Policy uncertainty, lender of last resort and the real economy

February 08, 2021 at 09:59 am EST

Martina Jasova, Caterina Mendicino, Dominik Supera

Working Paper Series

No 2521 / February 2021

Disclaimer: This paper should not be reported as representing the views of the European Central Bank (ECB). The views expressed are those of the authors and do not necessarily reflect those of the ECB.

Abstract

We show that a reduction in lender of last resort (LOLR) policy uncertainty positively aects bank lending and propagates to investment and employment. We exploit a unique policy that reduced uncertainty regarding the availability of future LOLR funding for banks as a quasi-natural experiment. Using micro-level data on banks, rms and loans in Portugal, we generate cross-sectional variation in banks' exposure to uncertainty and nd that the size of the haircut subsidy - the gap between private market and central bank security valuations - plays a key role in the propagation of the shock to lending and the real economy.

JEL classication: E44, E52, E58, G21, G32

Keywords: Bank Credit, Haircut Subsidy, Central Bank Liquidity, Policy Uncertainty,

Firm-level Employment and Investment

ECB Working Paper Series No 2521 / February 2021

Non-technical Summary

How do banks respond to a sudden reduction in uncertainty surrounding the lender of last resort (LOLR) liquidity policy during periods of nancial distress? What are the implications for lending and the real economy?

This paper addresses these questions using granular bank-, rm- and loan-level data from Portugal and a unique policy change by the European Central Bank that reduced uncertainty regarding the availability of LOLR funding for euro area banks as a quasi-natural experiment. Our identication strategy relies on exploiting banks' cross-sectional variation in the ex-ante exposure to the LOLR policy uncertainty in a dierence-in-dierences research design.

On December 8, 2011, the ECB unexpectedly announced a new very Long-Term Re- nancing Operation (vLTRO) with an extraordinarily long maturity of 36 months. At the time of the vLTRO banks were facing uncertainty about their ability to fully satisfy their funding needs through the ECB over an extended period of time. Hence, the vLTRO represents a suitable setting to investigate the eects of a reduction in uncertainty surrounding the future availability of central bank liquidity.

Importantly, the vLTRO was introduced when the ECB was providing banks with a haircut subsidy. We dene the haircut subsidy as a favorable gap between the central bank and private market haircut on the value of risky securities pledged to obtain repo funding. We argue that the size of the haircut subsidy plays a crucial role in determining the impact of a reduction in LOLR policy uncertainty. In the presence of a substantial haircut subsidy, uncertainty regarding the long-term availability of LOLR funding exposed banks to the risk of having to rely on more costly market funding. The size of the haircut subsidy is our main measure of cross sectional banks' exposure to LOLR policy uncertainty.

We compare the evolution of lending to rms by banks were more exposed to the policy uncertainty relative to banks with less exposure. To isolate the causal eect of the policy,

ECB Working Paper Series No 2521 / February 2021

we compare lending outcomes of the same rm borrowing in the same time from at least two dierently exposed banks, i.e. this is implemented by using rm-time xed eect in line with Khwaja and Mian (2008).

This paper exploits a unique micro-level dataset that links detailed information on banks, rms and loans. First, we match bank security holdings data with the bank pledging of securities both with the ECB and the private repo markets to measure the size of the haircut subsidy. Second, we merge the haircut subsidy measured at the cross-section of banks with the universal credit registry that collects loan-level information for all credit relationships of Portuguese rms to investigate lending outcomes. Finally, we match the data with rm- level balance sheet and linked employee-employer dataset to study the real outcomes of the reduction of policy uncertainty.

We show that a lengthening of the maturity of central bank liquidity by itself is, however, not sucient to stimulate lending. The size of the haircut subsidy is key in the propagation of a reduction in LOLR policy uncertainty to bank lending and the real economy. In the absence of a high haircut subsidy, uncertainty about the future availability of LOLR liquidity is not expected to have a large eect on the borrowing capacity of banks and thus on their lending behavior. This is because borrowing on the private repo markets is not substantially more costly than borrowing from the central bank. Thus, a high haircut subsidy is a necessary condition for the policy change to have real eects.

We nd that a decrease in LOLR policy uncertainty has a positive and economically sizable impact on banks' credit supply to rms both on the intensive and extensive margin. Banks more exposed to the reduction in LOLR policy uncertainty increased their lending to rms and oered loans of longer maturity.

The loan-level results also translate into economically relevant rm-level credit outcomes. We show that the reduction in LOLR policy uncertainty strongly transmitted to rm-level investment and employment during the sovereign debt crisis.

