the crisis there is no extra effect to be found. When buying the credit protection, the exposure towards real estate is lowered for both the full sample as the large BHC subsample at a 1% and 5% significance level respectively. In the five-year window the results is similar to the four-year window, but for the large BHCs the effect is now significant at a 1% level. This means there is strong evidence that using credit derivatives has an impact on the real estate exposure of BHCs. Also the coefficients are sufficiently large to say that there also is an economic significance. Because of this the fourth hypothesis is accepted. Using credit derivatives impacts a BHC’s real estate risk.

36

Table 6: Random effects model on real estate risk exposures on the 3, 4, and 5-year windows on the full sample and large BHC sample

3-year window 4-year window 5-year window

VARIABLES Full sample Large BHC

sample Full sample Large BHC sample Full sample Large BHC sample

Credit protection sold 3.912 3.951 7.086*** 4.446** 6.258*** 3.859***

(2.612) (2.680) (1.439) (1.900) (1.068) (1.445)

Credit protection sold in crisis 6.167 8.258 1.887 0.920 2.156 1.608

(8.606) (12.88) (2.805) (3.232) (1.749) (2.164)

Credit protection bought -3.229 -2.900 -6.695*** -4.377** -6.043*** -3.841***

(2.497) (2.591) (1.304) (1.741) (0.970) (1.324)

Credit protection bought in crisis -5.082 -5.582 -1.398 -0.693 -1.784 -1.417

(8.496) (14.03) (2.665) (3.047) (1.645) (2.023) Mortgages 1.931*** 0.468 1.826*** 0.957 1.346*** 0.492 (0.421) (0.913) (0.403) (0.846) (0.408) (0.873) Market liquidity 0.982** -0.398 0.837** -0.608 0.398 -1.230* (0.410) (0.705) (0.359) (0.609) (0.383) (0.704) Funding liquidity 1.451* -1.376 2.296** 1.528 1.327 -0.200 (0.788) (1.156) (0.932) (1.318) (0.947) (1.041) Non-performing loans -0.0167 0.00289 -3.536* -3.464 -5.071*** -4.350* (0.0213) (0.0224) (1.819) (2.739) (1.570) (2.340) Loan charge-offs -0.352*** -0.224 -32.55*** -22.56* -31.47*** -20.05* (0.0921) (0.152) (8.517) (13.55) (7.019) (11.09)

Loan loss provision 0.396*** 0.301** 35.88*** 32.21*** 31.89*** 27.70***

(0.0680) (0.126) (6.505) (10.68) (5.207) (8.062) Size -0.0952** -0.182** -0.0801* -0.0343 -0.0857** -0.0567 (0.0406) (0.0754) (0.0410) (0.0685) (0.0419) (0.0636) Capital ratio -2.175** -2.765 -2.261** -1.638 -4.795*** -6.940*** (0.977) (1.905) (0.940) (1.355) (1.284) (2.555) GDP growth -8.254*** -9.677*** -8.516*** -8.214*** -6.066*** -4.673*** (1.285) (2.352) (1.016) (1.455) (0.950) (1.480) Constant 1.447* 4.078** 1.325* 1.270 1.991*** 2.511* (0.767) (1.753) (0.779) (1.438) (0.733) (1.302) Observations 2,391 556 2,317 712 2,029 629 Number of BHCs 80 18 75 22 73 21

Note: The regressions are made using a random effects model. For readability purposes the yearly dummy variables are not shown. Robust standard

37

5. Conclusion

This study examined the relationship between the use of derivatives and the risk of publicly traded bank holding companies in the United States in the period between 2000 and 2018. There is ambiguity when it comes to the risk of banks and the use of derivatives. A number of papers have found evidence that concludes that banks increase their risk by using derivatives, other papers claim the opposite. This study is an attempt to add more knowledge to this discussion, but I am afraid this study found ambiguity as well.

In a two-staged analysis this study test four hypotheses:

H1: The use of interest rate derivatives by BHCs impacts the interest rate risk of BHCs. H2: The use of foreign exchange derivatives by BHCs impacts the foreign exchange risk of BCHs.

H3: The use of credit derivatives by BHCs impacts the credit risk of BHCs. H4: The use of credit derivatives by BHCs impacts the real estate risk of BHCs.

