Capital requirements, risk shifting and the mortgage market [Provisional – Please do not cite without the permission of the author] Tomasz Wieladek(1),* Abstract I study the effect of changes to bank-specific capital requirements on mortgage loan supply with a new loan-level dataset containing all (over 4 million) mortgages issued in the UK between 2005Q2 and 2007Q2. If a rise in capital requirements is binding, and the Miller-Modgliani (1958) theorem fails, affected banks will need to raise their capital to risk-weighted asset ratios by either cutting assets or raising capital. The former would require a contraction of mortgage loan supply. The latter may lead to riskier lending to raise capital through higher retained earnings. But competing non-bank finance companies or unaffected banks could provide a source of credit substitution, leaving aggregate mortgage loan supply unchanged. My results suggest that a rise of a 100 basis points in capital requirements leads to a 4.8% loan contraction. Borrowers with verified incomes and low LTV ratios experience the largest loan reductions, consistent with risk shifting. Competing banks extend credit to this group of borrowers by roughly the same amount. But no evidence for credit substitution from non-bank finance companies is found. Keywords: Capital requirements, loan-level data, mortgage market. JEL classification: G21, G28.

_______________________________________________________________________________ (1) Bank of England. Email: tomasz.wieladek@bankofengland.co.uk * I would like to thank Julia Giese, Vicky Saporta and Martin Weale for helpful comments on an earlier draft. I am indebted to Belinda Tracey for substantial help at the beginning of this project. All remaining errors are my own. The views represented in this paper are only those of the author and not the Bank of England.

Introduction

Current financial reform proposals focus on the introduction of macroprudential

policy to maintain financial stability and increase the resiliency of financial institutions to

adverse shocks. The Basel III agreement allows regulators to set counter-cyclical capital

buffers for this purpose. Yet the impact and transmission mechanism of these tools is still

not well understood (Galati and Moessner, 2011). This should not be surprising, since

regulators in most countries imposed a constant capital requirement, by bank and over

time, in accordance with the regulatory framework of Basel I, in the past. But as

extensively documented in Francis and Osborne (2009) and Aiyar, Calomiris and

Wieladek (2014a), UK regulators adjusted bank-specific capital requirements over time to

address legal, operational and interest rate risks, which were not allowed for in Basel I. In

this paper I exploit this unique regulatory regime to test if changes in lender level capital

requirements affect mortgage loan supply with loan-level data on all new mortgages

issued in the United Kingdom between 2005Q2 and 2007Q2. I also examine if the affected

institutions shift lending towards riskier borrowers to boost short-term profitability and

grow capital through higher retained earnings. Finally, I study to which extent locally

competing banks, which were not affected by a rise in capital requirements, and non-

bank finance companies, also referred to as the ‘shadow banking’ system, provide a source

of credit substitution in response to the loan contraction by the affected bank.

For capital requirements to be an effective macroprudential policy instrument and

affect the aggregate loan supply, three conditions need to be satisfied: i) Bank equity

needs to be more expensive than bank debt; ii) Capital requirements need to be a binding

constraint on a bank’s choice of capital and iii) there needs to be only limited substitution

from other sources of finance. The first condition implies a failure of the Miller-

Modigliani (1958) theorem for banks, as otherwise changes in the capital requirement do

not need to affect a financial institution’s balance sheet. But economic theory provides

good reasons for why condition i) should be satisfied, such as asymmetric information

2

(Myers and Majluf, 1984) and the difference in tax treatment between debt and equity.1

Similarly, empirical work documenting the impact of adverse shocks to capital on loan

growth, as in Bernanke (1983) and Peek and Rosengren (1997, 2000) provides support for

this assumption. Several empirical studies, namely Ediz et al (1998), Alfon et al (2005),

Francis and Osborne (2009), Aiyar, Calomiris and Wieladek (2014a) and Bridges et al

(2014), also demonstrate that capital requirements were a binding constraint on UK

banks’ capital choices during the 1998-2007 period. This suggests that the second

condition is likely to be satisfied as well.

Figure 1 illustrates what happens following an increase in the bank-specific capital

requirement if conditions i) and ii) hold. The affected bank will have to either raise

capital from outside investors, grow capital through retained earnings or cut back on risk-

weighted assets. Several studies use the same underlying FSA capital requirements data to

test this last implication. All of them consider the impact on lender-level loan growth to

the private non-financial corporate (PNFC) sector.2 Aiyar, Calomiris and Wieladek

(2014a), Bridges et al (2014) and Francis and Osborne (2012) find a loan contraction of

5.7, 5.63 and 5 per cent, following a one hundred basis points increase in capital

requirements, respectively. There is also related work on Dynamic Provisioning, an

alternative macroprudential instrument, in Spain. In particular, Pedryo, Jimenez, Ongena

and Saurina (2012) provide evidence for the impact of Dynamic Provisioning on lending

growth with a detailed loan-level dataset on loans granted and applied for by firms

operating in Spain. In contrast to this previous body of work, I examine the effect of

changes in the level of capital requirements on individual household mortgage loans, the

granularity of which allows me control for loan demand and address issues of reverse

causality substantially better than in previous studies. Furthermore, as shown in figure 1,

banks may choose to raise capital to satisfy their capital requirement through the retained

1 See Aiyar, Calomiris and Wieladek (2014a) for an extensive discussion of economic theory and empirical evidence in support of the validity of the first assumption. 2 Bridges et al (2014) is the only other work to consider the impact on mortgage loan growth, but they examine the impact on bank-level, as oppose to loan-level, mortgage lending. 3 To make their results comparable to the other studies, the authors kindly provided me with a figure that refers to the effect over one year and is a weighted average of their results for the commercial real estate and other PNFC lending categories.

3

earnings channel. Under Basel I, the regulatory framework that applied during this

period, household mortgages were assigned a risk-weight of 50% regardless of their

individual characteristics. Faced with a rise in their capital requirement, banks were

therefore incentivised to reallocate lending towards more profitable high risk borrowers

to grow capital via retained earnings (Kahane, 1977; Kim and Sontero, 1988). In that case,

raising capital requirements could have the perverse effect of making bank balance sheets

actually riskier and less resilient. Both Furlong (1988) and Sheldon (1996) test for this

type of risk shifting with bank balance sheet data as a result of the introduction of the

1981 leverage ratio and Basel I in the US and Switzerland, respectively. While they do not

find any support for risk shifting, Haldane (2013) argues that this could be a result of the

interplay with other regulations that were introduced coincidentally. In contrast, my

study examines risk shifting in response to individual bank-level capital requirements

with loan-level data within the same regulatory framework, meaning that this is a much

more powerful test of the risk-shifting hypothesis.

But, as clearly shown in figure 1, the effect on individual banks need not translate

to aggregate loan supply, as credit substitution from other sources may offset this effect.

In the presence of credit substitution, macroprudential policy, if aimed at affecting the

loan supply, will be less effective. Only two previous papers test for such credit

substitution. Aiyar, Calomiris and Wieladek (2014a) test for the presence of PNFC

lending substitution between competing foreign branches and UK regulated banks,

following changes in capital requirements. They match competing banks based on their

exposure to 14 different PNFC sectors. Alternatively one could test for substitution

between foreign branches and regulated banks that belong to the same banking group.

The results from this exercise, together with testing for substitution from equity and

corporate bond markets, are presented in Aiyar, Calomiris and Wieladek (2014b). Both

papers find strong evidence for loan substitution between foreign branches and regulated

banks. But evidence for loan substitution from either corporate bond or equity markets is

absent. I test for loan substitution as well and adopt a similar matching approach. In

particular, I test if banks which compete for business in the same local mortgage market

4

expand their loan supply in response to changes in capital requirements on the affected

banks. I adopt the same strategy to test for loan substitution between banks and non-bank

finance companies (often referred to as the shadow banking system) during this sample.

In summary, I contribute to this literature in several important dimensions: To my

knowledge, I am the first to examine the impact of changes in capital requirements on the

mortgage loan supply with loan-level data. As inference in my study is based on many

thousands of loan-level observations per financial institution and capital requirement

change, econometric bias due to reverse causality is much less likely than in the other

studies. More importantly, following the approach in Kwhaja and Mian (2008) and Aiyar,

Calomiris, Hooley, Korniyenko and Wieladek (2014), geographical time dummies allow

me to control for loan demand better than in most previous work. This is also the first

paper to document risk-shifting following changes in capital requirements. Finally, I am

the first to formally test for mortgage loan substitution from banks that were unaffected

by changes in capital requirements and non-bank finance companies, frequently referred

to as the ‘shadow banking system’.

I find that a rise in an affected banks capital requirement of about 100 basis points,

conditional on all observable borrower, loan and lender balance sheet characteristics,

leads to a decline in loan size of about 4.8%. My results suggest that this effect is

asymmetric: Mortgage loan supply only reacts to a rise, but not to a decline, in capital

requirements. I also find evidence for risk shifting: This loan contraction does not affect

borrowers with an impaired credit history or high Loan-to-Value (LTV) ratios. Similarly,

borrowers with verified income and low LTV ratios are affected to a greater extent.