Our focus on the lending and real eects of uncertainty regarding central bank LOLR pol-

ECB Working Paper Series No 2521 / February 2021

icy brings a dierent perspective to the literature that emphasizes the impact of government, trade and monetary policy uncertainty on the rm-level and aggregate economic outcomes (e.g. Fernandez-Villaverdeet al., 2015; Baker et al., 2016; Husted et al., 2019; Caldara et al., 2020).

Overall, we provide evidence on the positive impact of a reduction in LOLR policy uncertainty on investment and employment through the bank-lending channel. By showing that a reduction in LOLR policy uncertainty makes banks choose to no longer delay their lending, we document that the real option channel (Bernanke, 1983) is also important for banks.

Our results suggest that in the absence of an explicit commitment to the long-term provision of LOLR funding, a lengthening of the maturity of central bank liquidity can reduce policy uncertainty and be a valid policy tool to stimulate lending. This provides new evidence on the bank lending eects of central bank policies that go beyond those of conventional monetary policy (e.g. Jimenez et al., 2012) and unconventional monetary policies (e.g. Rodnyansky and Darmouni, 2017).

ECB Working Paper Series No 2521 / February 2021

Introduction

Since the start of the Global Financial Crisis, nancial sector bailouts and central bank unconventional policies have been surrounded by a great deal of uncertainty. Many central banks revived lender of last resort (LOLR) operations and implemented important temporary changes to their liquidity operations. This, often, generated uncertainty about the circumstances and terms under which they would provide liquidity assistance.1 While the existing literature has provided important insights on the impact of government, trade and monetary policy uncertainty on the rm-level and aggregate economic outcomes (e.g. Fernandez- Villaverde et al., 2015; Baker et al., 2016; Husted et al., 2019; Caldara et al., 2020), evidence on the real eects of uncertainty regarding central bank LOLR policy is still scant.

Banks are at the core of the monetary policy transmission mechanism (e.g. Kashyap and Stein, 2000; Bernanke and Blinder, 1992; Jimenez et al., 2012). Thus, the assessment of how uncertainty regarding central bank policies aects economic outcomes requires to understand its impact on bank lending decisions. Estimating the causal eect of LOLR policy uncertainty on the real economy through its eect on the supply of credit poses important identication challenges. Crucially, uncertainty simultaneously aects the demand and supply of credit. Moreover, it is generally dicult to identify exogenous shocks to policy uncertainty, as well as, to measure the exposure to uncertainty in the cross-section of banks and rms.

We address these challenges by studying the causal eect of a sudden reduction in LOLR policy uncertainty on bank lending and its propagation to the real economy by using a unique policy change and granular micro-level data. We focus on the European Central Bank (ECB)'s 2011 very Long-Term Renancing Operation (vLTRO) as a quasi-natural experiment of a sudden reduction in LOLR policy uncertainty. The vLTRO extended the maturity of central bank liquidity from short-term to extraordinarily long three years and,

1The dierential treatment of Bear Stearns, Lehman Brothers and AIG increased uncertainty regarding the availability of the U.S. Federal Reserve liquidity support. Similarly, in the U.K., the Northern Rock crisis increased uncertainty regarding the conduct of LOLR operations by the Bank of England (Hauser, 2014).

ECB Working Paper Series No 2521 / February 2021

thus, reduced uncertainty regarding the future availability of LOLR funding for a prolonged period of time. Importantly, the vLTRO was introduced when the ECB was providing banks with a haircut subsidy, i.e. a favorable gap between the central bank and private market haircut on the value of risky securities pledged to obtain repo funding. In the presence of a substantial haircut subsidy, uncertainty regarding the long-term availability of LOLR funding exposed banks to the risk of having to rely on more costly market funding.

Our identication strategy relies on exploiting banks' cross-sectional variation in the ex- ante exposure to the LOLR policy uncertainty in a dierence-in-dierences research design. We use the size of the haircut subsidy at the individual bank level as a measure of exposure to the LOLR policy uncertainty. We analyze the credit supply impact of policy uncertainty by comparing the credit outcomes of the same rm borrowing from at least two dierently exposed banks (Khwaja and Mian, 2008; Jimenez et al., 2019). We show that during the European sovereign debt crisis, banks more exposed to the reduction in LOLR policy uncertainty provided more credit to rms, in particular of longer maturity. Further, we also document that the loan-level eects are economically sizable and translate in investment and employment eects.

