• No results found

Investigating the existence of a bank lending channel in Sweden

N/A
N/A
Protected

Academic year: 2021

Share "Investigating the existence of a bank lending channel in Sweden"

Copied!
34
0
0

Loading.... (view fulltext now)

Full text

(1)

Master thesis, 15 ECTS

Master’s Programme in Economics/ Master thesis (II), 15 Credits Spring 2020

Investigating the existence

of a bank lending channel

in Sweden

(2)

Abstract

This paper uses quarterly Swedish bank-level data from 1999 to 2019 in an attempt to study the existence of a bank lending channel in Sweden. The bank lending channel is concerned with the role of credit in an economy, where it is argued that monetary policy can affect the real economy through a credit channel, in which banks shift their supply of loans to the public. The bank lending channel is hence part of the broader credit channel, where the external finance premium decides how large of an impact the bank lending channel actually has on the economy. Using a Vector Error Correction Model, the short- and long-run dynamics of the bank lending channel is observed in this paper. The paper finds statistically significant evidence for the existence of a bank lending channel in Sweden by primarily observing the Stockholm Interbank Offered Rate (STIBOR) and the supply of loans made by Swedish banks to the public. By using the Johansen test for cointegration, the paper estimates both the short- and long-run coefficients of the model. The long-run results are in line with the empirical literature, where STIBOR and real GDP are both statistically significant at the 1% level.

Key words: Riksbank, monetary policy, credit channel, bank lending channel, monetary transmission channel, VAR, VECM, cointegration

(3)

Table of contents

1. Introduction ... 1

2. Literature Review ... 4

2.1. Early studies ... 4

2.2. Bank lending and low/negative interest rates ... 5

2.3. Quantitative easing ... 6

2.4. Studies on Sweden ... 7

3. Theoretical Framework ... 8

3.1. CC-LM ... 8

3.1.1. The loan market ... 8

3.1.2. The money market ... 10

3.1.3. The commodity market ... 10

3.1.4. Graphical illustration of the CC-LM model ... 11

3.2. Expected results ... 12

4. Data ... 13

5. Methodology ... 15

5.1. Time-series data and choice of econometric model ... 15

5.1.1. Vector Error Correction Model (VECM) ... 15

5.1.2. Stationarity and Cointegration ... 15

5.1.3. Econometric model (VAR and VECM) ... 17

6. Results ... 19

6.1. Stationarity and the Augmented Dickey-Fueller test ... 19

6.2. Vector Error Correction Model (VECM) ... 20

7. Discussion and conclusion ... 27

References ... 29

(4)

1

1. Introduction

The transmission channel of monetary policy is a crucial topic in macroeconomics, especially as central banks have become increasingly more interventionist in the open markets after the financial crisis of 2008. Central banks have turned to large scale asset-purchasing programmes as interest rates have reached levels close to zero (and in some cases even gone to negative rates) in many countries, making conventional expansionary monetary policy less effective.1 However, the study of monetary policy transmission and its observable effects on the real economy is always relevant, and this paper is concerned with the role of credit in monetary transmission. More precisely, the purpose of this paper is to investigate the existence of a bank lending channel in Sweden. Hallsten (1998) and Westerlund (2003) have also examined this channel of monetary transmission, however as both papers are relatively old there is certainly room for an updated and slightly modified study on the bank lending channel in Sweden, especially as monetary policy has changed significantly post-2008 with the introduction of negative interest rates.2

The bank lending channel is part of a broader view on the role of credit in economics, namely the Credit View/Credit Channel of monetary policy. Here, the theory suggests the existence of two channels: the bank lending channel and the balance sheet channel. Essentially, the balance sheet channel considers the borrowers net worth, and how monetary policy affects the value of their assets, eventually affecting their eligibility for new loans (since the value of their collateral they can provide changes). The bank lending channel however concerns the supply of loans provided by banks, and how expansive/contractionary monetary policies affects the supply of bank loans to the public. Since traditionally, in what is often referred to as the pure money view, as rates goes up, money is substituted for bonds and illiquidity increases, and vice versa when the central bank is conducting expansionary policies. Further, in Bernanke & Gertler (1995) the authors argue for the existence of an external finance premium which increases the effects of monetary policy through the credit channel. The external finance premium is described as a wedge, which reflects the difference between someone choosing to self-finance (internal, e.g. through the issuance of bonds) and the cost of external financing (applying for a bank loan). If the central bank conduct contractionary policies, affecting the bank’s willingness to supply loans (e.g. through higher reserve requirements), the size of this wedge increase as the terms

1 Negative interest rates have been introduced to several countries though, including Sweden, Denmark, the

Eurozone (ECB), etc. Their effectiveness is however another topic of discussion.

(5)

2

for the bank loan becomes less beneficial for the applicant of the loan. The same process holds for expansionary policies, where looser requirements on the bank increases their willingness to supply loans. This wedge, the difference between internal and external financing, is the premium in the external financing premium, and it captures the effects of monetary policy on bank loans, which in the end affects the real economy through the size of the external finance premium.

Firstly, in order to conduct a quantitative analysis on this topic, one must choose the appropriate measure of monetary policy. In this paper, the choice of measure is the 3-month Stockholm Interbank Offered Rate (STIBOR), which is the rate at which the “STIBOR-banks”3 lend to each other on a daily basis. The interbank rate has been used in other similar studies (Borio & Gambacorta (2017); Loupias et. al (2001)) as well as the STIBOR specifically (Westerlund (2003)). The STIBOR/interbank rate is also especially suitable in this type context (banks and monetary policy) since it is concerned with how banks respond to monetary policy, and as the STIBOR is a reference rate in Sweden which shows the average of interest rates on the daily loans between banks, it is highly relevant for this paper. For example, if the central bank conducts contractionary policies, such as increased reserve requirements, it increases the illiquidity in the banking system, causing lending to the public and between banks (interbank lending) to decrease. This would cause the interbank rate to increase since banks are now lending less between each other. This is why using the interbank rate is appropriate when studying monetary policies and their effects on the banking system, as interbank lending is crucial part of the daily business for banks.

For the statistical regression in this study, a Vector Error Correction Model (VECM) is used to examine how changes in monetary policy (STIBOR) affects the total bank lending to the public (aggregate bank loan data). The VECM, which is an extension of the Vector Autoregression model (VAR), shows both the short- and long-run relationships. However, this is only possible if there is at least one cointegrating relationship in the model. As presented throughout section 5 and 6, the Johansen test for cointegration shows that there are two cointegrating equations in the model, which is why VECM is the correct model to use in this paper. As presented in section 6, the results from the VECM are in line with the expected results presented in section 3.2., which tells us that when the central bank conducts expansionary monetary policy, which leads to a lower STIBOR, causes the supply of loans made by banks to increase. Along with this, the

(6)

3

liquidity ratio also has the same inverse effect on bank lending, meaning that when liquidity requirements loosen, the banks increase their lending activity. Additionally, real GDP and total assets on the banks’ balance sheets have a positive effect on their supply of loans. While all the results are in line with the literature, only STIBOR and real GDP are statistically significant (see table 4 in section 6.2).

While the idea that financial markets can have significant effects on the real economy has long been around, along with the concept of a credit channel, Bernanke and Blinder (1988) were certainly early in theorizing this concept. In their paper, they modify the textbook IS-LM model by adding a loan/credit market, and while allowing for imperfect substitution between bonds and money they develop a simple model where the credit channel of monetary transmission can be analysed. Kashyap and Stein (1994) further contributes to this topic by using disaggregated bank data, while using the multi-asset model of Bernanke & Blinder (1988) as their main theoretical framework (namely bonds, money, and loans), and finding evidence in favour of a bank lending channel.45 Further evidence for the existence of a BLC is provided in Kashyap & Stein (2000), where the main contribution lies in their large data-set and (again) differences in bank-specific characteristics such as liquidity ratios and total asset size. The papers mentioned in this section, while using slightly different theoretical models and data-sets, all have the same objective, which is to study the existence of a bank lending channel in the transmission of monetary policy. Furthermore, the importance of this channel not only lies in its existence but also to what extent it affects the real economy, and consequently to what degree policymakers should account for it in their process of decision-making.

The rest of this paper is structured as follows: Section 2 presents the literature review in order to provide some comparable results from similar studies and other regions. Section 3 provides the theoretical framework which the study in this paper is based on. Sections 4 and 5 explains how the data is constructed and which econometrical model is used, along with other descriptions of the methodology for this paper. Section 6 presents the relevant results and lastly section 7 provides a brief discussion of the results and concludes the paper.