Interestingly, there is strong evidence that locally competing banks, which were not

affected by changes in capital requirements, expand lending to exactly this group of

borrowers. Quantitatively, this loan expansion seems to completely offset the loan

contraction associated with higher capital requirements on the affected banks. But I do

not find evidence for credit substitution from non-bank finance companies.

5

The loan supply impact I find in this study is greater than the .95% documented

in Bridges et al (2014), but due to the large confidence bands in both studies, this effect is

not likely to be statistically significantly different. Furthermore, I only have data for

2005Q2 – 2007Q2, a period when banks were most likely more capital constrained than

the period they examine (1990-2011), which means that the same change in capital

requirements is likely to have a more powerful effect on mortgage lending. Finally,

differences in controlling for loan demand may be partially responsible for this result.

Similarly, I find that locally competing institutions can offset the loan contraction

associated with a rise in capital requirements entirely. The closest study to mine, Aiyar,

Calomiris and Wieladek (2014b) finds that PNFC loan substitution from foreign branches

can only offset the loan contraction associated with higher capital requirements by about

43%. This quantitative difference may be due to the fact that is presumably easier for low

risk borrowers, namely those with low LTV ratios, to switch their loan provider.

Mortgage products are also standardised, while PNFC loans might involve more bespoke

arrangements. As a result the elasticity of substitution for the former should be higher

than for the latter.

But it is important to emphasize that these effects are based on observed sample

averages during this particular sample period. In theory, higher capital requirements

could increase lending at banks with very low or negative net worth; if capital ratio

requirements help to prevent or overcome a so‐called “debt overhang” problem, which

can occur at very low capital ratios, then in principle, higher capital could encourage

lending. Furthermore, my results measure short‐term loan‐supply reactions. It is not

surprising that a decline in the loan supply is associated with a transition to higher capital

requirements, but in the longer run, improvements in the stability of the banking system

that result from higher capital requirements could improve banks’ ability to raise funds in

the market and thereby mitigate the short-run declines in loan supply documented here.

6

My results suggest that changes in capital requirements can have important

effects on the size of individual mortgage loans, the affected banks’ risk shifting behaviour

and credit substitution by their competitors. While the findings from this study are based

on microprudential capital requirement changes and thus subject to the Lucas Critique,

since bank behaviour might change in the presence of macroprudential regulation, they

may still offer important lessons for economic policy and theory. In particular, the results

suggest that rises in capital requirements intended to make a bank more resilient, through

risk shifting, may actually have the opposite effect. Furthermore, competition in the local

lending market may mute the loan contraction, associated with a rise in capital

requirements on an individual institution, entirely. Awareness of these issues will likely

help policy makers with the implementation of Basel III, which should address these

problems, and the conduct of macroprudential policy to ensure financial resilience, whilst

minimising any possible distortions to the mortgage market.

The remainder of the paper is structured as follows: Section 2 describes the UKs

regulatory regime and the data. Section 3 describes the empirical approach for each of the

proposed hypotheses, presents the results and examines them for robustness. Section 4

concludes.

2. UK capital requirement regulation and data

In this section I describe the UK’s regulatory regime and the detailed loan-level

database that make my investigation possible.

Bank-specific capital requirements in most countries were set at a fixed value at or

above the minimum of 8 per cent of risk-weighted assets since the introduction of Basel I

in 1988. But in the UK, regulators varied bank-specific capital requirements, otherwise

known as minimum trigger ratios4, to address operational, legal or interest rate risks,

which were not accounted for in Basel I (Francis and Osborne, 2009). Individual financial

institutions were subject to different capital requirements over time and these were

4 A trigger ratio is the technical term for capital requirement, since regulatory intervention would be triggered if the bank capital to risk-weighted asset ratio fell below this minimum threshold.

7

subject to review either on an on-going basis or every 18-36 month. This regulatory

regime was first implemented by the Bank of England, with the Financial Services

Authority (FSA) taking over in 1997. The FSA based regulatory decisions for banks on a

system of guidelines called ARROW (Advanced Risk Responsive Operating frameWork),

which covers a wide array of criteria related to operational, management, business as well

as many other risks. The ARROW approach also encompassed prudential risks, but this

was not one of the core supervision areas. Indeed, in this high-level review into UK

financial regulation prior to the financial crisis of 2008, lord Turner, the chief executive

of the FSA, concluded that: ‘Risk Mitigation Programs set out after ARROW reviews

therefore tended to focus more on organisation structures, systems and reporting

procedures, than on overall risks in business models’ (Turner, 2009). Similarly, the

inquiry into the failure of the British bank Northern Rock concluded that ‘under

ARROW I there was no requirement on supervisory teams to include any developed

financial analysis in the material provided to ARROW Panels’ (FSA, 2008). Indeed Aiyar,

Calomiris and Wieladek (2014a) show that, while bank size and writeoffs appear to be

important determinants of the level of capital requirements in the cross-section, bank

balance sheet variables can typically not predict time variation in capital requirements.

More importantly, they show that, while capital requirement changes were not frequent

events, there was a substantial degree of heterogeneity in bank capital requirements both

in the cross-section and over time from 1998Q3 – 2007Q2. This regulatory framework

therefore provides a uniquely suitable environment to examine the effect of changes to

capital requirements on bank behaviour.

The Bank of England has kindly made these regulatory returns data, collected

from the BSD3 form, available for my investigation. I collect data on a total of 58

regulated banks’ lending to UK households. But banks are not the only source of

household mortgage finance in the UK. Building societies, which made up 16% of all

mortgage lending to households, are owned by their customers as oppose to external

shareholders. Like banks, building societies were subject to different minimum capital

requirements across institutions and time, but the review was only conducted once a year

8

on a rolling basis. Regulatory data for building societies, of which there are 59 in the UK,

from the QFS1 form was provided by the FSA. The third group of mortgage providers are

non-bank finance companies, which were not subject to changing capital requirements.

Instead, they were subject to a 1% capital requirement for institutions with a capital of

greater than 100,000 pound sterling. These mortgage providers are also typically referred

to as the ‘shadow banking system’ and made up about 10% of UK mortgage lending

during the time period examined here.

My study covers the time period 2005Q2 to 2007Q2, as loan-level data on

mortgages was not collected before then and subsequent data may have been affected by

the UK mortgage market crisis associated with failure of the bank Northern Rock in

2007Q3. Furthermore, prior to 2008Q1, UK regulators relied on the risk-weights

associated with Basel I. Unlike the Basel II risk weights, which were adopted after this

date, these risk weights were fixed, assigning a weight of 50% to mortgage loans and

100% to PNFC loans, regardless of riskiness of the underlying loan. The change from

fixed (Basel I) to internal models based (Basel II) risk weights may introduce an additional

margin of adjustment following changes in capital requirements (Benford and Nier, 2007).

This additional regulatory change would make the task of understanding the transmission

of capital requirements to individual loans even more difficult. For both of these reasons

and to better isolate how changes in capital requirements affect bank behaviour, I choose

the sample period to finish in 2007Q2. Regulated institutions were affected by 20 capital

requirement changes during this time, all of which are shown in figure 2, with summary

statistics provided in table 4. All of the other lender level balance sheet data was provided

by the Bank of England’s Statistics and Regulatory Data Division.5 The control variables I

derive from these data are described in greater detail in table 1, with the corresponding

summary statistics provided in table 3.

The source of loan-level data is the FSA’s Product Sales Database (PSD), which

starts in April 2005. This includes data on new mortgages issued, retail investments, and

5 All banks operating in the United Kingdom are legally required to provide this information to the Bank of England.

9

home purchase plans among other things. Any financial institution, whether bank,

building society or non-bank finance company, is required to report these data at a

quarterly horizon to the FSA, within 20 working days of the end of each quarter. In

addition to information about the size of the loan and the value of the property

purchased, this database also collects important borrower and mortgage contract

information. In particular, borrower age and income provide information on borrower

age in years and income in pound sterling. Flexibility of repayment and maturity provide

information on the presence of early repayment fees, taking the value of one if that is the

case and zero otherwise, and the maturity of the mortgage in years. Purpose of

remortgage provides information on whether extra money was raised and if so for what

purpose. Income verification and impaired credit history indicate if the applicant’s

income was verified and if they have a bad credit history, taking a value of one if that is

the case and zero otherwise. Type of employment indicates if the borrower is employed,

self-employed, retired or other. Borrower type provides information on whether the

borrower is a first-time buyer, a person moving home, a social tenant6 or remortgaging

their existing home, for example. Greater detail on these variables can be found in the

bottom half of table 1. Table 2 shows the mean and the standard deviation of the natural

logarithm of loans and the LTV ratio by bank type and selected borrower characteristics

for the time period 2005Q2 – 2007Q2. Consistent with economic intuition, individuals

whose income status is not verified and who are self-employed, and could therefore be

perceived as riskier borrowers, typically obtain mortgages with lower LTV ratios. Non-

bank finance companies tend to provide mortgages with much higher LTV ratios than

either building societies or banks.