The first stage this study has estimated to what degree U.S. bank holding companies are exposed to interest rate risk, credit risk, foreign exchange risk, and real estate risk. The returns of these BHCs have been generated and regressed on the four different risk factors using three, four, and five-year rolling windows. These windows have shown that the risk factors are not static, but quite dynamic. The first stage did show some surprising results, but they could be explained through economic argument, regulatory policy and monetary policy.

In the second stage the analysis continues by taking the estimated risk factors from the first stage and regressing them on financial accounting data, obtained from the quarterly FR Y-9C reports by the Federal Reserve. In the second stage panel regression techniques have been applied with fixed effects to measure if derivatives increase the risk of a BHC or not.

Hypothesis 1 has not been fully accepted. On average there is no statistical evidence that using interest rate derivatives impacts the interest rate risk. However, there is some evidence that the largest BHCs can negatively impact their interest rate risk by hedging.

Hypothesis 2 has been accepted. The results were somewhat ambiguous, but there is clear evidence that foreign exchange derivatives impact the foreign exchange risk of both the average BHC in the full sample as a large BHC.

38

Hypothesis 4 has been accepted. The three-year window showed no significance, but the four- and five-year window showed clear evidence with high significance that both the average BHC in the sample as the largest BHCs impact their real estate risk when using credit derivatives. Buying credit protection leaves a BHC less exposed to real estate, whereas selling credit protection increases the exposure to real estate.

Concluding, this study has not been able to determine whether using derivatives categorically increases a bank’s exposure to interest rates, credit, foreign exchange, and real estate. But there is evidence that it does. More research is needed into this subject to come to conclusive answer. There are some issues to be found in the data of this study. The FR Y-9C reports seem to change from time to time, which makes it very difficult to abstract from these reports what actually is the true number.

39

References

Acharya, V., Drechsler, I. and Schnabl, P., 2014. A Pyrrhic Victory? Bank Bailouts and Sovereign Credit Risk. The Journal of Finance 69(6), 2689-2739.

Adams-Kane, J., 2018. Commercial real estate: How vulnerable are U.S. Banks? Milken Institute, 1-25.

Adrian, T. and Shin, H.S., 2009. Money, liquidity, and monetary policy. American Economic Review: Papers & Proceedings 99(2), 600-605.

Aït-Sahalia, Y., Andritzky, J., Jobst, A., Nowak, S. and Tamirisa, N., 2012. Market response to policy initiatives during the global financial crisis. Journal of International Economics 87, 162-177.

Allen, F. and Carletti, E., 2006. Credit risk transfer and contagion. Journal of Monetary Economics 53, 89-111.

Allayanis, G. and Ofek, E., 2001. Exchange rate exposure, hedging, and the use of foreign currency derivative. Journal of International Money and Finance 20, 273-296.

Alnassar, W.I. and Bin Chin, O., 2015. Why banks use credit derivatives? Review paper, Procedia Economics and Finance 26, 566-574.

Altunbas, Y., Gambacorta, L. and Marqués-Ibáñez, D., 2010. Does monetary policy affect bank risk-taking? European Central Bank Working Paper Series 1160, 1-44.

Ampudia, M. and Heuvel, S. Van den, 2018. Monetary policy and bank equity values in a time of low interest rates. European Central Bank Working Paper Series 2199, 1-41.

Ashraf, D., Altunbas, Y. and Goddard, J., 2007. Who Transfers Credit Risk? Determinants of the Use of Credit Derivatives by Large US Banks. The European Journal of Finance 13(5), 483-500.

40

Bartram, S. M., Brown, G. W. and Fehle, F.R., 2009. International evidence on financial derivatives usage. Financial Management, 38(1), 185-206.

Becketti, S., 1993. Are derivatives too risky for banks? Economic Review 78(3), 27-42. Bell, A. and Jones, K., 2015. Explaining fixed effects: Random effects modeling of time-series cross-sectional and panel data. Political Science and Methods 3(1), 133-153.

Bell, A., Fairbrother, M. and Jones, K., 2017. Fixed and random effects models: making an informed decision. Quality & Quantity 53(2), 1051-1074.

Berndt, A., Jarrow, R.A., Kang, C., 2007. Restructuring risk in credit default swaps: An empirical analysis. Stochastic processes and their Applications 117, 1724-1749.

Bessler, W. and Kurmann, P., 2014. Bank risk factors and changing risk exposures: Capital market evidence before and during the financial crisis. Journal of Financial Stability 13, 151-166.