4 Although, the authors conclude that further research is needed in order to obtain significant and applicable

results. Specifically, they argue that their time-series data is too short and additionally further disaggregation is needed for the bank-specific data.

5 The authors, somewhat confusingly, uses the term “balance sheet” in the title of their paper. But again, it is

based on the bank’s balance sheets, which is why it still concerns the “bank lending channel” and not the “balance sheet channel”. See Black & Rosen (2007) for a clear distinction between the two channels, where the authors conduct an empirical analysis with the purpose of differentiating between the bank lending channel and the balance sheet channel, which are both part of the credit channel/view.

(7)

4

2. Literature Review

2.1.

Early studies

Bernanke and Blinder (1988) provide one of the earliest papers on the existence of a bank lending channel which, in their view, should be considered as a significant factor in conducting monetary policy. They introduce a modified version of the simple textbook IS-LM model, where they add credit as a third asset to the previous two-asset world (bonds and money in IS-LM), thereby arguing for the existence of a monetary transmission channel which works through what they call the “credit channel”.6 They further show some empirical evidence on the possible existence of a credit channel and how credit became more correlated than money with the gross national product (GNP) of the US.

Bernanke and Gertler (1995) summarize the role of the credit channel in monetary transmission (or what they refer to as the “black box”). They argue that while it is important to discuss the role of credit in monetary transmission, the credit view should not solely be considered as an independent channel of monetary transmission, but also as an amplifier of sorts. This is largely due to credit frictions in the market, which is generally referred to as the external finance

premium, which the authors also explain in their paper. The external finance premium is best

described as a wedge, reflecting the difference in internal financing where firms raise the necessary funds themselves and external financing, where funds are raised through a loan from the bank. The larger the difference, the larger the amplifying effect is from the credit channel on the real economy when the central bank conducts monetary policy (either contractionary or expansionary policies). The authors also provide a clear distinction within the credit channel, regarding the already discussed bank lending channel, and a balance sheet channel which is concerned with the borrower, and not the lender (the bank). The balance sheet channel implies that factors such as the borrowers net worth affects the size of the external finance premium, for example where wealthier borrowers can provide more collateral and hence increase the probability that the terms for their bank loan is more attractive/beneficial to them (in the case where a bank loan is necessary).

So far, this section on early studies of the bank lending channel has provided an overview on the important aspects of the bank lending channel, such as the inclusion of credit in IS-LM transforming it into a three-asset model, a distinction between the bank lending channel and balance channel, the external finance premium and its significance in conducting monetary

(8)

5

policy. As a final point, Kashyap and Stein (1994, 2000) contributes significantly to the topic of the bank lending channel by conducting a study on a large bank-data set covering all reportingbanks in the US, allowing for a thorough analysis with close to a million observations (quarterly data). They use a two-step regression in order to allow for different (macro) shocks in each period for different regions. Since their data is relatively large and is separated by the different federal reserve regions and further by categorizing banks by their asset size, this method is especially useful for their study. The difference here lies in the disaggregated data, compared to Bernanke and Blinder (1988), and therefore allows for more specified and interpretable results, however it makes the data collection and the analysis rather cumbersome. Furthermore, Kashyap and Stein (2000) find evidence for the existence a of bank lending channel, and they show that especially smaller banks which are more illiquid tend to be affected more by changes in monetary policy.7

2.2.

Bank lending and low/negative interest rates

Here, the focus lies on studies done on the bank lending channel post-2008. Ever since the financial crisis of 2008, central banks around the world have kept interest rates close to zero, and in some cases taken these to a negative territory. Monetary policy has certainly changed after the crisis, and thus the study on a bank lending channel should be reviewed with regards to these changes.

Borio and Gambacorta (2017) conducts an extensive study on an international level, where they use annual bank data from all major international banks, covering 20 years (1995-2014). They argue that in low interest rate environments, banks do not respond significantly to changes in the interest rate, and since margins on loan interest rates are already low, banks have increasingly sought different ways in making profit (e.g. wholesale banking). In their analysis they include real GDP growth and the 3-month interbank rate (similar to this paper) among other macroeconomic and bank-specific variables. They find statistically significant results which are in line with the expected results, meaning interbank rate negatively affects loan growth, however repeating the test where they only allow for a low interest rate environment, the effects are smaller and not statistically significant. Molyneux et al. (2020) also attempts to study bank lending in low/negative interest rate environment and find similar results for banks

7 The authors use three different ways of measuring monetary policy; the federal funds rate, and two other

custom-made measurements made by other authors. See the original paper’s description for measures of

(9)

6

in OECD countries where a negative interest rate policy (NIRP) has been applied. They apply a difference-in-differences regression analysis, and their data covers over 6500 banks from 2012-2016. They find that in these countries, bank lending was weaker than in countries where NIRP had not been adopted. Gambacorta and Marques-Ibanez (2011) provides a thorough analysis on bank lending and monetary policy by focusing especially on bank-specific characteristics. They cover more than 1000 banks from Europe and the US, and their quarterly data stretches from 1999 to 2009. While this does not cover the post-crisis period, it does focus on bank lending during the crisis, and they show that during a crisis the changes in bank lending are greater than otherwise normal circumstances.

2.3.

Quantitative easing

Here, I will briefly cover the effects of quantitative easing (QE) on bank lending.8 This is important since after the crisis of 2008, many central banks have turned to large scale asset purchases, including short-term treasury securities, and in some cases such as the Federal Reserve Bank in the US long-term treasuries and mortgage-backed securities. As QE has become a central topic in monetary policy, it is therefore important to discuss its implications in this context on bank lending.

Joyce and Spaltro (2014) studies the impact of QE as conducted by the Bank of England, where the purchases mainly consisted of government bonds. They use non-publicly available data from 1989 to 2010 consisting of 30 active financial institutions and find a small but statistically significant result that bank lending increased during the period when QE was being conducted. Lojschova (2017) focuses on the effects of the European Central Bank’s (ECB) asset purchase programmes and bank lending in Slovakia, using a data set consisting of 26 banks, ranging from 2009 to 2016. They use the Pooled Mean Group estimator to estimate both short and long-run relationships in their dynamic model. By using short-term interest rates and deposit ratios (similar to Joyce and Spaltro (2014)), they find that QE has increased both household consumption and bank lending. Rodnyansky and Darmouni (2017) carries out a similar study for the US, also showing statistically significant results, meaning as the central bank (in their case the FED) continues to purchase securities (government bonds and mortgage backed securities) bank lending increases (they use a difference-in-differences regression similar to

8 QE refers to the method of conducting expansionary monetary policy where the central bank commits to

(10)

7

Molyneux et al. (2020)). Finally, Bowman et al. (2015) studies the same relation in Japan, covering the period 2000-2009. Since the Bank of Japan (BOJ) was one of the first central banks to introduce QE (or rather QEP, quantitative easing policy as they called it), it is certainly important to study their case as well. Using a Generalized method of moments (GMM) procedure, the authors show that there is indeed a positive relation between the BOJ’s asset purchases and bank lending, however this effect was quite small but nonetheless statistically significant.

2.4.

Studies on Sweden

In Westerlund (2003), the author uses the 3-month Stockholm interbank offered rate (STIBOR) as the measure for monetary policy, and GDP is added as a control variable in the econometric model, similar to this paper. The analysis covers Swedish data from 1998-2003, and the results from the Full-information Maximum Likelihood (FIML) regression again imply that there appears to be bank lending channel in Sweden, since the results shows that as the three month STIBOR increases, the loan growth coefficients are negative, causing lending to the public to decline when contractionary policies take place. Hallsten (1998) uses the same model as in Bernanke and Blinder (1988) with a modification on the interest rate on bank loans, where the author instead uses the spread between the interest rate on bond and bank loans. Consistent with the previous literature mentioned above, the author finds that there is indeed a bank lending channel in Sweden. The analysis is conducted on Swedish data from 1986, and the measure for monetary policy is the overnight interest rate. The author uses the Autoregressive distributed-lag model (ARDL) in order to account for dynamic effects, where she explains that the amount of bank loans in previous periods play an important role in current and future levels of bank loan. This could, for example, be due to the relationship between the customer and the bank, where the customer might be inclined to stay with the bank for future loans as well (e.g. by creating different lock-in effects, making it costly to switch banks). This is also confirmed for this paper as well (see section 6.2.), seeing as bank loans are statistically significant (at the 1% level, lagged 1 quarter) and positively affects the supply of loans.