3. Empirical approach and results

In this section I describe the empirical framework to test each of the proposed

hypotheses and report the results.

6 In the UK individuals living in social housing, social tenants, have the right to buy the property they live in for a discount from the government, under the Right-to-Buy scheme originally introduced by Prime Minister Margaret Thatcher in the 1980s.

10

3.1 Do capital requirements affect mortgage loan supply?

In this section I examine whether changes in capital requirements affect loan

supply with the following regression equation:

𝑳𝑳𝑳𝑳𝑳𝑳𝑳𝑳𝒊𝒊,𝒋𝒋,𝒕𝒕 = 𝜶𝜶𝒋𝒋 + ∑ 𝜹𝜹𝒌𝒌𝑲𝑲𝑲𝑲𝒋𝒋,𝒕𝒕−𝒌𝒌𝟑𝟑𝒌𝒌=𝟎𝟎 + 𝝋𝝋𝝋𝝋𝑳𝑳𝝋𝝋𝒊𝒊,𝒋𝒋,𝒕𝒕 + ∑ 𝜽𝜽𝒌𝒌𝑷𝑷𝑷𝑷𝑳𝑳𝑷𝑷𝒋𝒋,𝒕𝒕−𝒌𝒌𝟑𝟑

𝒌𝒌=𝟎𝟎 + 𝜸𝜸𝑳𝑳𝝋𝝋𝒋𝒋,𝒕𝒕 + 𝑷𝑷𝝋𝝋𝒉𝒉𝒕𝒕 + 𝒆𝒆𝒊𝒊,𝒋𝒋,𝒕𝒕 (1)

where 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖,𝑗𝑗,𝑡𝑡 is the natural logarithm of the loan size for a new loan agreed between

by borrower i and lender j at time t. 𝐾𝐾𝐾𝐾𝑗𝑗,𝑡𝑡 is the minimum capital requirement ratio7 for

regulated financial institution j at time t. 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑗𝑗,𝑡𝑡 is the provisions to asset ratio for

regulated financial institution j at time t. 𝐵𝐵𝐿𝐿𝐵𝐵𝑖𝑖,𝑗𝑗,𝑡𝑡 is a vector of all the borrower and loan

characteristics listed in table 1, for individual i, borrowing from lender j at time t. Where

borrower and loan characteristics are not binary, they are modelled as additional sets of

fixed effects. 𝐿𝐿𝐵𝐵𝑗𝑗,𝑡𝑡 is a vector of lender characteristics for lender j at time t, all of which

are listed in table 1. 𝛼𝛼𝑗𝑗 is a lender fixed effect to account for lender unobservable time-

invariant characteristics. 𝑃𝑃𝐵𝐵ℎ𝑡𝑡 is a vector of postcode time effects to account for

unobservable geographical differences, in particular in loan demand, among the

approximately 10,000 different three character postcode districts of the UK over time. To

avoid perfect multi-colinearity, I drop the first variable from each group of fixed effects.

𝑒𝑒𝑖𝑖,𝑗𝑗,𝑡𝑡 is assumed to be a normally distributed error term. Since the main variable of

interest varies by institution and over time, all standard errors are clustered by bank-time.

This particular choice of specification is explained in greater detail below.

In this empirical model, capital requirements are assumed to affect the dependent

variable contemporaneously and with up to three lags. In the UK, mortgage loan

conditions change at monthly, sometimes weekly, frequency, depending on factors such

as the individual institutions balance sheet or general financial market conditions. The

FSA’s PSD records the loan size and the associated borrower and mortgage characteristics

when the purchase of a property is completed. This is typically several weeks, if not

7 Aiyar, Calomiris and Wieladek (2014a) point out that in practice, banks were subject to two different capital requirements: The trading and the banking book capital requirement. Since household lending is in a financial institution’s banking book, this is the capital requirement that I focus on.

11

months, after the mortgage terms were agreed. For these reasons data recorded in the

PSD lags actual mortgage market conditions by several weeks or months. Capital

requirements are available only at quarterly frequency. And although financial

institutions might start to change loan conditions several months before the

implementation of regulatory action, due to anticipation, it is difficult to know precisely

when the purchase of a property is completed. Furthermore, due to adjustment costs, it is

quite likely that the reaction to changes in capital requirements occurs with a lag. This is

why Aiyar, Calomiris and Wieladek (2014a, 2014b) and Aiyar, Calomiris, Hooley,

Korniyenko and Wieladek (2014) include the contemporaneous value as well as 3 lags of

the capital requirement ratio in their exploration of these data. For these reasons, and to

maintain comparability to previous work, the minimum capital requirement ratio enters

contemporaneously and with three lags.8

Similarly, the provisions to asset ratio enters model (1) contemporaneously and

with three lags. Aiyar, Calomiris and Wieladek (2014a) and Aiyar, Calomiris, Hooley,

Korniyenko and Wieladek (2014) both find that the write-off to risk-weighted asset ratio

is an important control variable to control for potential regulatory changes due to

deterioration in the quality of the loan portfolio. Unfortunately, writeoffs not are

available from the Quarterly Financial Statement form that building societies submit to

the FSA, which is why I use provisions instead, since these are available for both, banks

and building societies. The remaining balance sheet variables only enter

contemporaneously.

In this study I aim to identify to loan supply effect of changes in capital

requirements, and hence it is important to control for loan demand. This is a challenging

task in most empirical studies. But the detail of the dataset in this paper allows me to

follow the approach presented in Kwhaja and Mian (2008) and Aiyar, Calomiris, Hooley,

Korniyenko and Wieladek (2014). In both of their applications, they exploit geographical

information on the loan destination and use geographical time dummies to control for

8 I note that my results are robust to the inclusion of the contemporaneous value alone or the lagged value alone and that the inclusion of shorter lag lengths gives quantitatively similar results.

12

loan demand. To the extent that loan demand varies over time and across geographical

boundaries, this allows them to interpret their estimates as supply effects. The PSD

dataset provides detailed postcode level information on the property that was purchased

with the loan. In the UK, each property is assigned a 6-character postcode. There is a total

of 1.7 million postcodes corresponding to 29 million delivery addresses, approximately 17

addresses per postcode. The first three characters indicate the postal district, for example

‘CB1’ corresponds to the district – ‘Cambridge 1’. My dataset contains roughly 4,250,000

loan-level observations and 10,000 distinct three character postcodes. I therefore use

three character postcode time dummies to control for loan demand. To the extent that

mortgage loan demand shocks differ across the 10,000 three character level postcodes of

the UK and the associated time dummies pick those up, the regression estimates on the

capital requirement can thus be interpreted as a supply effect. I report the sum of

coefficients on these variables and corresponding F tests, testing the hypothesis that the

sum of coefficients is statistically different from zero at conventional levels of

significance, in all of the tables. In the presence of lender and time fixed effects, the

econometric interpretation of ∑ 𝛿𝛿𝑘𝑘3𝑘𝑘=0 is clear: this sum of coefficients reflects the

difference in the loan size of a bank/building society that experienced a change in its

capital requirement with respect to one that did not.

Economic theory has a clear prediction for the sign of the main coefficient of

interest, ∑ 𝛿𝛿𝑘𝑘3𝑘𝑘=0 . Given that equity is expensive and capital requirements a binding

constraint on an individual lender’s choice of capital, one would expect a decline in loan

size following a regulatory action leading to a rise in capital requirements, which is

consistent with a negative loan supply effect. At loan level, this can happen in two

different ways. The affected lender can either reject the application of a borrower that

requires a large loan and lend to one with a smaller loan instead or just offer a smaller

loan. Unfortunately, the PSD does not have information on loan applications, which

means that I cannot formally distinguish between these two adjustment channels. But my

examination of loan substitution in the next section suggests that the former channel

seems to be the relatively more important one.

13

I am not the first to study the impact of minimum capital requirements on loan

supply. Francis and Osborne (2009), Aiyar, Calomiris and Wieladek (2014a) and Bridges

et al (2014) examine the impact of lender-level changes in capital requirements on

lender-level loan growth. In contrast, in addition to the impact on loan size, I can also

examine if this effect varies with borrower and loan characteristics. The detail of my

dataset allows me to control for econometric bias arising from the omission of borrower

and loan characteristics. The postcode time effects also allow me to credibly control for

loan demand. Finally, unlike in previous studies, the dataset contains many thousand loan

observations per capital requirement change. Econometric bias from reserve causality is

therefore unlikely. In other words, I argue that equation (1) is substantially better

identified than the approaches used in previous work.