Our paper has several important implications. First, we provide empirical evidence on the impact of policy uncertainty on investment and employment through its eects on the supply of credit. By showing that a reduction in LOLR policy uncertainty makes banks choose to no longer delay their lending, we document that real option channel (Bernanke, 1983; Rodrik, 1991) is also important for banks. We nd that banks more exposed to the reduction in policy uncertainty not only started to invest more in credit, but they also granted loans of longer maturities. Second, we uncover new insights into the importance of central bank commitment and forward guidance (e.g. Campbell et al., 2012). Our results suggest that in the absence of an explicit long-horizon commitment to the provision of LOLR funding, a lengthening of the maturity of central bank liquidity reduces uncertainty regarding the availability of central bank funding over an extended period of time and is a valid policy

ECB Working Paper Series No 2521 / February 2021

tool to stimulate lending and the real economy. Thus, central bank commitment or explicit guidance about its future policy intentions is also crucial in the context of lender of last resort policy. Third, the focus on LOLR policy (e.g. Bagehot, 1873; Friedman and Schwartz, 1963; Drechsler et al., 2016), allows us to provide new evidence on the bank lending eects of central bank policies that goes beyond those of conventional and unconventional monetary policies (e.g. Jimenez et al., 2012; Rodnyansky and Darmouni, 2017). Importantly, the size of the haircut subsidy is key in the propagation of LOLR policy to bank lending and the real economy.

This paper exploits a unique micro-level dataset that links detailed information on banks, rms and loans. First, we use the data on banks' security pledging with the ECB and match it with the private repo market haircuts to measure the size of the haircut subsidy. Second, we merge the haircut subsidy measured at the cross-section of banks with the universal credit registry that collects loan-level information for all credit relationships of Portuguese rms to investigate lending outcomes. Finally, we link the data with rm-level balance sheet and employee-employer datasets to study the real eects of the policy. Our paper is, to the best of our knowledge, the rst work that examines the transmission of LOLR policy changes on the real economy at this level of coverage, match and granularity.

The 2011 vLTRO represents a suitable setting to investigate the eects of a reduction in uncertainty surrounding the future availability of central bank liquidity. In October 2008, the ECB started to fully satisfy the demand for short-term liquidity from Eurozone banks against eligible collateral under xed rate full allotment policy. However, the ECB never committed to provide unlimited liquidity and, thus, to act as a LOLR, for an extended period of time. The lack of long-horizon guidance about its intentions regarding the provision of liquidity generated uncertainty for banks regarding the future possibility to fully satisfy their funding needs through the central bank for an extended period of time.2 By extending the maturity

In addition, in April and July 2011 the ECB increased interest rates, after almost two years of interest rate cuts. This created uncertainty regarding the stance of the ECB monetary policy over the medium term,

ECB Working Paper Series No 2521 / February 2021

of available funds to three years, the vLTRO suddenly reduced uncertainty regarding the longer-term availability of LOLR liquidity.3

We argue that the size of the haircut subsidy plays a crucial role in determining the impact of a reduction in LOLR policy uncertainty. In case of a withdrawal from the unlimited liquidity policy, banks borrowing from the ECB against securities with a high haircut subsidy would incur a drastic reduction in their borrowing capacity by switching to the private repo market. Using data on private market haircuts, we calculate that the average subsidy for Portuguese banks was around 70 pp before the introduction of the 2011 vLTRO. In the extreme case of a complete reliance on the private market, banks' borrowing capacity would be reduced by 57% of their equity capital (EUR 20 bn). Due to this substantial haircut subsidy, Portuguese banks heavily relied on the LOLR funding and were, largely exposed to uncertainty about its future availability. Thus, banks with ex-ante larger haircut subsidy beneted more from the vLTRO.

Our results highlight that a lengthening of the maturity of central bank liquidity by itself is not sucient to stimulate lending. Specically, we argue that the necessary condition for this policy change to work is a sizable exposure of banks to LOLR policy uncertainty, as measured by a large haircut subsidy. We document this by extending our analysis to a similar long-term liquidity operation introduced in 2009. Unlike the 2011 vLTRO, the 2009 policy was implemented in a period of negligible haircut subsidy. As a result, banks were less exposed to LOLR policy uncertainty as they could eectively substitute central bank funding with private market repo nancing without suering from a dramatic reduction in their borrowing capacity due to the haircut dierences.4 Our empirical analysis of the 2009 policy nds no impact of the lengthening of the maturity of central bank liquidity on bank

and raised concerns of a possible tightening cycle.

3Based on banks' public announcements, banks reported the reduction in uncertainty regarding the guarantee of long-term funding as the crucial reason for their participation in the vLTRO.

4The average subsidy for a Portuguese bank was around 4 pp before the introduction of the 2009 policy compared to 70 pp before 2011 vLTRO.

ECB Working Paper Series No 2521 / February 2021

lending in a period of low haircut subsidy.