(11)

8

3. Theoretical Framework

The theoretical framework of this paper is based on the model from Bernanke & Blinder (1988), where the authors develop a model based on the textbook IS-LM model, where the investment and savings (IS) curve is replaced by a commodities and credit (CC) curve. While the model is quite simple, it does provide useful guidelines as to how the bank lending channel affects the real economy, along with some basic assumptions. Going forward, this model will for simplicity be referred to as the CC-LM model.

3.1.

CC-LM

To begin with, in the IS-LM model bonds and money are thought to be perfect substitutes in the standard two-asset LM curve. In the CC-LM model, this relationship is abandoned as a third asset, credit, is introduced and, following Bernanke and Blinder (1988), imperfect substitutability is assumed to hold between bonds and credit (bank loans). This is because some entities such as smaller and/or newly established firms do not have the necessary reputation yet to issue bonds in the securities markets in order to attract enough investment interest. As a result, these firms must turn to external sources of funding, such as applying for loan at a bank. This makes the issuance of bonds and applying for bank loans, both with the purpose of financial funding, imperfect substitutes.9 Now, both borrowers and lenders are concerned with two interest rates, the one on bonds and on bank loans. Next, I will define the three necessary market equilibrium conditions. Note that “bonds” in this context relates to the issuing of bonds with the purpose of creating credit for oneself (e.g. think of firms issuing bonds to the public in order to finance some investment). The problem essentially concerns the choice between internal (issuing bonds) and external (bank loans) funding.

3.1.1. The loan market

Let the interest rate on bank loans be denoted by 𝜌 and the interest on bonds by 𝑖, and suppose that banks cannot issue certificates of deposit (CD’s).10 In order to include aggregate demand for transactions in the economy, GDP/income is also included in the model, denoted by 𝑦, and

9 These informational problems are not as relevant for larger and well-known firms as it is for smaller and

younger firms, since they not only have the necessary reputation but also the resources available to conduct internal financing.

10 This is an assumption made in Bernanke and Blinder (1988) and Hallsten (1998), which makes the problem

easier to interpret later when the bank’s loan supply intensity is introduced in section 3.1.2. This makes the variable for the loan supply intensity to stay between 0 and 1.

(12)

9

it positively affects loan demand since higher aggregate demand increases new investment and consumption. Increasing bond interest rates affects loan demand positively, since the loan interest rate becomes more attractive than the bond interest rate (again, this concerns the bond interest rate a firm would have to pay to its bond holders). The demand for lending can now be formally expressed as:

𝐿𝑑 = 𝐿(𝜌

, 𝑖 +, 𝑦+)

As for the loan supply, a simplified balance sheet of a bank is presented with the following assets and liabilities:

Assets Liabilities

Reserves (R) Bonds (𝐵𝑏)

Loans (𝐿𝑠)

Deposits (D)

Reserves consists of two components; required reserves (𝜏𝐷) and excess reserves (E), where 𝜏 is the reserve ratio on deposits. Replacing R with 𝐸 + 𝜏𝐷, the adding-up constraint for the bank becomes; 𝐵𝑏+ 𝐿𝑠 + 𝐸 = 𝐷(1 − 𝜏). 𝐷(1 − 𝜏) is therefore the size of deposits which the bank can choose to divide between either supplying loans and/or buying bonds (for liquidity reasons, as loans are less liquid than bonds)11. Introducing a loan supply intensity variable, 𝜆, which is interpreted as the bank’s willingness to supply loans, the loan supply functions is expressed as 𝐿𝑠 = 𝜆 (𝜌

+, 𝑖−

) 𝐷(1 − 𝜏). This allows for the expression of a loan market equilibrium:

𝐿 (𝜌 −

, 𝑖

+, 𝑦+) = 𝜆 (𝜌+, 𝑖−) 𝐷(1 − 𝜏)

(1)

The supply coefficient λ is simply considered as the bank’s willingness to supply loans, and it can take any value between 0 and 1, where 1 means the bank is willing to supply all of its loanable funds 𝐷(1 − 𝜏) to the public.12 Note that in this context, it is assumed that banks

11 “Commitment” loans such as house loans, but also spot loans, are obviously less liquid than bonds which can

be sold in the markets directly.

12 As in Hallsten (1998), the loan supply coefficient can also be expressed as 𝜆 = 𝐿𝑠

𝐷(1−𝜏) which again means that

(13)

10

cannot issue certificate of deposits (CD’s). Loanable funds are, as mentioned above, defined as the total amount of deposits 𝐷 excluding the reserve requirement 𝜏𝐷.

3.1.2. The money market

The money market is derived by using the traditional LM-curve. On the demand side, the money market, which in this model is simplified as total demand of deposits where cash is ignored, is affected by the interest rate and income (transactions motive)13. For example, holding money is less attractive as the interest rate on bonds increase, causing an inverse relationship between money demand and interest rates.

Now turning to the supply side of money, the model introduces a variable 𝜀 in order to capture the bank’s holdings of excess reserves. Using the expression for their loanable funds (𝐷 − 𝜏𝐷 = 𝐷(1 − 𝜏)), the expression for a bank’s excess reserves becomes 𝐸 = 𝜀(𝑖)𝐷(1 − 𝜏). Note that the reserves R consists of excess reserves (E) and required reserves (𝜏𝐷). Hence, the money multiplier (e.g. how much lending the bank can conduct, which is determined by the central bank) becomes 𝑚(𝑖) = 1

[𝜀(𝑖)(1−𝜏)+𝜏]

14. The money multiplier times the bank reserves then equals the supply of deposits/money. The money market equilibrium results in the following expression:

𝐷𝑑( 𝑖

−, 𝑦+) = 𝑚 ( 𝑖+) 𝑅

(2)

The demand for money increases as wealth increases (transactions motive), and/or as interest rates decrease (bonds are less attractive). The supply of money increases as bank reserves increase (more liquidity) and as interest rates rise (for the same reason as in the demand side, which implies imperfect substitution between bonds and money).

3.1.3. The commodity market

The commodity/goods market follows the same reasoning in the textbook IS curve. Higher market interest rates and loan interest rates increases the cost of financing new investments.

13 Essentially, the transactions motive concerns the idea that people with higher income increases their money

demand (increased liquidity demand) because they are thought to spend more precisely because their income increase. Hence, transactions increase as income increases.

14 𝑅 = 𝐸 + 𝜏𝐷 => 𝑅 = 𝜀(𝑖)𝐷(1 − 𝜏) + 𝜏𝐷 =>𝑅

𝐷= 𝜀(𝑖)(1 − 𝜏) + 𝜏 => 𝑅 ∗ 1

(14)

11

This, naturally, leads to less economic activity and lower aggregate demand, eventually leading to decreased income. Equilibrium in the goods markets is therefore simply expressed as:

𝑦 = 𝑌 (𝜌 −

, 𝑖 −)

(3)

3.1.4. Graphical illustration of the CC-LM model

Using the equilibrium condition in the money/deposits market (eq. 2) to replace deposits (D) in the loan market (eq. 1), one can solve for the loan interest rate 𝜌 as a function of the interest rate (i), income (y), and reserves (R). Note that reserves are directly affected by monetary policy, hence the CC curve differs in an important way from the IS curve as it includes the effects of monetary policy. The loan market equilibrium becomes:

𝐿(𝜌, 𝑖, 𝑦) = 𝜆(𝜌, 𝑖)𝑚(𝑖)𝑅(1 − 𝜏) Using this to solve for the loan interest rate:

𝜌 = ∅( 𝑖

+, 𝑦+, 𝑅−) (4)

Using this expression (eq. 4) to replace the loan interest rate in equation 3, the resulting expression is what Bernanke and Blinder (1988) call the CC curve:

𝑦 = 𝑌(𝑖, ∅(𝑖, 𝑦, 𝑅)) (5)

Where the difference from the textbook IS model is in the monetary policy effect, captured by R, the amount of reserves the bank is required to hold.

(15)

12

Here, the CC line includes the standard IS line (y) but also the credit channel affected by monetary policy (R in equation 5). For example, suppose a shift occurs in the bank’s willingness to supply loans for some reason, such as looser reserve requirements which increases the amount of loanable funds for the bank or simply that the bank perceives loans to be less risky because of a change in some exogenous factor. This would cause the loan supply function 𝜆 (𝜌

+ , 𝑖

−) to increase, because banks are now supplying more loans and the bank loan interest rate would then have to decrease, all else equal. This would shift both the CC and LM curves outwards, increasing both GDP and interest rates. In this manner, the CC-LM model implies that the bank lending channel is “more” expansionary than in the standard IS-LM model, as the reserve requirements are included in both functions (CC and LM) here.