Table 5 presents estimates of equation (1) for loan size. Column one shows that,

once lender and postcode time fixed effects are included, the estimate of ∑ 𝜹𝜹𝒌𝒌𝟑𝟑𝒌𝒌=𝟎𝟎 is -.056,

implying that if capital requirements rise by 100 basis points, the average loan size issued

by the affected bank decreases by about 5.6%. Once provisions and size are included, the

effect now declines to about 5%. The inclusion of additional bank balance sheet variables,

such as core funding, the mortgage to total asset ratio, the liquid to total asset ratio and

the return on assets in column four leads to an estimate of 4.8%. Standard errors are

clustered by bank and time, though the results remain the same if I only cluster by bank

or only cluster by time. Colum five presents results from a regression where the capital

requirement ratio variables in specification (1) have been interacted with the fraction of

mortgage in total lending. The sum of coefficients on this interaction is not statistically

significant. Column six separately reports the impact of coefficient for institutions that

experienced a rise in capital requirements, and those that experienced a decline. I only

detect a statistically significant effect for the former, but not the latter. This suggests that

capital requirements affect the loan supply in an asymmetric way: Rises in capital

requirement lead to a loan contraction, but declines do not necessarily lead to a loan

expansion. In sum, the effect of a 100 basis point rise in the capital requirement ratio

14

leads to a loan contraction of about 4.8%, which seems to be mostly the result of capital

requirement increases, rather than decreases.

Of course, lenders could choose to grow capital organically by raising profits and

retaining a greater share of earnings, rather than contracting lending. One way to achieve

this goal in the short run is to lend more to riskier borrowers, who can be charged extra

fees and/or higher interest rates. This was a particularly attractive margin of adjustment

under Basel I, which assigned a risk weight of 50% to all mortgage loans regardless of

borrower and loan characteristics. In other words, in the presence of risk shifting, high

(low) risk borrowers should be affected relatively less (more). Previous empirical work by

Furlong (1989) and Sheldon (1996) does find not any evidence for this type of risk-

shifting following the introduction of the 1981 US leverage ratio and Basel I in

Switzerland, respectively. Yet as argued by Haldane (2013), this could be due to the

coincident introduction of other types of financial regulation. I examine risk shifting in

table 6. In particular, I interact the capital requirement ratio coefficients with several

borrower and loan characteristics. Column one shows estimates for a regression model

where the capital requirement ratio is interacted with several dummy variables that take

the value of one if the individual is a first time buyer, self-employed or has an impaired

credit history or not. Interestingly, the interaction on first time buyers is statistically

significant, positive and of almost equal size to the coefficient on KR. This suggests that

first time buyers, who typically tend to be riskier borrowers, are not affected by the loan

contraction. A similar pattern applies to individuals who are self-employed and have an

impaired credit history. This is consistent with risk shifting following a increase in capital

requirements. Similarly, columns two and five suggest that individuals whose income was

verified at the time of application and low LTV loans are affected the most. Column three

and four suggests that individuals taking out interest only mortgages (high maturity

loans) are affected less. Finally, all of the interaction terms are included in column six.

This suggests that individuals who take out low LTV mortgages and those with verified

income experience the largest credit contraction following a 100 basis point rise in capital

requirements. In contrast, first time buyers, individuals who are self-employed or have an

15

impaired credit history do not seem to be adversely affected by the mortgage loan

contraction associated with an increase in capital requirements. Overall, this evidence

seems to be consistent with the idea that lenders reallocated lending from low risk to

high risk borrowers following an increase in capital requirements.

The findings in this section suggests that affected lenders cut back mortgage

lending by about 4.8% following a 100 basis point rise in capital requirements and that

this effect is asymmetric: Lenders only seem to cut back mortgage loan supply in response

to an increase in capital requirements, but the evidence for an expansion of mortgage loan

supply after a decrease in capital requirements is weaker. Furthermore, I also document

that the effect varies by borrower and loan type. In particular, individuals who are first-

time buyers, have an impaired credit history or are self-employed are less affected by the

loan contraction associated with a rise in capital requirements. On the other hand, low

LTV ratio borrowers are affected the most. Given their large deposits or long repayment

histories, this group is clearly least likely to default, which is why this result might seem

strange at first sight. However, it is important to keep in mind that lenders can always

choose to raise capital through greater retained earnings, which could be achieved

through a reallocation of lending towards higher risk, higher profit, borrowers. The

results presented in this paper are indeed consistent with this type of risk-shifting.

3.2. Do unaffected lenders partially offset the loan contraction associated with capital

requirement policy?

One of the conditions for capital requirements policy to affect the aggregate credit

supply and systemic credit risk is the limited substitution from alternative sources of

finance. Aiyar, Calomiris and Wieladek (2014a) provide evidence that foreign branches

expand lending in response to an increase in capital requirements on regulated banks

through the competition channel. Aiyar, Calomiris and Wieladek (2014b) show that

foreign branches react to capital requirement changes in the regulated subsidiary

belonging to the same banking group. They also explore substitution from corporate bond

and equity markets, but do not find evidence for this channel.

16

But no previous work has looked at potential loan substitution from either

regulated lenders which were not subject to changes in capital requirement (referred to as

‘unaffected’ lenders subsequently) or the shadow banking system in the mortgage market.

I.e. Do individuals switch to unaffected or non-bank lenders following changes in capital

requirements on regulated institutions? While this is clearly a relevant issue for both

economic theory and policy, there is no previous empirical evidence on this question. For

ease of exposition, I focus on the loan supply response of unaffected lenders first and

discuss non-bank finance companies thereafter.

Here I examine whether, and to which extent, the size of mortgage loans that

unaffected lenders and non-bank finance companies issue, reacts to changes in capital

requirements on competing financial institutions. The challenge in this task is, of course,

to relate capital requirements of affected lenders to lending decisions by unaffected

lenders. I create a lender-specific reference groups for this purpose. The idea behind this

approach is that unaffected institutions should respond the most in geographical areas in

which they have an established presence and compete actively with lenders affected by

capital requirement changes. The results in table 5 suggest that while increases in capital

requirements are associated with a loan contraction, decreases in capital requirements do

not necessarily lead to a loan expansion. Similarly, faced with a smaller loan than desired,

borrowers might be forced to switch lenders, but larger loans may not necessarily attract

new borrowers. In other words, rises in capital requirements are more likely to lead to

borrowing from a different lender, which why I only focus on this case in constructing

the reference group below. The reference group is constructed in the following way:

First, I calculate the level of the capital requirement at the local authority level implied

by all lenders that experienced an increase in capital requirements:

𝐾𝐾𝐾𝐾𝑞𝑞 = ∑ � 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑙𝑙𝑎𝑎𝑙𝑙𝑙𝑙𝑖𝑖𝑙𝑙𝑙𝑙 𝑏𝑏𝑏𝑏 𝑙𝑙𝑎𝑎𝑙𝑙𝑙𝑙𝑎𝑎𝑎𝑎 𝑚𝑚 𝑡𝑡𝑡𝑡 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑞𝑞𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑙𝑙𝑎𝑎 𝑙𝑙𝑎𝑎𝑙𝑙𝑙𝑙𝑖𝑖𝑙𝑙𝑙𝑙 𝑏𝑏𝑏𝑏 𝑎𝑎𝑙𝑙𝑙𝑙 𝑙𝑙𝑎𝑎𝑙𝑙𝑙𝑙𝑎𝑎𝑎𝑎𝑙𝑙 𝑡𝑡𝑡𝑡 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑞𝑞

�𝐾𝐾𝐾𝐾𝑚𝑚𝑚𝑚 . This is the effective capital

requirement in area q. I then construct the reference capital requirement, that is the

exposure of unaffected lender j is to the effective capital requirement:

𝐾𝐾𝑒𝑒𝑅𝑅𝐾𝐾𝐾𝐾𝑗𝑗 = ∑ � 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑙𝑙𝑎𝑎 𝐿𝐿𝑎𝑎𝑙𝑙𝑙𝑙𝑖𝑖𝑙𝑙𝑙𝑙 𝑏𝑏𝑏𝑏 𝑢𝑢𝑙𝑙𝑎𝑎𝑢𝑢𝑢𝑢𝑎𝑎𝑢𝑢𝑡𝑡𝑎𝑎𝑙𝑙 𝑙𝑙𝑎𝑎𝑙𝑙𝑙𝑙𝑎𝑎𝑎𝑎 𝑗𝑗 𝑡𝑡𝑡𝑡 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑞𝑞𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑙𝑙𝑎𝑎 𝐿𝐿𝑎𝑎𝑙𝑙𝑙𝑙𝑖𝑖𝑙𝑙𝑙𝑙 𝑏𝑏𝑏𝑏 𝑢𝑢𝑙𝑙𝑎𝑎𝑢𝑢𝑢𝑢𝑎𝑎𝑢𝑢𝑡𝑡𝑎𝑎𝑙𝑙 𝑙𝑙𝑎𝑎𝑙𝑙𝑙𝑙𝑎𝑎𝑎𝑎 𝑗𝑗 𝑡𝑡𝑡𝑡 𝑎𝑎𝑙𝑙𝑙𝑙 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑙𝑙

�𝐾𝐾𝐾𝐾𝑞𝑞𝑗𝑗 .