Portugal presents an ideal laboratory to investigate the bank lending and real eects of a sudden reduction in LOLR policy uncertainty in times of crisis for at least three reasons. First, the high level of granularity of the data allows us to overcome a number of identication challenges and properly trace the transmission of the policy uncertainty shock to bank lending and its propagation to the rm-level outcomes and the real economy. Second, the Portuguese banking sector was highly exposed to the LOLR policy uncertainty. Due to the fear of contagion from the Greek sovereign debt crisis, in May 2010 Portuguese banks lost access to the international wholesale funding market and increased their reliance on the short- term ECB funding. At the time of the introduction of the vLTRO, Portuguese securities were considered risky and had extremely high haircuts in the private market. Thus, banks were highly exposed to uncertainty regarding the availability of future LOLR liquidity.5 Third, the focus on Portugal mitigates concerns about confounding eects related to the change of collateral rules for certain types of securities during the vLTRO. Together with the announcement of the vLTRO, the ECB also relaxed collateral rules on risky asset-backed securities and allowed national central banks to temporarily accept non-marketable securities. In sharp contrast with other European countries, the use of these types of securities was very limited in Portugal.6 Overall, Portugal represents a valuable choice for our analysis as it allows us to overcome important identication challenges.

We formally exploit the variation in banks' exposure to the reduction in LOLR policy uncertainty entailed in the vLTRO using a dierence-in-dierences framework. To isolate the causal eect of the reduction in policy uncertainty on lending, we compare the credit outcomes of the same rm borrowing from at least two dierently exposed banks. Our

Portugal received the second largest uptake of the vLTRO relative to the size of the domestic banking sector among all euro area countries.
6In Section 1, we use security-level data to carefully document that the change in collateral rules con- comitant with the vLTRO did not aect in a relevant way the composition of pledged securities used by Portuguese banks. In contrast, this increase in the collateral availability had a signicant impact on the pledging behaviour of banks during the vLTRO in other European countries.

ECB Working Paper Series No 2521 / February 2021

identifying assumption is that in the absence of the policy, the lending of more and less exposed banks would have followed parallel trends. We measure the exposure to the policy using the size of the haircut subsidy for each bank as the dierence between the ECB and the private market haircut-adjusted value of its securities pledged with the ECB to obtain liquidity.

We nd that the introduction of long-term operations improves lending outcomes by reducing banks' uncertainty regarding the availability of central bank funding over an extended period of time. This has a positive and economically sizable impact on banks' credit supply to rms both on the intensive and extensive margin. Banks more exposed to the reduction in LOLR policy uncertainty increased their actual lending to rms, allowed borrowers to increase credit line drawdowns and oered loans of longer maturity. In terms of elasticity, one standard deviation increase in bank exposure to the reduction in funding uncertainty is associated with a 3.2 percent increase in lending to non-nancial rms in Portugal. The eects are positive and signicant not only at the loan-level but importantly also at the rm-level credit. We compare the actual credit development to the counterfactual without the policy intervention and nd that although the policy did not stop the ongoing credit contraction, it signicantly reduced its pace. The observed credit contraction in the vLTRO period was about 5.75%. We estimate that in the absence of the policy, credit would have reduced by additional 2.15 percentage points (EUR 892 million).

While the reduction in policy uncertainty had a positive impact on lending volumes, it also led to a temporary loosening of lending standards. Specically, more exposed banks were more likely to establish new relationships with riskier rms and these newly extended loans defaulted more often within the subsequent three years. This provides evidence on the risk-taking channel of LOLR policy and it is in line with the risk-taking channel of monetary policy already documented for both conventional and unconventional monetary policy shocks.

Finally, we explore whether the loan-level results translate into economically relevant

ECB Working Paper Series No 2521 / February 2021

rm-level credit outcomes. A valid concern is that as rms received more funding from more exposed banks, their borrowing from other less exposed banks could have been reduced by the same proportion (a zero-sum outcome). This would imply negligible rm-level credit eects.7 We show that the reduction in LOLR policy uncertainty strongly transmitted to rm-level investment and employment during the sovereign debt crisis. To this end, we match the credit register data for all non-nancial rms in Portugal with granular rm-level census and employer-employee data. While the observed drop in investment in 2011{2012 was about 18.5%, we estimate that without the vLTRO, rms' investment would have contracted by additional 2.2 percentage points. In case of employment, we nd that while the year-on-year labor market contacted by 9.7%, in the absence of the vLTRO the employment drop would have been 2.0 percentage points more severe. Thus, the real option channel for banks prove to have important eects not only on bank lending decisions, but also for the real economy.

Literature review. This paper contributes to the growing literature that assesses the real and nancial eects of economic policy uncertainty (e.g. Pastor and Veronesi, 2012; Baker et al., 2016; Kelly et al., 2016; Jens, 2017; Husted et al., 2019; Caldara et al., 2020). We add to this literature by examining the eects of uncertainty surrounding LOLR policies. Our results highlight that the real option channel of uncertainty is important for banks and translates into rm real eects.