3.2.

Expected results

Following the literature (see section 2) and the theoretical framework presented previously, it is expected that as the Riksbank conducts expansionary monetary policies, banks should be incentivized to supply more loans, and therefore decrease the size of the wedge in terms of the external financial premium discussed on several occasions in this paper. As the focus is on banks and their choices, the Stockholm Interbank Offered Rate (STIBOR) is used as the policy indicator in order to measure how banks react to changes in monetary policy (see sections 1, 2 and 4 for further discussion on this). As Bernanke and Blinder (1988) find, the role of credit appears to be increasingly more important for the growth of a country with regards to spending and consumption. As an additional example, Westerlund (2003) also uses the STIBOR as a policy indicator, and GDP as an additional control variable, much like this paper, and finds that the results are in line with the expected results. That is, an increase in STIBOR (implying that contractionary policies have taken place) causes bank lending to decrease. Therefore, the expected results in this paper is to observe the same inverse relationship between monetary policy and bank lending.

(16)

13

4. Data

This paper uses quarterly aggregate data for banks active in Sweden between 1999-2019, including foreign banks which have active branches in Sweden. This is in order to avoid controlling for mergers, bankruptcies, and dissecting for other bank-specific deviations which would significantly increase the workload in collecting the data which is not possible with the limited time-frame of this paper. Instead, aggregate data from for all active banks in Sweden is used here, going back to 1999 without any missing values. The variable to be explained is total lending to the public from banks, and the main explanatory variable for this is some kind of measure of monetary policy, which in this paper is the 3-month Stockholm Interbank Offered Rate (STIBOR). Earlier papers on the bank lending channel (including Sweden) finds it suitable to use this as a measure of monetary policy, (see Westerlund (2003); Borio & Gambacorta (2017); Loupias et. al (2001)). Note that the STIBOR concerns the (nominal) rate offered between the major banks in Sweden, namely the STIBOR-banks.15 The STIBOR/interbank rate is a measure of the average interest rate on the loans made between the STIBOR banks and therefore acts as a good measure of the bank’s stance with regards to the economic climate and monetary policy, and is therefore used frequently in these types of studies (see the papers mentioned in this section and section 1).

Other variables are also accounted for in the model presented in this paper. The liquidity ratio of the banking system (again, using aggregated data) is used to capture the liquidity position of the banks and control for this (see e.g. Molyneux et. al (2020); Westerlund (2003); Loupias et. al (2001)). The liquidity ratio is calculated as the total liabilities and equity over total assets, meaning if the banking system has a liquidity ratio of 1, it is able to repay all of the liabilities and hence have a good liquidity position. The liquidity ratio has generally been around 0.9 in Sweden, with very small fluctuations (see table 1). Total assets for all banks are also included in the model, as asset size has proven to be an important factor in loan supply growth (see Kashyap and Stein (2000)). Finally, in order to account aggregate demand in the economy and inflation, real GDP is included as a macroeconomic variable simply because as spending, investments, and consumption increases, so will eventually the need for credit as well.16 The “transactions motive” discussed in section 3.1.2. is certainly applicable here. All variables were

15 Nordea, Swedbank, SEB, Handelsbanken, Danske Bank, and Länsförsäkringar Bank.

16 Several papers use real rather than nominal GDP. Kashyap and Stein (1994) conducts two tests, by using

nominal GDP once and real on the second test, however most of the results stay the same. Additionally, Borio and Gambacorta (2017) and Molyneux et al. (2020) uses real GDP as well in their model.

(17)

14

available in their original quarterly form. GDP data was collected from the Federal Reserve Economic Database (FRED), Statistics Sweden (SCB) provided the bank data such as assets and liabilities, and the STIBOR was available at the Riksbank statistics database. Real GDP (RGDP), total assets (TA), and total lending (LOAN) have all been transformed to their natural logarithm form while the liquidity ratio (LIQR, taking on values between 0 and 1) and STIBOR have not been changed. The following table provides some summary statistics for dataset used in this paper:

Table 1

Summary of dataset (1999Q2-2019Q4)

Variable Observations Mean Std. Deviation Min Max

𝐿𝑂𝐴𝑁 83 2815275 1088100 1203057 4831492 𝑆𝑇𝐼𝐵𝑂𝑅 83 1.975663 1.729406 -0.57 5.18 𝑇𝐴 83 3932995 1502581 1734904 6604759 𝐿𝐼𝑄𝑅 83 0.8957795 0.0285988 0.8369 0.9525 𝑅𝐺𝐷𝑃 83 881406.1 114485 674844.4 1081126 𝑙𝑛𝐿𝑂𝐴𝑁 83 14.76963 0.4162593 14.00038 15.39067 𝑙𝑛𝑇𝐴 83 15.10434 0.4165247 14.36646 15.7033 𝑙𝑛𝑅𝐺𝐷𝑃 83 13.68084 0.1311814 13.42224 13.89351

As mentioned before, the liquidity ratio does not vary too much, going from a minimum value of 0.83 to its highest value of 0.95. Additionally, the STIBOR takes on negative values here, which is a key difference from the earlier studies on Sweden mentioned previously (Hallsten (1998) and Westerlund (2003)) when studying monetary policy.

(18)

15

5. Methodology

5.1.

Time-series data and choice of econometric model

5.1.1. Vector Error Correction Model (VECM)

The purpose of this study is to examine the relationship between bank lending (total, including households and firms) and monetary policy in Sweden. To do this, the natural logarithm of the lending growth is regressed on the 3-month Stockholm Interbank Offered Rate (STIBOR), as well as other macroeconomic and bank-specific variables (presented in section 5.1.3). See table 1 in section 4 for a further discussion on the selected variables. Further, as the aim is to study how monetary policy affects bank lending over time, and as the variables are most likely interdependent, the Vector Error Correction Model (VECM) is the appropriate model to use as it shows both long- and short-run estimations and has been used in other similar studies (see Ramey (1993); De Bondt (1999); Hülsewig et. al (2004)). The VECM is suitable to use when there is evidence of cointegration between the variables, while the Vector Autoregression model (VAR) is the correct model if there is no cointegration and hence only the short-run coefficients can be estimated. The VECM will show the long-run coefficients for and within the error correction term (ECT) as well as the differenced VAR model which shows the short-run coefficients. The speed of adjustment, which is the coefficient for the ECT, shows how “fast” the errors from the previous time-period (quarters in this case) are corrected for within the current time-period. This section will continue by introducing the Johansen test for cointegration and Augmented Dickey-Fueller test for stationarity, and finally conclude with the presentation of the econometric model.

5.1.2. Stationarity and Cointegration

When conducting time-series analysis, an important factor is checking for stationarity in the variables. For a series to be stationary, they must have the same probability distributions (i.e., the same expected value, variance, and covariance for all time-periods). Without this, the model used might suffer from spurious regression and unreliable results. If, however, the series appears to be non-stationary, that would mean it has a unit root. One could begin by observing a graphical representation of the way the series behave, however one should always conduct an

(19)

16

appropriate unit root test to check for this issue. Here, the Augmented Dickey-Fueller (ADF) test is most commonly used and is expressed as follows17 :

∆𝑌𝑡= 𝛽0+ 𝛿𝑌𝑡−1+ ∑ 𝛽𝑖∆𝑌𝑡−𝑖 𝑝 𝑖=1 + 𝜀𝑡 𝐻0: 𝛿 = 0, 𝑌𝑡 𝑖𝑠 𝑛𝑜𝑡 𝑠𝑡𝑎𝑡𝑖𝑜𝑛𝑎𝑟𝑦 𝐻𝐴: 𝛿 < 0, 𝑌𝑡 𝑖𝑠 𝑠𝑡𝑎𝑡𝑖𝑜𝑛𝑎𝑟𝑦

Here, 𝛽0 is the constant term and 𝑝 the number of lags. ∆𝑌𝑡 the first difference of 𝑌𝑡 (i.e. 𝑌𝑡− 𝑌𝑡−1) and 𝛿, 𝛽𝑖 are coefficients. It is also important to use an appropriate test for choosing the number of lags for this test18. Here, the method presented in Ng and Perron (1995) will be used for this purpose, where the optimal number of lags is calculated by setting an upper bound for 𝑝* so that 𝑝*= 𝑝𝑚𝑎𝑥. Using this upper bound as a first step, one checks for the value of the t-statistic and whether it is larger than 1.6. If it is, then the 𝑝*= 𝑝𝑚𝑎𝑥. If not, one simply reduces the lag length with one unit and runs the test again until the first statement holds. Schwert (2002) presents a common “rule of thumb” for choosing 𝑝𝑚𝑎𝑥:

𝑝𝑚𝑎𝑥= (12 ∗

𝑇 100

0.25

)

Finally, in order to check for cointegration this paper uses the Johansen test for cointegration which tells how many cointegrating relationship there are in model. The Johansen test produces a trace statistic which can be used to decide the number of cointegrating equations by observing the maximum rank (which is equal to the maximum number of cointegrating equations). The null hypothesis is simply that there are no cointegrating relationship, while the alternative hypothesis suggests that there are cointegrating equations in the model, which is based on the maximum rank. The maximum rank is observed where the trace statistic is smaller than the 5% critical value. The results presented in section 6 and table I in Appendix shows that the maximum rank is two, meaning we can estimate the long-run coefficients by using the VECM ,instead of only using a VAR model, and thereby observe two long-run equations.