17

𝐾𝐾𝑒𝑒𝑅𝑅𝐾𝐾𝐾𝐾𝑗𝑗 is therefore the extent to which lender j is exposed to local mortgage markets

that were affected by an increase in capital requirements. In the presence of credit

substitution, an increase 𝐾𝐾𝑒𝑒𝑅𝑅𝐾𝐾𝐾𝐾𝑗𝑗 should lead to larger loans issued by lender j.

An important choice in constructing the reference group is the granularity of

geographical area. The PSD provides geographical information on the location of the

property at both post code and regional level. Choosing a narrow geographical area, such

as at the post code level, carries the risk that not all relevant lenders that actively compete

in the local mortgage market, and hence are available to borrowers, will be picked up by

the reference group. On the other hand, choosing a very wide geographical area, such as

one of the 12 regions in the UK, will probably lead to the inclusion of lenders that not are

competing in the local mortgage market. The Office of National Statistics (ONS) classifies

the UK according to the Nomenclature of Units for Territorial Statistics (NUTS).9 Figure 3

shows all three levels of this classification and how they are spread across the UK. The

broadest NUTS classification consists of 12 areas, corresponding to the 12 regions of the

UK, while the most detailed classification, level three, classifies the UK into 134

geographical areas. This is the most detailed level for which statistics, such for example

local unemployment rates, are provided. While the choice of the geographical area is, of

course, to a certain extent arbitrary, the most detailed NUTS classification appears to have

the right degree of granularity: The size of geographical area included in the classification

makes the risk of excluding lenders that compete in the local mortgage market and

including those who do not seem small.

In this section, I try to examine to which extent unaffected lenders offset the loan

contraction as a result of an increase in capital requirements on regulated institutions. In

particular, I construct a reference capital requirement for this purpose. This allows me to

examine if unaffected lenders expand lending in NUTS areas that they are exposed to and

that are home to competing lenders which have been subject to an increase in capital

requirements. While I can identify the NUTS area that lenders lend to, the PSD only

9 NUTS was created by the European Office for Statistics (Eurostat) as a single hierarchical classification of spatial units used for statistical production across the European Union.

18

allows me observe actual loan outcomes as oppose to loan applications. It is therefore not

possible to know if unaffected lender borrowers applied for a loan elsewhere first and

then switched following rejection, due to an increase in the capital requirement of the

lender they first applied to. Similarly, mortgage contracts and the associated deals offered

by lenders may change following an increase in capital requirements, prompting

individuals to borrow from an alternative lender within the same local authority, even

before an application was made. By matching lenders that were subject to an increase in

capital requirements to those that were not, based on their exposure to a particular NUTS

area, I need to assume that either one of these channels operates. Unless there is a

complete lack of overlap of borrowers between two lenders within the same area, the

approach I propose here is not an unreasonable first attempt to test my hypothesis.

In the presence of loan substitution between unaffected and affected lenders, due

to an increase in the capital requirement on the latter, one would therefore expect a

positive correlation between the reference capital requirement and lending by an

unaffected lender: If the average mortgage size extended by a lender, that was subject to

an increase in the capital requirement, declines, individuals that require larger loans may

borrow from an unaffected lender instead. Formally, I test for loan substitution between

competing lenders who were and were not subject to a rise in their capital requirement

with the following regression model:

𝑳𝑳𝑳𝑳𝑳𝑳𝑳𝑳𝒊𝒊,𝒋𝒋,𝒕𝒕 = 𝜶𝜶𝒋𝒋 + ∑ 𝝈𝝈𝒌𝒌𝑲𝑲𝒆𝒆𝑹𝑹𝑲𝑲𝑲𝑲𝒋𝒋,𝒕𝒕−𝒌𝒌𝟑𝟑𝒌𝒌=𝟎𝟎 + 𝜸𝜸𝑲𝑲𝒆𝒆𝑹𝑹𝑳𝑳𝝋𝝋𝒋𝒋,𝒕𝒕 + 𝜸𝜸𝑳𝑳𝝋𝝋𝒋𝒋,𝒕𝒕 + 𝝋𝝋𝝋𝝋𝑳𝑳𝝋𝝋𝒊𝒊,𝒋𝒋,𝒕𝒕 + 𝑷𝑷𝝋𝝋𝒉𝒉𝒕𝒕 + 𝒆𝒆𝒊𝒊,𝒋𝒋,𝒕𝒕 (2)

where 𝐿𝐿𝑃𝑃𝐿𝐿𝐿𝐿𝑖𝑖,𝑗𝑗,𝑡𝑡 is the logarithm of the loan size for a new loan taken out by

borrower i from lender j, who was not subject to the rise in capital requirements, at time

t. 𝐾𝐾𝑒𝑒𝑅𝑅𝐾𝐾𝐾𝐾𝑗𝑗,𝑡𝑡 is the reference group capital requirement for lender j time t. Since the

horizon at which borrowers switch from an affected to an unaffected lender is unclear, I

use the contemporaneous value of the capital requirement together with three lags, as in

model (1). 𝐾𝐾𝑒𝑒𝑅𝑅𝐿𝐿𝐵𝐵𝑗𝑗,𝑡𝑡 are the corresponding reference groups for all of the affected lender

balance sheet control variables, including provisions, used in model (1). 𝐵𝐵𝐿𝐿𝐵𝐵𝑖𝑖,𝑗𝑗,𝑡𝑡 is a vector

of all the borrower and loan characteristics listed in table 1, for individual i, borrowing

19

from lender j at time t. Where borrower and loan characteristics take values that are not

binary, they are modelled as additional sets of fixed effects. 𝐿𝐿𝐵𝐵𝑗𝑗,𝑡𝑡 is a vector of lender

characteristics for lender j at time t, all of which are listed in table 1. 𝛼𝛼𝑗𝑗 is a lender fixed

effect to account for lender unobservable time-invariant characteristics. 𝑃𝑃𝐵𝐵ℎ𝑡𝑡 is a vector of

postcode time effects to account for unobservable geographical differences, in particular

in loan demand, among the approximately 10,000 different three character postcode

districts of the UK over time. To avoid perfect multi-colinearity, I drop the first variable

from each group of fixed effects. 𝑒𝑒𝑖𝑖,𝑗𝑗,𝑡𝑡 is assumed to be a normally distributed error term.

In the presence of lender j fixed and postcode time effects, ∑ 𝜎𝜎𝑘𝑘3𝑘𝑘=0 can be

interpreted as the sum of coefficients on the difference between lender j’s time-invariant

capital requirement and the reference group capital requirement. As before, the presence

of postcode time effects allows me to control for demand and interpret this sum of

coefficients as a loan supply effect. Economic theory gives clear predictions about the sign

of ∑ 𝜎𝜎𝑘𝑘3𝑘𝑘=0 . When capital requirements of locally competing lenders rise, the affected

lenders can either offer smaller loans to all borrowers or stop lending to some borrowers.

In the case of the latter, these borrowers may turn to local competitors who would be

willing to lend to them instead. If this type of loan substitution operates, then one would

expect the sign of ∑ 𝜎𝜎𝑘𝑘3𝑘𝑘=0 to be positive. In other words, one would expect that lenders

which were not affected by an increase in their capital requirement, to provide loan

substitution to offset the local credit contraction.

Table 7 present estimates of model (2). The standard errors in all of the

specifications are clustered by bank-time, though the results are robust to alternative

assumptions on the standard errors of the model. Column one of table 7 reports estimates

of ∑ 𝜎𝜎𝑘𝑘3𝑘𝑘=0 , controlling for unaffected lender and postcode time fixed effects, as well as all

of the mortgage and borrower characteristics listed in table 1. Columns two and three add

several reference group balance sheet control variables. Columns four to five add the

unaffected lender balance sheet variables. Regardless of specification, ∑ 𝜎𝜎𝑘𝑘3𝑘𝑘=0 is always

positive and statistically significant with a value of between .039 and .053. Unaffected

20

lenders thus increase the size of the mortgages they issue by between 3.9% and 5.3%

following an increase in the capital requirement of a locally competing institution.

Quantitatively, these numbers are very similar to the loan contraction of between 4% and

5.6% documented in table 5, but with the opposite sign. This suggests that the loan

contraction as a result of an increase in capital requirements is completely offset locally.

Table 8 examines how ∑ 𝜎𝜎𝑘𝑘3𝑘𝑘=0 varies by borrower and loan characteristic,

analogous to table 6 in the previous section. This allows me to explore which type of

borrowers are awarded larger loans following increases in capital requirements on

competing institutions. Interestingly, column one suggests that individuals who are self

employed, have impaired income or are first time buyers are affected less by the loan

contraction. Though the coefficients on the impaired income and first time buyer

interaction are not individually statistically significant, the sum of these coefficients and

∑ 𝝈𝝈𝒌𝒌𝟑𝟑𝒌𝒌=𝟎𝟎 are not statistically different from zero. Columns two, five and six show that

individuals whose income has been verified and those who take out low LTV mortgages

benefit the most from this loan expansion. Quantitatively, the coefficients on these

interactions are almost identical, but of opposite sign, to their counterparts in table 6.