The paper also relates to the growing strand of the monetary economics literature which has highlighted the importance of commitment and forward guidance for interest rate policies (e.g. Wright, 2012; Campbell et al., 2012) or asset purchases (e.g. Krishnamurthy et al., 2017; Swanson, 2017). The relevance of providing guidance on central bank policy intentions has not been assessed in the realm of lender of last resort policies. Our paper lls this important gap.

By showing that managing the maturity structure of the central bank liquidity is an

See, for instance, Jimenez et al. (2019) for an example of how large loan-level eects of a credit supply shocks in Spain result in close to zero rm-level aggregate eects.

ECB Working Paper Series No 2521 / February 2021

important tool for the lender of last resort, we also contribute to existing work that analyzes banks' borrowing from the central bank during crisis times (e.g.Cassola et al., 2013; Drechsler et al., 2016; Berger et al., 2017). By reducing uncertainty about the availability of future liquidity, the central bank can provide support to banks and stimulate lending. The positive and sizable eects of the LOLR policy uncertainty on bank lending and the real economy are crucially linked to the size of the haircut subsidy.

This paper complements existing work on the eect of central bank policies related to the vLTRO. Existing papers focus on the vLTRO liquidity uptake (Andrade et al., 2018) or on the relaxation of the collateral rules concomitant with the vLTRO (e.g., Cahn et al., 2018; van Bekkum et al., 2018; Carpinelli and Crosignani, 2018) and provide evidence of its eect on the lending in a number of countries.8 We show that the lengthening of the maturity of central bank liquidity can aect bank lending independently of the relaxation of the collateral rules. Importantly, the eects are related to the exposure of banks to uncertainty regarding the long-term provision of LOLR funding. We exploit two distinctive features of our setup that allow us to examine this mechanism. First, we focus on Portugal where, dierently from other European countries, the relaxation of collateral rules was very limited.9 Second, we present a novel measure of the exposure to the policy - the haircut subsidy - that allows us to capture ex-ante bank's exposure to the reduction in LOLR policy uncertainty entailed in the vLTRO.10

8At the time of the vLTRO, the ECB (i) reduced the rating threshold for certain asset-backed securities and (ii) allowed national central banks, as a temporary solution, to accept as collateral additional performing credit claims that satisfy specic eligibility criteria. van Bekkum et al. (2018) exploit the lowering of the rating requirement for eligible residential mortgage-backed securities in the Netherlands. Carpinelli and Crosignani (2018) explore the introduction of a regulatory intervention by the Italian government that allowed banks to manufacture" collateral by guaranteeing securities, such as retained bank own bonds, otherwise ineligible at the ECB. Andrade et al. (2018) studies the impact of the endogenous vLTRO liquidity uptake in France, while Cahn et al. (2018) exploit the concurrent relaxation of the collateral rules for French banks.

In Section 1, we provide detailed evidence of the negligible role played by these newly eligible securities in the pledging of Portuguese banks during the vLTRO.
10In our analysis we also compare the impact of the 2011 vLTRO with that of the 2009 LTRO to further document the importance of the LOLR policy uncertainty channel in explaining the bank lending eects of the lengthening in the maturity of central bank liquidity. To the best of our knowledge, the 2009 LTRO was

ECB Working Paper Series No 2521 / February 2021

Finally, this paper connects to the literature on the bank lending channel of monetary policy (e.g. Jimenez et al., 2012, 2014) and in particular to the latest papers on unconventional monetary policies (e.g. Chakraborty et al., 2018; Rodnyansky and Darmouni, 2017; Altavilla et al., 2018, 2019; Bottero et al., 2019; Heider et al., 2019). We contribute to this strand of the literature by documenting the bank lending channel of LOLR policy. In addition, by showing that policy uncertainty that hits banks propagates to the real economy, we also link to existing work that quanties the aggregate eects of shocks propagated through the credit channel (see, e.g., Jimenez et al., 2019; Amiti and Weinstein, 2018; Chodorow-Reich,2014; Huber, 2018; Luck and Zimmermann, 2020).

The rest of the paper is organized as follows. Section 1 shows the institutional back- ground. Section 2 discusses the data. Section 3 presents the empirical strategy. Section 4 reports the loan-level eects, Section 5 presents rm-level credit and real outcomes. Section

concludes.