17 Based on the Dickey-Fuller test presented in Dickey & Fuller (1979)

18 If the number of lags are too high, the power of the model will be weak, while a too small number of k lags

(20)

17

5.1.3. Econometric model (VAR and VECM)

As mentioned in the previous section, the Johansen test showed a rank of 2, meaning we have two long-term equations to consider. These equations will be presented shortly, while the VAR(k) is presented here:

𝑦𝑡 = 𝛼 + 𝐴1𝑦𝑡−1+ 𝐴2𝑦𝑡−2+ ⋯ + 𝐴𝑘𝑦𝑡−𝑘+ 𝜀𝑡

Since we are using five variables in our model, 𝑦𝑡 is therefore a vector with length 5, and 𝐴𝑡 is a 5x5 matrix. In our model, we found that the optimal amount of lags is 3 (see table II in Appendix).

However, as the Johansen test showed a maximum rank of 2, it means our long-term econometric model (VECM) needs to consider two long-term relations/error correction terms, identified through the Johansen normalization (see Johansen (1995)). In this section, we begin by introducing the VECM model in equation forms, whereas the complete output for both short- and long-run results are presented in section 6.2, tables 4-6.

The econometric model (VECM) used in this paper takes the following equation forms:

∆𝑙𝑛𝐿𝑂𝐴𝑁𝑡= 𝛼1+ ∑ 𝛽1,𝑖∆𝑙𝑛𝐿𝑂𝐴𝑁𝑡−𝑖 𝑘−1 𝑖=1 + ∑ 𝜗1,𝑗∆𝑆𝑇𝐼𝐵𝑂𝑅𝑡−𝑗 𝑘−1 𝑗=1 + ∑ 𝜃1,𝑚∆𝑙𝑛𝑇𝐴𝑡−𝑚 𝑘−1 𝑚=1 + ∑ 𝜑1,𝑛∆𝐿𝐼𝑄𝑅𝑡−𝑛 𝑘−1 𝑛=1 + ∑ 𝜔1,𝑟∆𝑙𝑛𝑅𝐺𝐷𝑃𝑡−𝑟 𝑘−1 𝑟=1 + 𝛾1𝐸𝐶𝑇1,𝑡−1+ ℶ1𝐸𝐶𝑇2,𝑡−1 + 𝜀1,𝑡 (5.1) ∆𝑙𝑛𝑇𝐴𝑡 = 𝛼2+ ∑ 𝜃2,𝑚∆𝑙𝑛𝑇𝐴𝑡−𝑚 𝑘−1 𝑚=1 + ∑ 𝛽2,𝑖∆𝑙𝑛𝐿𝑂𝐴𝑁𝑡−𝑖 𝑘−1 𝑖=1 + ∑ 𝜗2,𝑗∆𝑆𝑇𝐼𝐵𝑂𝑅𝑡−𝑗 𝑘−1 𝑗=1 + ∑ 𝜑2,𝑛∆𝐿𝐼𝑄𝑅𝑡−𝑛 𝑘−1 𝑛=1 + ∑ 𝜔2,𝑟∆𝑙𝑛𝑅𝐺𝐷𝑃𝑡−𝑟 𝑘−1 𝑟=1 + 𝛾2𝐸𝐶𝑇1,𝑡−1+ ℶ2𝐸𝐶𝑇2,𝑡−1 + 𝜀2,𝑡 (5.2) ∆𝑆𝑇𝐼𝐵𝑂𝑅𝑡 = 𝛼3+ ∑ 𝜗3,𝑗∆𝑆𝑇𝐼𝐵𝑂𝑅𝑡−𝑗 𝑘−1 𝑗=1 + ∑ 𝛽3,𝑖∆𝑙𝑛𝐿𝑂𝐴𝑁𝑡−𝑖 𝑘−1 𝑖=1 + ∑ 𝜃3,𝑚∆𝑙𝑛𝑇𝐴𝑡−𝑚 𝑘−1 𝑚=1 + ∑ 𝜑3,𝑛∆𝐿𝐼𝑄𝑅𝑡−𝑛 𝑘−1 𝑛=1 + ∑ 𝜔3,𝑟∆𝑙𝑛𝑅𝐺𝐷𝑃𝑡−𝑟 𝑘−1 𝑟=1 + 𝛾3𝐸𝐶𝑇1,𝑡−1+ ℶ3𝐸𝐶𝑇2,𝑡−1 + 𝜀3,𝑡 (5.3)

(21)

18 ∆𝐿𝐼𝑄𝑅𝑡 = 𝛼4+ ∑ 𝜑4,𝑛∆𝐿𝐼𝑄𝑅𝑡−𝑛 𝑘−1 𝑛=1 + ∑ 𝛽4,𝑖∆𝑙𝑛𝐿𝑂𝐴𝑁𝑡−𝑖 𝑘−1 𝑖=1 + ∑ 𝜗4,𝑗∆𝑆𝑇𝐼𝐵𝑂𝑅𝑡−𝑗 𝑘−1 𝑗=1 + ∑ 𝜃4,𝑚∆𝑙𝑛𝑇𝐴𝑡−𝑚 𝑘−1 𝑚=1 + ∑ 𝜔4,𝑟∆𝑙𝑛𝑅𝐺𝐷𝑃𝑡−𝑟 𝑘−1 𝑟=1 + 𝛾4𝐸𝐶𝑇1,𝑡−1+ ℶ4𝐸𝐶𝑇2,𝑡−1 + 𝜀4,𝑡 (5.4) ∆𝑙𝑛𝑅𝐺𝐷𝑃𝑡 = 𝛼5+ ∑ 𝜔5,𝑟∆𝑙𝑛𝑅𝐺𝐷𝑃𝑡−𝑟 𝑘−1 𝑟=1 + ∑ 𝛽5,𝑖∆𝑙𝑛𝐿𝑂𝐴𝑁𝑡−𝑖 𝑘−1 𝑖=1 + ∑ 𝜗5,𝑗∆𝑆𝑇𝐼𝐵𝑂𝑅𝑡−𝑗 𝑘−1 𝑗=1 + ∑ 𝜃5,𝑚∆𝑙𝑛𝑇𝐴𝑡−𝑚 𝑘−1 𝑚=1 + ∑ 𝜑5,𝑛∆𝐿𝐼𝑄𝑅𝑡−𝑛 𝑘−1 𝑛=1 + 𝛾5𝐸𝐶𝑇1,𝑡−1+ ℶ5𝐸𝐶𝑇2,𝑡−1 + 𝜀5,𝑡 (5.5)

Where the error correction terms ECT are expressed as:

𝐸𝐶𝑇1,𝑡−1= [1 ∗ 𝑙𝑛𝐿𝑂𝐴𝑁𝑡−1− 𝛿1,𝑗𝑆𝑇𝐼𝐵𝑂𝑅𝑡−1− 𝜇1,𝑚𝑙𝑛𝑇𝐴𝑡−1− 𝜎1,𝑛𝐿𝐼𝑄𝑅𝑡−1 − 𝜋1,𝑟𝑙𝑛𝑅𝐺𝐷𝑃𝑡−1] (5.6) 𝐸𝐶𝑇2,𝑡−1= [1 ∗ 𝑙𝑛𝑇𝐴𝑡−1− 𝛿2,𝑗𝑆𝑇𝐼𝐵𝑂𝑅𝑡−1− ℵ2,𝑠𝑙𝑛𝐿𝑂𝐴𝑁𝑡−1− 𝜎2,𝑛𝐿𝐼𝑄𝑅𝑡−1 − 𝜋2,𝑟𝑙𝑛𝑅𝐺𝐷𝑃𝑡−1] (5.7)

Note that lag length is now reduced to 𝑘 − 1. 𝛽, 𝜗, 𝜃, 𝜑, 𝜔 are now the coefficients for the short-run dynamic adjustment for the long-run model. 𝛾 is the speed of adjustment, where a negative number would suggest that the variable converges to a long-run equilibrium (or in other words correcting for any disequilibria) with some percent each time-period (speed of adjustment). These results and the speed of adjustment to a long-run equilibrium are all presented throughout section 6.