Overall, this suggests that those borrowers that were most affected by the loan

contraction also benefit the most from the loan expansion. Clearly, because the dataset

lacks information on loan applications and explicit identities of individual borrowers, I

cannot claim that these are indeed the same people moving from one lender to another.

However, since it is the same group of borrowers that is affected and the magnitudes of

the coefficients are so similar, but of opposite sign, it seems likely that this is the case.

Table 9 estimates model (2) for non-bank finance companies. In other words, the

dependent variable, together with all of the loan and borrower characteristics, now refers

to loans issued by non-bank finance companies, often referred to as ‘the shadow banking

system’. The summary statistics in table 4 suggest that, compared to the sample overall,

these lenders typically lend to higher LTV borrowers. Whether or not they provide a

source of credit substitution is an important economic policy question. The estimates of

21

∑ 𝜎𝜎𝑘𝑘3𝑘𝑘=0 are not statistically significant in any of the specifications in table 9. This suggests

that non-bank finance companies did not provide a source of credit substitution in

response to an increase in capital requirements on regulated lenders. This should not be

surprising as non-bank finance companies specialise in high LTV borrowers, while most

of the loan contraction associated with higher capital requirements affected low LTV

borrowers. Furthermore, most of the local credit contraction seems to have been offset by

those banks and building societies that did not experience an increase in their capital

requirement.

In summary, the results presented in this section suggest that locally competing

and unaffected banks and building societies expand their loan supply in response to the

loan contraction associated with an increase in the capital requirements of affected

lenders. Quantitatively, the size of the loan expansion is almost identical to the

contraction associated with the increase in capital requirements. Furthermore, I also find

that the type of borrowers who were affected the most by the loan contraction, are also

those that were affected the most by the loan expansion. This suggests that it is probably

individuals in the same group of borrowers that switch lenders locally. But no evidence of

credit substitution from non-bank finance companies, also known as the ‘shadow banking

system’ is found.

4. Conclusion

Countries around the world have introduced macroprudential regulation to

increase the resilience of the financial system to socially costly financial crises. One

proposed instrument, which is also embedded in Basel III, is a time-varying capital

requirement. But, to date, there is only little understanding of how this instrument will

affect the mortgage loan supply, lenders’ preferences for risk or credit substitution from

providers outside the banking system. The UK’s unique regulatory regime, where banks

and building societies were subject to time-varying capital requirements, together with a

new loan-level database, containing a large set of loan and borrower characteristics on all

22

new mortgages issued in the UK between 2005Q2 to 2007Q2, allows me to provide the

first empirical examination of these important questions.

The purpose of this paper is to examine the impact of changes to capital

requirements on mortgage loan supply. Banks and building societies in the UK were

subject to time-varying capital requirements, which varied by institution and over time.

Economic theory suggests that if an increase in capital requirements is binding and the

Miller-Modgliani (1958) theorem fails, then the affected institution will either need to

raise capital or reduce risk-weighted assets to satisfy the new requirement. Previous

work, such as Aiyar, Calomiris and Wieladek (2014a), Francis and Osborne (2009) or

Bridges et al (2014) tests this last implication on balance sheet level PNFC lending. In

contrast, I examine the effect of changes to an institution’s capital requirement on

individual household mortgage loans, the granularity of which allows me to better

control for loan demand and address issues of reverse causality. The affected institution

may also choose to raise capital through the retained earnings channel. Under Basel I, the

regulatory framework during this period, household mortgages were assigned a risk

weight of 50% regardless of their individual characteristics. Faced with a rise in the

capital requirement, banks were therefore incentivised to reallocate lending towards

more profitable high risk borrowers to grow capital via retained earnings. To my

knowledge, I am the first to formally test for this risk shifting channel following changes

in capital requirements. Finally, I am also the first to examine mortgage loan substitution

from banks that were unaffected by changes in capital requirements and non-bank

finance companies, frequently referred to as the ‘shadow banking system’.

I find that a rise in an affected banks capital requirement of about 100 basis points

leads to a decline in loan size of about 4.8%. My results suggest that this effect is

asymmetric: Mortgage loan supply only reacts to a rise, but not to a decline, in capital

requirements. I also find strong evidence for risk shifting: This loan contraction does not

affect borrowers with an impaired credit history or high Loan-to-Value (LTV) ratios.

Conversely, borrowers with verified income and low LTV ratios are affected to a greater

23

extent. Interestingly, there is strong evidence that locally competing banks, which were

not affected by changes in capital requirements, expand lending to exactly this group of

borrowers. Quantitatively, this loan expansion seems to completely offset the loan

contraction associated with higher capital requirements of affected banks. I do not find

evidence for credit substitution from non-bank finance companies.

But it is important to emphasize that these effects are based on observed sample

averages during this particular sample period. In theory, higher capital requirements

could increase lending at banks with very low or negative net worth: If capital

requirements help to prevent or overcome a so‐called “debt overhang” problem, which

can occur at very low capital ratios, then in principle, higher capital could encourage

lending. Furthermore, my results measure short‐term loan‐supply reactions. It is not

surprising that a decline in the loan supply is associated with a transition to higher capital

requirements, but in the longer run, improvements in the stability of the banking system

that result from higher capital requirements could improve banks’ ability to raise funds in

the market and thereby mitigate the short-run declines in loan supply documented here.

The findings in this study are also based on changes to microprudential capital

requirements and therefore subject to the Lucas Critique, since bank behaviour might

change in the presence of macroprudential regulation. But they may nevertheless offer

important lessons for economic policy and theory. Specifically, my results suggests that

increases in capital requirements intended to make the banking system more resilient,

through risk shifting, may actually have the opposite effect. Furthermore, competition in

local lending markets may mute the loan contraction, associated with greater capital

requirements on an individual institution, entirely. Both of these issues are likely to be

addressed by Basel III, which envisions a banking system wide capital requirement and

risk weights that vary with the riskiness of the underlying loan. But awareness of them

will help policy makers with the implementation of Basel III and the conduct of

macroprudential policy to ensure financial resilience, whilst minimising any possible

distortions to the mortgage market.

24

References

Aiyar, S, Calomiris, C and Wieladek, T (2014a), ‘Does macro-pru leak? Evidence from a UK policy experiment,’ Journal of Money, Credit and Banking 46(1), pages 181-214.

Aiyar, S, Calomiris, C and Wieladek, T (2014b), ‘Identifying Sources of Credit Substitution when Bank Capital Requirements are Varied’, Economic Policy 29(77), pages 45-77 Aiyar, S, Calomiris, C, Hooley, J, Korniyenko, G and Wieladek, T (2014), ‘The International Transmission of Bank Minimum Capital Requirements’, Journal of Financial Economics, 113(3), pages 368-382. Alfon, I, Argimón, I and Bascuñana-Ambrós, P (2005), ‘How individual capital requirements affect capital ratios in UK banks and building societies,’ Working Paper 515, Bank of Spain. Benford, J and Nier, E (2007), ‘Monitoring Cyclicality of Basel II Capital Requirements,’ Bank of England Financial Stability Paper No.3. Bernanke, B (1983), ‘Nonmonetary effects of the financial crisis in the propagation of the great depression,’ American Economic Review 73, 257-276. Bridges, J, Gregory, D, Nielsen,M, Pezzini, S, Radia, A and Spaltro, M (2012), ‘The Impact of Capital Requirements on Bank Lending,’ Bank of England WP 486. Ediz, T, Michael, I, and Perraudin, W (1998), ‘The impact of capital requirements on UK bank behavior,’ Federal Reserve Bank of New York Economic Policy Review, Vol. 4(3), Pages 15-22. Francis, W and Osborne, M (2009), ‘Bank regulation, capital and credit supply: Measuring the impact of Prudential Standards,’ FSA Occasional paper no. 36. FSA (2008), ‘The supervision of Northern Rock: a lesson learned review’ Furlong, F (1988), ‘Changes in Bank Risk-Taking, Federal Reserve Bank of San Francisco Economic Review, Spring, pp 45-56. Galati, G and R Moessner (2011), ‘Macroprudential policy – A literature review’, BIS Working Paper No. 337 Haldane, A (2013), ‘Constraining Discretion in Bank Regulation’, Speech at Federal Reserve Bank of Atlana, April 2013. Jimenez, G, Salas, J S,Ongena, S and Peydro, J (2011), ‘Macroprudential Policy, Countercyclical Bank Capital Buffers and Credit Supply: Evidence from the Spanish Dynamic Provisioning Experiment,’ European Banking Center Discussion Paper.