LOLR, policy uncertainty and haircut subsidy

1.1 The liquidity framework of the ECB

Until 2008, the ECB provision of liquidity was implemented in the form of repurchase agreements against eligible collateral through auctions at variable rate.11 As a reaction to the nancial crisis, in October 2008, the existing tender procedure was replaced by a xed-rate full allotment (FRFA). With FRFA, all bank bids were fully satised regardless of bids placed by other banks in the Eurozone as long as the bank could pledge sucient collateral. Thus, banks could borrow unlimited amounts of liquidity against eligible collateral.12

so far unexplored by the literature.

See Cassola et al. (2013) for details on the primary auctions of liquidity before 2008 and for the analysis of banks' bidding behavior under the multiple rate auction during the 2007 sub-prime market crisis.
We focus on regular ECB liquidity operations and abstract from the Emergency Liquidity Assistance program, which is administered by the national central bank to support banks with insucient eligible

ECB Working Paper Series No 2521 / February 2021

On 8 December 2011, the ECB unexpectedly announced a new very Long-Term Renanc- ing Operation (vLTRO) with an extraordinarily long maturity of 36 months.13 The newly available long-term funding was oered at exactly the same conditions as for the existing short-term repo.14 More than 800 banks participated to the vLTRO and the ECB allotted approximately EUR 1 trillion of funding. To date, this is the largest liquidity provision in the history of modern central banking.

1.2 LOLR policy uncertainty and vLTRO

Why was the vLTRO so popular? We argue that at the time of the vLTRO banks were facing uncertainty about their ability to fully satisfy their funding needs through the ECB over an extended period of time. The ECB only provided short-horizon guidance on the future availability of LOLR funding.15 In fact, in October 2011 (one and a half months before the vLTRO announcement), the ECB announced to maintain the FRFA policy only until mid-2012. In the absence of a long-horizon commitment to the future course of actions, the introduction of a new long-term operation reduced uncertainty regarding the availability of LOLR funding over an extended period of time.

Banks' public announcements also point towards the reduction in uncertainty regarding the availability of funding over the incoming years as the crucial reason for their participation in the vLTRO. The banks communicated that they took the opportunity to borrow from the ECB at three years, which made funding more stable and took pressure o the use of weekly borrowing operations" (Caixa Economica Monte Pio, Annual Report, 2011), as it

collateral.

Previously, the ECB o ered liquidity to banks with a maximum maturity of one year, i.e, weekly main re- nancing operations (MRO), and 1-,3-,6-monthlong-term operations (LTRO), and on two special occasions (in 2009 and 2011) the ECB introduced 12-month LTRO.
The vLTRO interest rate was a oating rate computed as an average of the weekly MRO rates set by the ECB over the horizon of three years and was paid at the maturity of the operation. Banks were required to pay a oating rate that fully mirrored any changes in the MRO rates over the horizon of three years. Thus, vLTRO was not associated with a reduction in interest rate risk compared to other shorter-term operations.
See Internet A.1 for ECB announcements concerning FRFA.

ECB Working Paper Series No 2521 / February 2021

guaranteed the same [liquidity] position for the coming two years" (Banco Carregosa, Report and Accounts, 2011), represented a structural improvement in the prole of maturities" (Caixa Geral de Depositos, Annual Report 2012), improved nancing structure by replacing short-term maturities by long term funding" (Santander, Annual Report, 2011) and enabled them to stabilize [the] structural liquidity prole" (Banco Popolare, Annual Report, 2012).16 Thus, the guidance horizon used by the ECB had left a great deal of uncertainty regarding the availability of LOLR funding. The vLTRO clearly contributed to reduce this uncertainty.

In addition, in April and July 2011 (i.e., around six months prior to the vLTRO an- nouncement) the ECB increased interest rates, after almost two years of interest rate de- creases. This increased uncertainty regarding the stance of the ECB monetary policy over the medium term, and raised concerns of a possible tightening cycle. This, potentially, also contributed to enhanced uncertainty regarding the future provision of LOLR liquidity.

1.3 vLTRO and the Portuguese banking sector

1.3.1 Liquidity provisions

Figure 1 Panel (a) summarizes the development of ECB liquidity received by Portuguese banks. In May 2010, Portuguese banks lost access to international wholesale markets and increased their dependence on the ECB liquidity operations.17 Prior to the introduction of the vLTRO, Portuguese banks borrowed from the ECB in short maturities (on average 4 months). The vLTRO allowed banks to costlessly swap their existing short-term funding into a stable and predictable source of nancing of three-year maturity. Figure 1 Panel (b) shows that the average maturity of bank debt from the ECB increased from 4 months right before the vLTRO announcement to 32 months in the post- period. In total, the vLTRO provided EUR 20.2bn of long-term liquidity to Portuguese banks in December 2011 and

See Internet A.2 for further details.
Alves et al. (2016) nd that the banks did not freeze lending to the real economy as they eectively substituted their source of funding with the ECB liquidity.