Additionally, in order to check whether the time-series data in this paper suffers from autocorrelation in the residuals, the Breusch-Godfrey LM (Lagrange Multiplier) test will be applied.

(22)

19

6. Results

This section will present the estimation results from the Vector Error Correction Model (VECM) as well as the results from the Augmented Dickey-Fueller test. Some results and outputs for other tests which are considered unnecessary to present here can instead be found in the appendix. Later, section 7 attempts to further discuss the results presented here and draw some thoughtful conclusions, along with some comparisons with previous similar studies.

6.1.

Stationarity and the Augmented Dickey-Fueller test

A good starting point when conducting time-series analysis is to test whether the time-series are stationary or not (in order to avoid spurious regression). First, plotting the series on a graph is a useful first step in observing the data and checking for stationarity in their level form. After this, the same type of graph is constructed using the first-difference form of the series. See figures 2 and 3 below.

Figure 2 suggests that several of the time-series may be non-stationary in level form. Figure 3 shows that all the first differences revolve around 0, although STIBOR experienced a significantly big drop as consequence of the (extremely) stimulating monetary policies set in effect in response to the crisis of 2008. However, the following output from the ADF-test shows that in their first difference form all variables appear to be stationary:

Figure 3 Figure 2

(23)

20

Table 2

Augmented Dickey-Fueller test with 4 lags19 (Checking against the 5% critical value of -2.908) 𝐻0: 𝑒𝑥𝑖𝑠𝑡𝑒𝑛𝑐𝑒 𝑜𝑓 𝑎 𝑢𝑛𝑖𝑡 𝑟𝑜𝑜𝑡 𝐻𝐴: 𝑛𝑜 𝑢𝑛𝑖𝑡 𝑟𝑜𝑜𝑡

Variable ADF test-statistic

Level form 𝑙𝑛𝐿𝑂𝐴𝑁 𝑆𝑇𝐼𝐵𝑂𝑅 𝑙𝑛𝑇𝐴 𝐿𝐼𝑄𝑅 𝑙𝑛𝑅𝐺𝐷𝑃 -1.179 -1.773 -1.280 -1.862 -0.482 First difference 𝑑. 𝑙𝑛𝐿𝑂𝐴𝑁 𝑑. 𝑆𝑇𝐼𝐵𝑂𝑅 𝑑. 𝑙𝑛𝑇𝐴 𝑑. 𝐿𝐼𝑄𝑅 𝑑. 𝑙𝑛𝑅𝐺𝐷𝑃 -3.012 -4.325 -3.069 -4.901 -5.427

If the test-statistic is greater than the absolute critical value of -2.908, the null hypothesis for the existence of a unit root in the time-series data is rejected, and hence the data is said to be stationary. The output for the ADF-test gives clear evidence of stationarity in the variables at their respective first difference forms, meaning they are all I(1), and since the Johansen test for cointegration shows that there are at least two cointegrating equations (see table I in Appendix), the VECM is the suitable model in this case. Having resolved the issue of stationarity, the next step is to run the VECM model.

6.2.

Vector Error Correction Model (VECM)

The results from the Johansen test for cointegration (see table I in Appendix) implied that there are two cointegrating equations in our model, and hence the Vector Error Correction Model (VECM) is suitable to use. The results from VECM presented here in table 4 first shows the short-run coefficients, including the speed of adjustments for the ECT’s, followed by table 5 and 6 which presents both error correction terms and hence show the long-run results.

(24)

21

Additionally, the Breusch-Godfrey LM test shows that there is no autocorrelation in the residuals, see table 3 below.

Table 3, Breusch-Godfrey LM test

Breusch-Godfrey LM test for autocorrelation

𝐻0 = 𝑁𝑜 𝑎𝑢𝑡𝑜𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 𝑎𝑡 𝑙𝑎𝑔 𝑜𝑟𝑑𝑒𝑟𝑠 Lag Chi2 df Prob > chi2 1 36.5612 25 0.635 2 50.8655 25 0.166

First, the optimal lag order was decided by using the Akaike Information Criterion (AIC), where the output showed that 3 was the optimal number of lags using either AIC or BIC, as both had the same number of lags as the optimal lag order (see table II in Appendix for further information). Further, the trace statistic from the Johansen test for cointegration showed a maximum rank of two, implying that there are two cointegrating equations (see table I in Appendix). As a result of this, we can estimate the VECM model and observe its long- and short-run coefficients and the respective speed of adjustments. For this, see tables 4-6 below to observe the complete output for the VECM:

Table 4, Short-run coefficients of the VECM

Estimated short-run coefficients from the VECM (including adjustment term)

Time-period: 2000Q1-2019Q4, N=80 observations, 𝑅2 = 0.4716

𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒 (−𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑙𝑎𝑔𝑠) Coefficient Standard error

Dependent variable = ∆𝒍𝒏𝑳𝑶𝑨𝑵 𝛾1 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) -0.4323*** 0.1238 ℶ1 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) 0.2431 0.0806 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−1) -0.0196 0.1739 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−2) -0.0722 0.1599 ∆𝑙𝑛𝑇𝐴(−1) 0.4797 0.1387 ∆𝑙𝑛𝑇𝐴(−2) -0.0505 0.1489 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−1) 0.0001 0.0114 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−2) 0.0010 0.0104

(25)

22 ∆𝐿𝐼𝑄𝑅(−1) -0.2975 0.2068 ∆𝐿𝐼𝑄𝑅(−2) -0.3682 0.2123 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−1) 0.1321 0.4619 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−2) 0.1987 0.4374 𝛼 (constant) 0.0078 0.0139 Dependent variable = ∆𝒍𝒏𝑻𝑨 𝛾2 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) 0.1561 0.1708 ℶ2(𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) -0.2810*** 0.1112 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−1) 0.1515 0.2398 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−2) -0.1260 0.2205 ∆𝑙𝑛𝑇𝐴(−1) -0.2800 0.1913 ∆𝑙𝑛𝑇𝐴(−2) 0.0546 0.2054 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−1) 0.0263 0.0157 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−2) -0.0147 0.0143 ∆𝐿𝐼𝑄𝑅(−1) 0.0590 0.2852 ∆𝐿𝐼𝑄𝑅(−2) -0.8677 0.2928 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−1) 0.5946 0.6370 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−2) 0.6343 0.6032 𝛼 (constant) 0.0168 0.0191 Dependent variable = ∆𝑺𝑻𝑰𝑩𝑶𝑹 𝛾3 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) 0.0503 0.3671 ℶ3 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) -0.1082 0.8907 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−1) 1.7809 1.9201 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−2) 3.052 1.7656 ∆𝑙𝑛𝑇𝐴(−1) -3.5992 1.5314 ∆𝑙𝑛𝑇𝐴(−2) -2.1041 1.6444 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−1) 0.2714 0.1261 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−2) -0.0656 0.1152 ∆𝐿𝐼𝑄𝑅(−1) -0.5321 0.2829 ∆𝐿𝐼𝑄𝑅(−2) 0.2035 0.3438 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−1) 14.3915** 5.0991 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−2) 19.8348*** 4.4828

(26)