Bessler, W., Kurmann, P. and Nohel, T., 2015. Time-varying systematic and idiosyncratic risk exposures of US bank holding companies. Journal of International Financial Markets,

Institutions, & Money 35, 45-68.

Bessler, W. Leonhardt, A. and Wolff, D., 2016. Analyzing hedging strategies for fixed income portfolios: a Bayesian approach for model selection. International Review of Financial Analysis 46, 239-256.

Bessler, W., Conlon, T. and Huan, X., 2019. Does corporate hedging enhance shareholder value? A meta-analysis. International Review of Financial Analysis 61, 222-232.

Booth, J.R., Smith, R.L. and Stolz, R.W, 1984. Use of interest rate futures by financial institutions. Journal of Banking Research 15(1), 15-20.

Bredin, D. and Hyde, S., 2011. Investigating sources unanticipated exposure in industry stock returns. Journal of Banking & Finance 35, 1128-1142

41

Calistru, R.A., 2012. The credit derivative market – a threat to financial stability? Procedia, Social and Behavioral Sciences 58, 552-559.

Carmichael, B. and Coën, A., 2015. Real estate as a common risk factor in bank stock returns, Journal of Banking and Finance 94, 118-130.

Choi, J.J. and Elyasiani, E., 1997. Derivate exposure and the interest rate and exchange rate risks of U.S. Banks. Journal of Financial Services Research 12, 267-286.

Choi, J.J. and Jiang, C., 2009. Does multinationality matter? Implications of operational hedging for exchange risk exposure. Journal of Banking & Finance 33, 1793-1982.

Chue, T.K. and Cook, D., 2008. Emerging market exchange rate exposure. Journal of Banking & Finance 32(7), 1349-1362.

Clifford, D.R.N., Smith, W. and Smithson, C.W., 1993. On the Determinants of Corporate Hedging. The Journal of Finance 48(1), 267-284.

Cyree, K.B., Huang, P. and Lindley, J.T., 2011. The Economic Consequences of Banks’ Derivatives Use in Good Times and Bad Times. Journal of Financial Services Research 41, 121-144.

Diamond, D.W. and Rajan, R.G., 2000. A Theory of Bank Capital, The Journal of Finance 55, 2431-2465.

Doidge, C., Griffin, J., and Williamson, R., 2006. Measuring the economic importance of exchange rate exposure. Journal of Empirical Finance 13(4), 550-576

Fama, E.F. and French, K.R., 1989. Business conditions and expected returns on stocks and bonds. Journal of Financial Economics 25, 23-49.

Fiordelisi, F., Galloppo, G. and Ricci, O., 2014. The effect of monetary policy intervention on interbank markets, equity indices and G-SIFIs during financial crisis. Journal of Financial Stability 11: 49-61.

42

Gay, G.D. and Nam, J., 1998. The underinvestment problem and corporate derivatives use. Financial Management 27(4), 53-69.

Genay, H. and Podjasek, R., 2014. What is the impact of a low interest rate environment on bank profitability. Chicago Fed Letter 324.

Gorton, G.B. and Rosen, R.J., 1995. Banks and Derivatives. Rodney L. White Center for Financial Research, Working Paper, 6-95.

Gounopolos, D., Molyneux, P., Staikouras, S.K., Wilson, J.O.S. and Zhao, G., 2012. Exchange rate risk and equity performance of financial intermediaries. International Review of Financial Analysis 29, 271-282.

Greenspan, A., 1999. Financial Derivatives. Before the futures industry association, Boca Raton. Florida.

Hausman, J.A., 1978. Specification tests in econometrics, Econometrica 46(6), 1251-1271. Hendershott, P.H. and Kane, E.J., 1990. Causes and consequences of the 1980s commercial construction boom, Journal of Applied Corporate Finance 5(1), 61-70.

Hentschel, L. and Kothari, S.P., 2001. Are corporations reducing or taking risks with derivatives? Journal of Financial and Quantitative Analysis 36(1), 93-118.

Huan, X. and Parbonetti, A., 2017. Do financial derivatives make banks less risky? Evidence from eleven developed markets. Working paper.

Huizinga, H. and Laeven, L., 2012. Bank valuation and accounting discretion during a financial crisis, Journal of Financial Economics 106, 614-634.

Hull, J.C., 2015a. Risk Management and Financial Institutions 4th _{edition, Hoboken: Wiley &}

Sons, Inc.

Hull, J.C., 2015b. Options, Futures, and Other Derivatives 9 th edition, Harlow: Pearson Education Limited.