ECB Working Paper Series No 2521 / February 2021

an additional EUR 26.8bn in February 2012.18 Portuguese banks were the second largest recipient of vLTRO relative to the size of banking sector.

1.3.2 Pledged securities

Figure 1 Panel (c) illustrates the main categories of collateral pledged with the ECB by Portuguese banks. While government bonds (red) were always a relevant source of collateral, bank-issued securities (blue) played an increasingly more important role as a collateral for liquidity operations over time. The majority of these bonds are (risky) domestic securities issued by Portuguese banks and government and are associated with high haircut subsidy.

Together with the vLTRO the ECB also increased collateral availability by (i) reducing the rating threshold for certain asset- backed securities (ABS) and (ii) allowing national central banks, as a temporary solution, to accept as collateral additional performing credit claims (i.e. bank loans) that satisfy specic eligibility criteria." (ECB press release 8 De- cember 2011). However, dierently from a number of other European countries, Portugal did not take advantage of the relaxation of collateral rules.

Regarding the use of asset-backed securities, Figure 1 Panel (c) illustrates that the share of securitized assets (light grey), pledged by Portuguese banks, did not increase during the period of the vLTRO.19 As for the additional credit claims, the Bank of Portugal introduced only a very limited set of changes to the existing collateral framework and only for the pledging of collateral in February 2012, i.e. the second allotment date. As a result, the share of non-marketable securities (dark grey), did not exceed 5% of the total collateral pledged and its use did not increase substantially during the vLTRO. Thus, our results are by and large not driven by the relaxation of collateral rules on securitized assets, or by modications

18The policy was administered in two operations on December 21, 2011 and February 29, 2012.

19We also do not observe any large movements in pledging of securitized assets in the cross-section of banks. Prior to the vLTRO, the average share of securitized assets to total pledged assets with the ECB was about 23.2% with a 24.8% standard deviation. Following the vLTRO, the importance of securitized assets has slightly decreased (18% average and 21.5% standard deviation). For details see Internet Appendix Figure H2.

ECB Working Paper Series No 2521 / February 2021

of the collateral framework at the national level.

1.3.3 Haircut subsidy

The ECB provides funding against adequate collateral and it applies haircuts, i.e. a reduction to the value of the collateral pledged by banks. The size of the haircuts varies depending on the riskiness and maturity of the underlying collateral. Prior to 2008, the haircuts applied by the ECB were similar to the private market haircuts on repo loans. However, after September 2008, the ECB started oering haircuts on risky securities that were below private market haircuts. We refer to the gap between the private market and central bank haircut on the value of risky securities, as haircut subsidy. Drechsler et al. (2016) highlight that the changes in the haircut policy essentially worked as a subsidy for distressed economies in the euro area.

Figure 1 Panel (d) shows the average private market and ECB haircut for securities pledged by Portuguese banks with the ECB.20 The haircut subsidy for those securities was stable and on average 5 pp between 2007 and 2010 but it increased dramatically during the European sovereign debt crisis and reached 70 pp in 2011. This change was triggered by a rating downgrade of Portuguese sovereign bonds and a subsequent increase in private market haircuts by about 75 pp. While private market repo clearing houses responded by increasing the haircut on the securities issued by peripheral economies, the ECB kept those haircuts at signicantly lower levels. In fact, ECB haircuts increased only slightly by about 5 pp.

In the extreme case of a complete reliance of banks on the private repo market, the reduction in the borrowing capacity of the Portuguese banking sector would have been reduced by around EUR 20 billions (57% of their equity capital) at the time of 2011 vLTRO. By extending the maturity of the ECB liquidity operations to three years, the vLTRO signicantly reduced uncertainty regarding the long-term availability of LOLR funding and the potential need to rely on private repo nancing at higher market haircuts. Thus, the reduction in

20Private repo market haircuts are obtained from LCH Clearnet. See Section 2.

ECB Working Paper Series No 2521 / February 2021

uncertainty was particularly strong for banks holding securities with large haircut subsidy.

Data

For the purpose of our analysis, we build a novel dataset that matches data from the Euro- pean Central Bank, private repo markets and the Bank of Portugal. Below we describe the data following a top-down approach:

LOLR data. We use the Eurosystem Collateral Database to extract information on the securities pledged with the ECB to obtain LOLR funding. We observe the following characteristics of the pledged assets at bank-security-month level: ISIN-code, nominal value, ECB haircut adjusted value, haircut category, quantity, issuance date, and maturity date.

We also use data from the private repo market - LCH Clearnet. For each security, we observe monthly series of private market haircuts. We match the ECB and private market data to construct the measure of the banks' haircut subsidy.