23 Statistically significant at 10%=*, 5%=**, 1%=*** 𝛼 (constant) 0.0023 0.1535 Dependent variable = ∆𝑳𝑰𝑸𝑹 𝛾4 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) 0.0738 0.0751 ℶ4 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) -0.0525 0.0489 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−1) -0.0748 0.1054 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−2) -0.0028 0.0969 ∆𝑙𝑛𝑇𝐴(−1) -0.0139 0.0841 ∆𝑙𝑛𝑇𝐴(−2) -0.0002 0.0903 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−1) 0.0033 0.0069 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−2) -0.0113 0.0063 ∆𝐿𝐼𝑄𝑅(−1) -0.2133 0.1254 ∆𝐿𝐼𝑄𝑅(−2) -0.0694 0.1287 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−1) 0.4598 0.2801 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−2) 0.1341 0.2652 𝛼 (constant) 0.0158 0.0084 Dependent variable = ∆𝒍𝒏𝑹𝑮𝑫𝑷 𝛾5 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) 0.0309 0.0313 ℶ5 (𝑠𝑝𝑒𝑒𝑑 𝑜𝑓 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡) -0.0135 0.0203 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−1) -0.1140*** 0.0439 ∆𝑙𝑛𝐿𝑂𝐴𝑁(−2) -0.0129 0.0404 ∆𝑙𝑛𝑇𝐴(−1) 0.1176*** 0.0350 ∆𝑙𝑛𝑇𝐴(−2) 0.0339 0.0376 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−1) -0.0049 0.0028 ∆𝑆𝑇𝐼𝐵𝑂𝑅(−2) -0.0028 0.0026 ∆𝐿𝐼𝑄𝑅(−1) 0.0865 0.0522 ∆𝐿𝐼𝑄𝑅(−2) 0.0243 0.0536 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−1) -0.0265 0.1167 ∆𝑙𝑛𝑅𝐺𝐷𝑃(−2) 0.0316 0.1105 𝛼 (constant) 0.0230 0.0035

(27)

24

Table 5, Long-run coefficients (𝐄𝐂𝐓𝟏)

Estimated long-run coefficients: 𝐸𝐶𝑇1 Speed of adjustment 𝛾1= −0.4323

Variable (-number of lags) Coefficient Standard error

𝑙𝑛𝐿𝑂𝐴𝑁 1 - 𝑆𝑇𝐼𝐵𝑂𝑅 0.69396*** 0.09417 𝑙𝑛𝑇𝐴 -0.01116 0.00015 𝐿𝐼𝑄𝑅 2.41016 2.58889 𝑙𝑛𝑅𝐺𝐷𝑃 -12.1929*** 1.25194 𝛼 (constant) 153.0864 - Statistically significant at 10%=*, 5%=**, 1%=***

Table 6, Long-run coefficients (𝐄𝐂𝐓𝟐)

Estimated long-run coefficients: 𝐸𝐶𝑇2 Speed of adjustment ℶ2 = −0.2810

Variable (-number of lags) Coefficient Standard error

𝑙𝑛𝑇𝐴 1 - 𝑙𝑛𝐿𝑂𝐴𝑁 -1.44093 1.25605 𝐿𝐼𝑄𝑅 4.37877 4.01419 𝑙𝑛𝑅𝐺𝐷𝑃 -17.07618*** 1.94119 𝑆𝑇𝐼𝐵𝑂𝑅 1.05651*** 0.14601 𝛼 (constant) 73.7451 -

Looking at the error correction terms (from equations 5.6-5.7), with the speed of adjustment included (coefficients for the ECTs) and the expression for our error correction terms ( 𝛾1𝐸𝐶𝑇1

and ℶ2𝐸𝐶𝑇2) they are expressed as follows (note that the signs from the table become reversed in the ECT):

−0.4323 ∗ 𝐸𝐶𝑇1,𝑡−1

= −0.4323[1 ∗ 𝑙𝑛𝐿𝑂𝐴𝑁𝑡−1− 0.69396𝑆𝑇𝐼𝐵𝑂𝑅𝑡−1+ 0.01116𝑙𝑛𝑇𝐴𝑡−1

− 2.41016𝐿𝐼𝑄𝑅𝑡−1+ 12.1929𝑙𝑛𝑅𝐺𝐷𝑃𝑡−1]

(28)

25

−0.2810 ∗ 𝐸𝐶𝑇2,𝑡−1

= −0.2810[1 ∗ 𝑙𝑛𝑇𝐴𝑡−1− 1.05651𝑆𝑇𝐼𝐵𝑂𝑅𝑡−1+ 1.44093𝑙𝑛𝐿𝑂𝐴𝑁𝑡−1

− 4.37877𝐿𝐼𝑄𝑅𝑡−1+ 17.07618𝑙𝑛𝑅𝐺𝐷𝑃𝑡−1]

(6.2)

Where we have expressed the signs accordingly and added the values from the long-run output to end up with the final expressions for our error correction terms with the correct speed of adjustments.

The adjustment term 𝛾1= −0.4323 suggests that deviations from the long-run equilibrium in our model are corrected for with a speed of 43.23% in the current quarter. Further, the Stockholm interbank lending rate shows a small but negative effect on the supply of loans, implying that contractionary monetary policy (higher STIBOR) causes banks to reduce their lending. Real GDP shows a positive effect on lending, which could mean that when the economy is booming, lending also increases as demand for credit increases. Both coefficients are statistically significant at the 1% level.

Total assets have a positive effect on loans, meaning that as the banks’ balance sheet grows, it also positively affects their supply of loans. The results also show that a higher liquidity rate (increasing constraints on banks) imposed on the banks causes lending to decrease. Neither of these results however are statistically significant at the 1%, 5%, or 10% level, making them inappropriate to base any conclusions on. However, by merely observing the coefficients we can at least argue that they are in line with the assumptions made throughout this paper, although they are not statistically significant.

The short-run coefficients were mostly statistically insignificant, however some of the coefficients still provides valuable economic interpretations. Here, some of the results are briefly summarized while they are more thoroughly discussed in the next section. ∆𝑙𝑛𝐿𝑂𝐴𝑁 is negatively affected by an increase in the liquidity ratio (contractionary monetary policy) in both the first and second quarter, while real GDP had a positive impact on the supply of loans (booming economy, higher demand for credit). ∆𝑆𝑇𝐼𝐵𝑂𝑅 and an increasing balance sheet (∆𝑙𝑛𝑇𝐴) both showed a positive effect on the supply of loans. ∆𝑙𝑛𝑇𝐴 as a dependent variable was positively affected by both real GDP and the supply of loans, while liquidity ratio had a negative effect when lagged two quarters. Both the supply of loans and real GDP had a positive effect on ∆𝑆𝑇𝐼𝐵𝑂𝑅, which could imply that tighter monetary policy is imposed as the economy heats up and perhaps shows signs of overheating. Real GDP is negatively affected by an increasing STIBOR, while an increasing supply of loans positively affects real GDP. As for the

(29)

26

respective speed of adjustments, none where statistically significant except for the ones presented in the long-run output (𝛾1 and ℶ2)20.

As observed, the short-run coefficients were mostly statistically insignificant, while changes in real GDP in both quarters were statistically significant at the 1% level and affected the STIBOR positively. Additionally, ∆𝑙𝑛𝑇𝐴 and ∆𝑙𝑛𝐿𝑂𝐴𝑁 both lagged one quarter had a positive effect on real GDP and were both statistically significant at the 1% level. These again confirm the idea that expansionary monetary policy potentially causes banks to increase their supply of loans, and by doing this the economy continues to heat up as credit is “more easily” available and looser requirements increases the incentive to borrow.

As we saw from the Johansen test for cointegration, there are two cointegrating equations to consider. It is therefore necessary to present these results which represent the second error correction term 𝐸𝐶𝑇2 from equation 5.7, with the correct statistically significant negative coefficient, showing the speed of which potential disequilibria are corrected for in each time-period. Here, it appears the size of the banks total assets (𝑙𝑛𝑇𝐴) are also negatively affected by the STIBOR, which is also is in line with the empirical literature presented in section 2. Real GDP positively affects the banks total assets, again confirming the idea that as the economy is booming, economic activity ramps up and this is beneficial for the bank. Both these results are statistically significant at the 1% level. The coefficient (adjustment term ℶ2 = −0.2810) for this second error correction term is also statistically significant at the 1% level and has the expected negative sign, meaning deviations from the long-run equilibrium are corrected for with a speed of 28.10% current quarter.