25

Kahane, Y (1977), ‘Capital Adequacy and the Regulation of Financial Intermediaries’, Journal of Banking and Finance, Vol 1, Issue 2, pp 207-18. Khwaja, A and Mian, A (2008), 'Tracing the effect of bank liquidity shocks: evidence from an emerging market’, American Economic Review, Vol. 98, No.4, pages 1413A42. Kim, D, and Santomero, A (1988), ‘Risk in Banking and Capital Regulation’, Journal of Finance, Issue 43, Vol 5, pp 1219-33. Miller, M and Modigliani, F (1958), ‘The Cost of Capital, Corporation Finance and the Theory of Investment,’ American Economic Review, Vol. 48, Pages 261-297. Myers, S and Majluf, N (1984), ‘Corporate financing and investment decisions when firms have information that investors do not have,’ Journal of Financial Economics, Vol. 13, Pages 187-221. Peek, J and Rosengren, E (1997), ‘The international transmission of financial shocks: The case of Japan,’ American Economic Review, Vol. 87, Pages 495-505. Peek, J and Rosengren, E (2000), ‘Collateral damage: Effects of the Japanese bank crisis on real activity in the United States,’ American Economic Review, Vol. 90, Pages 30-45. Sheldon, G (1996), ’Capital Adequacy Rules and the Risk-Seeking Behaviour of Banks: A Firm-Level Analysis,’ Swiss Journal of Economics and Statistics, Vol 132, pp 709-734. Turner, A (2009), ‘The Turner Review: A regulatory response to the global banking crisis’

26

Table 1

Bank Level Variables Variable Definition Source Notes

BBKR – Banking Book capital

requirement ratio

FSA-set minimum ratio for capital-to-risk weighted assets (RWA) for the banking book.

BSD3/ QFS

BSD3 provides this information for Banks. QFS provides it for Building societies.

Size of Household in Total Lending

AL18/AL19 AL AL 18 – Household lending to UK residents AL 19 – Total lending to UK residents

Size Natural log of (BT40) BT BT40 – Total Assets

Liquidity (BT21+ BT32D)/BT40 BT BT21 - Cash BT32D – Holdings of Government Stock

Core Funding (BT2H +BT3H)/BT40 BT BT2H – Retail Sight Deposits BT3H – Retail Time Deposits

Provisions PL20B/BT40 PL/BT PL20B – Provisions Return on Assets PL15/BT40 PL/BT PL15 - Profits

Borrower and Mortgage Characteristic Variables - All from FSA PSD Variable Definition Notes

loan Loan size in pound sterling Borrower Age The age of the borrower

Income verified Takes value of 1 if income verified, 0 otherwise. Mortgage Term Remaining mortgage maturity

Mortgage Payment Protection

Takes value of 1 if payment insurance taken out, 0 otherwise

Employment Status Takes distinct numerical value for each status

Status: 1) Employed, 2) Self-employed, 3) Retired or 4) Other.

Borrower type Takes distinct numerical values for each type

Type: 1) First Time Buyer, 2) Home Mover, 3) Remortgagor, 4) Social tenant, 5) Other, 6)Unknown

Repayment type

Takes distinct numerical value for each type

Type: 1) Capital and Interest 2) Interest only (endowment), 3) Interest only (pension), 4) Interest only (Unknown), 5) Mix of ‘capital and interest’ and ‘interest only’ 6) Unkown

Remortgage reason

Takes distinct numerical value for each reason

Reason: 1) No extra money raised, 2) Extra money for home improvement, 3) Extra money for debt consolidation, 4) Extra money for home improvement and debt consolidation, 5) other

Rate Type Takes distinct numerical value for each rate type

Type: 1) Fixed, 2) Discount, 3) Trackers, 4) Capped, 5) other.

27

Table 2 – Loan size and LTV ratio by Lender and Borrower

Type Loan Size LTV Ratio Number of

observations Mean St. Dev. Median Mean St. Dev. Median

Whole sample 11.48 .76 11.54 63.25 24.98 67.79 5655181

Bank 11.47 .78 11.53 63.22 25.14 67.74 5267527

Non-bank finance 11.61 .61 11.58 74.38 19.34 80.67 387654

Employed 11.45 .73 11.51 64.30 24.51 68.53 4450023

Self-Employed 11.76 .75 11.81 49.77 22.88 45.16 955831

Income verified 11.44 .81 11.52 66.68 25.89 73.95 3269728

Income unverified 11.53 .70 11.57 58.93 22.83 61.53 2385453

Impaired Credit

History

11.47 .58 11.47 71.9 19.04 75.92 228972

28

Table 3 – Balance Sheet Data Description

Variable Mean St. Dev. Median 25th 75th

BBKR 10.07 1.38 10 8 11

Size 8.91 2.42 8.96 7.16 10.74

Liquid Assets 0.011 0.024 0.001 0.00 .01

Core Funding 0.42 0.23 0.43 0.24 0.55

Fraction of Mortgage Asset

on Balance sheet

0.45 0.31 0.39 0.21 0.71

Return on Assets 0.0033 0.006 0.0026 0.001 .004

Table 4 – Description of Changes in Capital Requirements

Number of

Changes

Mean Standard

Deviation

Capital Requirement increase 12 -0.89 0.67

Capital Requirement decline 8 0.81 0.47

29

Table 5 – The Impact of Capital Requirements on Loan Size

Dependent Variable: Log of Loan Size KR -0.056** -0.04* -0.05** -0.048** -0.044* (4.671) (2.902) (4.678) (4.332) (3.349) KR*Mortfrac -0.001 (0.99) KR positive -0.047** (4.04) KR negative 0.024 (0.24) Provisions -0.11** -0.108** -0.10** -0.10** -0.074* -0.11** (6.35) (6.06) (6.14) (6.37) (3.54) (6.35) Size 0.0172 0.0346 0.0326 0.0319 (0.0375) (0.0328) (0.0330) (0.0352) Corefund -0.00691** -0.00831*** -0.00873*** -0.00229 (0.00282) (0.00294) (0.00308) (0.00256) Mortfrac 0.00452 0.0131 0.00299 (0.00513) (0.0116) (0.00514) Liquidity -0.989* -0.991* -0.689 (0.585) (0.588) (0.503) ROA -0.398 -0.520 -0.209 (1.989) (1.984) (1.946) Borrower Controls X X X X X X Bank Fixed Effects X X X X X X Postcode time Effects

X X X X X X

Observations 4,257,706 4,257,706 4,257,706 4,257,706 4,257,706 4,257,706 R-squared 0.635 0.635 0.635 0.635 0.635 0.636

This table presents results from panel regressions of regulated financial institutions. The dependent variable is the logarithm of loans issued by these institutions. All specifications include bank fixed effects and postcode time effects. I use the contemporaneous and 3 lags of each of the first 2 variables: the level of the capital requirement ratio and the provisions to total asset ratio. I report the sum of coefficients and F-statistics in parenthesis for these variables. The bank balance sheet control variables are size, defined as log of total assets; the fraction of retail to total liabilities; the fraction of mortgage to total assets; the fraction of liquid to total assets and profits dived by total assets. All of them enter contemporaneously. For these variables, t statistics are reported in parenthesis below. Borrower and loan characteristics include: 1) Borrower age; 2) Binary dummy variables for the following types of interest rate: Fixed rate, Discounted variable rate, Tracker, Capped rate, Standard variable rate or Other ; 3) Term of mortgage in years; 4) Binary dummy variables for the method of repayment: Capital and interest, Interest only (endowment), Interest only (ISA), Interest only (pension), Interest only (type unknown), Mix of capital and interest and interest only, Unknown; 5) Binary dummy variables for the purpose of remortgage: No extra money raised, Extra money for home improvements, Extra money for debt consolidation, Extra money for home improvements and debt consolidation; 6) A dummy variable for mortgage protection payment insurance; 7) A dummy variable for income verification; 8) Binary dummy variables for type of employment: Employed, Self-employed, Retired, Other, Unknown. 9) A dummy variable for impaired credit history; 10) Binary dummy variables for the borrower type: 1) First Time Buyer, 2) Home Mover, 3) Remortgagor, 4) Social tenant, 5) Other, 6) Unknown. For statistical significance, I use the following convention throughout: *** p<0.01, ** p<0.05, * p<0.1. Standard errors are clustered by bank-time.