20 Some of the coefficients which had a positive sign indicates that the variables are not converging to a long-run

(30)

27

7. Discussion and conclusion

The purpose of this paper was to study the existence of a bank lending channel in Sweden, using the Stockholm Interbank Offered Rate STIBOR as a measure of monetary policy. Although Hallsten (1999) and Westerlund (2003) conducts a similar study, their studies are quite old and hence do not cover the recent periods with low and negative interest rates in Sweden. Using aggregate bank data for Sweden and covering the past 20 years (1999-2019), this paper attempts to shed new light on the bank lending channel and its current relevance. Similar to Westerlund (2003), this paper uses STIBOR as a monetary policy indicator for the banks, where contractionary monetary policy is observed through a higher STIBOR. By observing the long-run results from table 5in section 6.2, one can observe that increases in STIBOR (as a result of contractionary monetary policy) causes the banks’ lending to the public to decrease. Further, as a measure of general economic activity and prosperity, real GDP was used and it was also in line with our expected results, implying that as real GDP increases, it causes the supply of loans made by banks to increase as well. Both these results were statistically significant at the 1% level and in line with the empirical and theoretical literature. Additionally, while the short-run results were not statistically significant, they were mostly in line with our expectations. Real GDP and total assets had a positive impact on the supply of loans, showing signs of a booming economy which causes the demand for credit to increase. STIBOR however also showed a positive effect on loans, which is not in line with the literature or our expectations from section 3. One argument for this might be that as the economy is overheating, the central bank reacts by raising rates, but cannot combat the overheating economy in the short-run. As we saw for the long-run results however, these variables are (expectedly) inversely related.

Furthermore, total assets (𝑙𝑛𝑇𝐴) also appears to have a positive impact on the supply of loans, which is in line the expected results of this paper and as presented in Kashyap and Stein (1994, 20009), while a higher liquidity ratio (LIQR) (regulations imposed on banks) decreases their lending activity. Although none of these two variables were statistically significant, they nonetheless still provide some economically interpretable results, specifically with regards to the literature presented throughout section 2. As the banks’ total assets increase, for example as a result from either higher loan demand or firm-specific factors, it concurrently increases the supply of loans made to the public. The liquidity ratio imposes restrictions on the banks as they must keep some level of liquidity, where a higher ratio would mean the bank must hold more liquid assets on their balance sheet and hence their lending capacity decreases.

(31)

28

As mentioned previously, most of the short-run results were statistically insignificant, but they can nonetheless still provide some useful economic interpretations Total assets had a seemingly expected short-run results. As the bank’s lending activity increases, its balance sheet increases as well in terms of outstanding debt. The STIBOR and liquidity ratio both had negative signs when lagged two quarters, implying that an inverse relationship could be interpreted as contractionary policies are in effect. The liquidity ratio is positively affected by an increase in the STIBOR for both quarters, which could be the results of contractionary monetary policy affecting both aspects in the expected manner, higher rates and higher required liquidity for the bank. While most short-run results showed statistical insignificance, there were some variables which were statistically significant. The change in supply of loans and total assets for the bank in the short-run had a positive and statistically significant effect on real GDP when lagged one quarter, again confirming the idea that increased lending activity helps the economy to heat up. Conclusively, this study has presented long-run empirical evidence on the existence of a bank lending channel in Sweden, covering quarterly data between 1999-2019. As the previous research on a bank lending channel in Sweden (e.g., Hallsten (1998) and Westerlund (2003)) was studied before the financial crisis of 2008 and before negative repo-rates were introduced, this paper attempt to capture this new aspect of monetary policy as well. The results for the long-run, i.e. the coefficients for the error correction terms in the VECM, suggests that there is a bank lending channel (BLC) in Sweden, and therefore confirms the idea that the BLC, as part of the Credit Channel, works as a monetary transmission channel in Sweden. The results for both of the error correction terms were statistically significant and showed that our measure of monetary policy, the STIBOR, had the expected inverse relationship in the long-run with both the supply of loans made by banks, but also on the size of the bank’s balance sheet.

(32)

29

References

Akaike, H. (1974). A new look at the statistical model identification. IEEE transactions on

automatic control, 19(6), 716-723.

Bernanke, B. S., & Blinder, A. S. (1988). Credit, money, and aggregate demand (No. w2534). National Bureau of Economic Research.

Bernanke, B. S., & Gertler, M. (1995). Inside the black box: the credit channel of monetary policy transmission. Journal of Economic perspectives, 9(4), 27-48.

Black, L. K., & Rosen, R. K. (2007). How the credit channel works: Differentiating the bank lending channel and the balance sheet channel. (No. 2007-13). Working Paper, Federal Reserve Bank of Chicago, Chicago, IL

Borio, C., & Gambacorta, L. (2017). Monetary policy and bank lending in a low interest rate environment: diminishing effectiveness?. Journal of Macroeconomics, 54, 217-231.

Bowman, D., Cai, F., Davies, S., & Kamin, S. (2015). Quantitative easing and bank lending: Evidence from Japan. Journal of International Money and Finance, 57, 15-30.

De Bondt, G. J. (1999). Credit channels in Europe: a cross-country investigation. PSL Quarterly

Review, 52(210).

Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of the American statistical association, 74(366a), 427-431. Gambacorta, L., & Marques-Ibanez, D. (2011). The bank lending channel: lessons from the crisis. Economic Policy, 26(66), 135-182.

Hallsten, K. (1998). Bank loans and the transmission mechanism of monetary policy (No. 73). Sveriges Riksbank Working Paper Series, Sveriges Riksbank, Stockholm

Hülsewig, O., Winker, P., & Worms, A. (2004). Bank Lending and Monetary Policy Transmission: A VECM Analysis for Germany/Bankkredite und geldpolitische Transmission: Eine VECM Analyse für Deutschland. Jahrbücher für Nationalökonomie und Statistik, 224(5), 511-529.

Johansen S. 1995. Likelihood-Based Inference in Cointegrated Vector Autoregressive Models. Oxford University Press: Oxford.

(33)

30

Joyce, M., & Spaltro, M. (2014). Quantitative easing and bank lending: a panel data approach (No. 504). Bank of England Working Paper

Kashyap, A. K., & Stein, J. C. (2000). What do a million observations on banks say about the transmission of monetary policy?. American Economic Review, 90(3), 407-428.

Kashyap, A. K., & Stein, J. C. (1994). The impact of monetary policy on bank balance sheets (No. w4821). National Bureau of Economic Research.

Lojschova, A. (2017). Did quantitative easing boost bank lending? The Slovak experience. (No. 1/2017) Working Papers, Bratislava: National Bank of Slovakia.

Loupias, C., Savignac, F., & Sevestre, P. (2001). Monetary policy and bank lending in France: are there asymmetries?. ECB Working Paper, No. 101, European Central Bank (ECB),

Frankfurt a. M.

Molyneux, P., Reghezza, A., Thornton, J., & Xie, R. (2020). Did negative interest rates improve bank lending?. Journal of Financial Services Research, 57(1), 51-68.

Ng, S., & Perron, P. (1995). Unit root tests in ARMA models with data-dependent methods for the selection of the truncation lag. Journal of the American Statistical Association, 90(429), 268-281.

Ramey, V. (1993, December). How important is the credit channel in the transmission of monetary policy?. In Carnegie-Rochester Conference Series on Public Policy (Vol. 39, pp. 1-45). North-Holland.

Riksbank (2015) 2015 – Negative repo rate is introduced.

https://www.riksbank.se/en-gb/about-the-riksbank/history/historical-timeline/2000-2018/negative-repo-rate-is-introduced/ Rodnyansky, A., & Darmouni, O. M. (2017). The effects of quantitative easing on bank

lending behavior. The Review of Financial Studies, 30(11), 3858-3887.

Schwert, G. W. (2002). Tests for unit roots: A Monte Carlo investigation. Journal of Business & Economic Statistics, 20(1), 5-17.

Westerlund, J. (2003). A panel data test of the bank lending channel in Sweden. Working Papers 2003:16. Department of Economics, Lund University

References

Related documents

För att uppskatta den totala effekten av reformerna måste dock hänsyn tas till såväl samt- liga priseffekter som sammansättningseffekter, till följd av ökad försäljningsandel

Från den teoretiska modellen vet vi att när det finns två budgivare på marknaden, och marknadsandelen för månadens vara ökar, så leder detta till lägre

The increasing availability of data and attention to services has increased the understanding of the contribution of services to innovation and productivity in

Generella styrmedel kan ha varit mindre verksamma än man har trott De generella styrmedlen, till skillnad från de specifika styrmedlen, har kommit att användas i större

Parallellmarknader innebär dock inte en drivkraft för en grön omställning Ökad andel direktförsäljning räddar många lokala producenter och kan tyckas utgöra en drivkraft

Närmare 90 procent av de statliga medlen (intäkter och utgifter) för näringslivets klimatomställning går till generella styrmedel, det vill säga styrmedel som påverkar

På många små orter i gles- och landsbygder, där varken några nya apotek eller försälj- ningsställen för receptfria läkemedel har tillkommit, är nätet av

Det har inte varit möjligt att skapa en tydlig överblick över hur FoI-verksamheten på Energimyndigheten bidrar till målet, det vill säga hur målen påverkar resursprioriteringar