30

Table 6 – The Impact of Capital Requirements on Loan Supply by Borrower/Loan Characteristic

Dependent Variable: Log of Loan Size KR -0.067*** 0.06** -0.042* -0.047** 0.056*** 0.111*** (7.247) (3.794) (3.390) (4.455) (9.560) (16.46) KR*Self Employed 0.03*** 0.00368 (26.44) (0.483) KR*Impaired Credit 0.04*** -2.17e-07 (11.71) (1.11e-09) KR*First Time Buyer 0.062*** 0.009** (8.128) (5.002) KR*High Income 0.00518 -0.012*** (1.564) (8.348) KR*Low Income 0.00839 0.041*** (1.087) (20.07) KR*Income Verified -0.12*** -0.086*** (19.30) (20.23) KR*High Maturity 0.044*** 0.0051* (20.45) (3.172) KR*Low Maturity -0.0113 0.0451 (1.18) (24.32) KR*Flexible -0.0143 0.0106 (1.595) (2.530) KR*Fixed Interest Rate 0.000873 -0.0084*** 0.0321 (7.918) KR*Interest Only 0.02*** 0.0044* (10.92) (3.752) KR*SVR 0.0151 0.056*** (0.723) (17.98) KR*High LTV 0.00747 0.033*** 2.369 (13.21) KR*LOW LTV -0.178*** -0.19*** (18.29) (18.74) Balance Sheet Controls X X X X X X Borrower Controls X X X X X X Postcode Time Effects X X X X X X Bank Fixed Effects X X X X X X R-squared 0.635 0.639 0.637 0.635 0.826 0.829

This table presents results from panel regressions of regulated financial institutions. The dependent variable is the logarithm of loans issued by these institutions. The main explanatory variable is the capital requirement ratio, which enters contemporaneously and with three lags. This variable is interacted with the several binary dummy variables which reflect the following borrower/loan characteristics: 1) Self Employment; 2) Impaired credit history; 3) First Time Buyer; 4) High/Low Income/LTV/Maturity if borrower income/LTV/maturity is above/below the 75/25% of the relevant distribution; 5) Flexible mortgage features; 6) Income verification; 7) Interest only mortgage; 8) Fixed interest rate mortgage; 9) Standard variable rate mortgage. I report the sum of coefficients and F-statistics in parenthesis for all of these variables. The remainder of the model specification is identical to column four in table 5. For statistical significance, I use the following convention throughout: *** p<0.01, ** p<0.05, * p<0.1. Standard errors are clustered by bank-time.

31

Table 7 – The Impact of Reference Group Capital Requirements on Unaffected Lender Loan

Supply

Dependent Variable: Log of Loan Size Reference KR 0.039*** 0.047*** 0.039** 0.036* 0.053** (8.61) (10.51) (5.95) (3.67) (5.37) Reference Provisions -0.0459 -0.0325 -0.0358 -0.0476 (1.00) (0.42) (0.51) (0.85) Provisions -0.318 0.730 (1.70) (2.70) Reference Size -0.0905 -0.0761 -0.0563 (0.0895) (0.0919) (0.0922) Reference Corefund -0.00556* -0.00482 -0.00267 (0.00319) (0.00329) (0.00331) Reference Mortfrac 0.0164 0.0182 0.0222 (0.0163) (0.0162) (0.0170) Reference Liquidity -0.542 -0.490 -0.0629 (0.565) (0.581) (0.642) Reference ROA -4.031 -4.636 -5.137 (3.712) (3.658) (3.783) Size -0.0652 (0.0684) Corefund 0.261 (0.259) Mortfrac 0.115 (0.284) Liquidity 6.110 (10.54) ROA -12.23 (40.82) Borrower Controls X X X X X Postcode Time Effects X X X X X Bank Fixed Effects X X X X X Observations 1,278,430 1,181,375 1,181,375 1,181,375 1,181,375 R-squared 0.625 0.622 0.622 0.622 0.622

This table presents regression results of the reference group capital requirement on lending by regulated lenders which were not subject to a change in capital requirements, which I refer to as unaffected lenders below. The dependent variable is the logarithm of loans. I use the contemporaneous and 3 lags of each of the first 3 variables: the reference group capital requirement ratio, the reference group provisions to total asset ratio and the provisions to total asset ratio. I report the sum of coefficients and F-statistics in parenthesis for these variables. For all other variables, t-statistics are reported in parenthesis below the coefficients. The reference group capital requirement is constructed in the following way: First I compute the implied capital requirement of lenders affected by capital requirements changes for that particular geographical area. The reference group capital requirement is given by the sum of the product of this implied capital requirement and the unaffected lenders exposure to that particular geographical area. See the main text for more details. All variables preceded by the work ‘Reference’ are defined in an analogous way. See the footnote of table 5 for a detailed description of the lender balance sheet, borrower and loan characteristics. For statistical significance, I use the following convention throughout: *** p<0.01, ** p<0.05, * p<0.1. Standard errors are clustered by bank-time.

32

Table 8 – The Impact of Reference Group Capital Requirements on Unaffected Lender Loan

Supply by Borrower/Loan Characteristic

Dependent Variable: Log of Loan Size Reference KR 0.072*** -0.0985 0.054** 0.083*** -0.0318 -0.10 (7.45) (2.01) (4.68) (10.85) (0.64) (2.42) Reference KR*Self Employed -0.051** -0.002 (4.40) (0.031) Reference KR*Impaired Credit -0.0442 0.025*** (0.941) (9.153) Reference KR*First Time Buyer -0.0687 -0.000492 (1.14) (0.0016) Reference KR*High Income 0.00455 0.019* ((0.17) (3.032) Reference KR*Low Income -0.0192 -0.037* (1.710) (3.132) Reference KR*Income Verified 0.178** 0.113** (5.750) (5.397) Reference KR*High Maturity -0.0190 0.0129 (0.730) (0.957) Reference KR*Low Maturity 0.00250 -0.0564** (0.00849) (5.149) Reference KR*Flexible 0.0914 1.055 (0.763) (0.306) Reference KR*Fixed Interest Rate -0.0169 0.00885 (1.239) (1.618) Reference KR*Interest Only -0.07*** -0.00985 (9.682) (1.429) Reference KR*SVR -0.0406 -0.0299 (0.724) (1.461) Reference KR*HIGH LTV 0.00442 3.918 (0.947) (0.0493) Reference KR*LOW LTV 0.191 0.21* (2.725) (3.00) Balance Sheet Controls X X X X X X Reference Balance Sheet Controls X X X X X X Postcode Time Effects X X X X X X Bank Fixed Effects X X X X X X R-squared 0.622 0.625 0.624 0.622 0.834 0.836

This table presents regression results of the reference group capital requirement on lending by regulated lenders which were not subject to a change in capital requirements. The dependent variable is the logarithm of loans. The main explanatory variable, the reference group capital requirement, Reference KR, and the balance sheet controls are described in the footnote to table 7. This variable is interacted with the several binary dummy variables which reflect the following borrower/loan characteristics: 1) Self Employment; 2) Impaired credit history; 3) First Time Buyer; 4) High/Low Income/LTV/Maturity if borrower income/LTV/maturity is above/below the 75/25% of the relevant distribution; 5) Flexible mortgage features; 6) Income verification; 7) Interest only mortgage; 8) Fixed interest rate mortgage; 9) Standard variable rate mortgage. I report the sum of coefficients and F-statistics in parenthesis for all of these variables. See the footnote of table 5 for a detailed description of the borrower and loan characteristics. For statistical significance, I use the following convention throughout: *** p<0.01, ** p<0.05, * p<0.1. Standard errors are clustered by bank-time.

33

Table 9 – The Impact of Reference Group Capital Requirements on Non-Bank Finance Company

Loan Supply

Dependent Variable: Log of Loan Size Reference KR -0.0120 -0.0224 -0.0224 -0.0234 (1.816) (1.428) (1.308) (1.544) Reference Provisions -0.134 -0.865 -0.611 (1.053) (1.116) (0.366) Reference Size -0.0197 -0.0101 (0.0249) (0.0335) Reference Corefund -0.0600 -0.101 (0.0986) (0.118) Reference Mortfrac 0.0400 0.103 (0.0833) (0.103) Reference Liquidity -15.68 (26.97) Reference ROA 15.39 (27.59) Borrower Controls X X X X Postcode Time Effects X X X X Bank Fixed Effects X X X X Observations 175,616 175,616 175,616 175,616 R-squared 0.887 0.887 0.887 0.887

This table presents regression results of the reference group capital requirement on lending by nonbank finance companies. The dependent variable is the logarithm of loans. I use the contemporaneous and 3 lags of the reference group capital requirement ratio provisions and report the sum of coefficients and F-statistics in parenthesis for these variables. For all other variables, t-statistics are reported in parenthesis below the coefficients. The reference group capital requirement is constructed in the following way: First I compute the implied capital requirement of lenders affected by capital requirements changes for that particular geographical area. The reference group capital requirement is given by the sum of the product of this implied capital requirement and the unaffected lenders exposure to that particular geographical area. See the main text for more details. All variables preceded by the work ‘Reference’ are defined in an analogous way. See the footnote of table 5 for a detailed description of the lender balance sheet, borrower and loan characteristics. For statistical significance, I use the following convention throughout: *** p<0.01, ** p<0.05, * p<0.1. Standard errors are clustered by bank-time.

34

Figures

Figure 1

Lender i CAPITAL REQUIREMENT ↑

& Binding & M-M fails

Raise Capital from Retained Earnings or Outside Investor (costly)

Contract Risk-Weighted Assets

CREDIT CONTRACTION

Aggregate Credit Supply reaction?

Raise earnings via reallocation of lending to higher risk/ higher LTV borrowers

Non-bank finance company substitutes

Unaffected lender substitutes

Credit

Expansion

35

Figure 2: Histogram of changes in capital requirements

Source: BSD3 form

0

1

2

3

4

5

6

-2.5 -1.5 -1 -0.5 -0.25 0.25 0.5 0.75 1 1.5

36

Figure 3 – UK Nomenclature of Territorial Statistics (NUTS) Boundaries

Source: ONS.

37