Finally, we exploit the ECB monetary policy and market operations database. This data source provides us with detailed information on all ECB liquidity operations split by categories (weekly main renancing operations (MRO); longer-term renancing operations with 1, 3, 6, 12 month maturity; 36-month operations (vLTRO)) for all banks on a daily basis. The database allows us to directly observe the exact amount of 36-month vLTRO funding used by each bank.

Bank-level data. We rely on several sources maintained by the Bank of Portugal. The Securities Statistics Integrated System (SIET) contains information on the pool of all marketable securities held by banks such as quantity, book value, and market value at the bank-security-month level. We use SIET to dene the two alternative measure of banks' exposure to the LOLR policy uncertainty - (i) holding of eligible securities and (ii) holding of eligible securities that match the maturity prole of the vLTRO operation.21

21We also use variables from the bank balance sheet and prudential monthly databases to construct controls

ECB Working Paper Series No 2521 / February 2021

Credit register. Central de Responsabilidades de Credito (CRC) provides monthly loan-level information on the universe of outstanding loans to Portuguese rms above the reporting threshold of EUR 50.22 CRC includes data on loan amounts and key loan characteristics (maturity, currency, type of the loan, and the guarantee/collateral used to secure the loan, if any). CRC allows us to observe both drawn and potential credit (unused credit lines, credit cards, etc.). The analysis uses all outstanding loans granted by banks to non- nancial rms residing in Portugal and borrowing in euro currency between June 2011 and June 2012. In the core part of the analysis, we focus on private non-nancial rms with multiple bank borrowing relationships. This accounts for almost 1.5 million (bank-rm-month observations) and 116,918 bank-rm relationships (see Internet Appendix Table H1).

Firm-level data. Firm-level annual census contains balance sheet and nancial reports as well as regional and sectoral classication of rms. We use this information to control for rm characteristics (total assets, employment, age, industry, and district) as a substitute to rm xed eects. In addition, we use rm-level investment for the analysis of the real outcomes. Finally, we also use employee-employer (Quadros de Pessoal) data matched with the credit registry to study the eects on employment at the rm-establishment level.

Empirical strategy

3.1 Measuring exposure to the reduction in LOLR uncertainty

Haircut subsidy. We use the size of the haircut subsidy to measure banks' exposure to LOLR policy uncertainty. We construct a measure of ECB haircut subsidy at the bank level as a dierence between the ECB and the private market valuation of all pledged securities

for observable bank characteristics such as size, equity ratio, capital ratio, liquidity ratio, loan-to-assets ratio, and equity-to-assets ratio. We restrict the analysis to domestic banks and domestic subsidiaries of foreign banks. This leaves us with a nal sample of 30 banks.

22We exploit the universal coverage of micro-level credit data of rms. This provides an extremely rich data for small and medium enterprises (SMEs) which tend to be underrepresented in other countries.

ECB Working Paper Series No 2521 / February 2021

normalized to bank's total assets:23
HaircutSubsidyi =	s (ECB valuations	private market valuations) Qi;s	:
P
	total assetsi

This measure captures bank's total haircut subsidy taking into account the dierence in the two valuations of its securities pledged with the ECB. For each security s (reported at ISIN- level), we retrieve the ocial ECB haircut-adjusted valuation and the private market (LCH- Clearnet) valuation. Qi;s represents the total before-haircut value of the security pledged by bank i with the ECB. To minimize endogeneity, we construct the exposure measure as of September 2011, three months prior to the policy announcement. This measure of exposure captures a hypothetical reduction in borrowing capacity in the extreme case of complete reliance on private repo market. Banks with a larger haircut subsidy are the ones that beneted more from the reduction in policy uncertainty.

Table 1 shows that the average haircut subsidy prior to the vLTRO in 2011 was 2.48% of total assets. Out of a total of 30 banks, 15 banks do not pledge securities with the ECB and we label these banks as control banks. In addition, we observe a large cross-sectional variation in the haircut subsidy at the bank level (standard deviation is 3.91%).24

In the baseline analysis we use the exposure measure as a continuous variable as it allows us to capture the fact that the higher the haircut subsidy the stronger the eects for a given bank. As a robustness, we compare lending outcomes using a dummy exposure by splitting banks into exposed (treated) and non-exposed (control). In Internet Appendix Table C1, we also compare averages of bank's observables between these two groups of banks.

We show that exposed banks are on average larger and hold more securities. The two

23Our baseline measure of haircut subsidy is constructed using only securities pledged with the ECB. In the Internet Appendix Table B2, we consider an alternative measure of haircut subsidy based on all eligible securities held by each bank (both pledged and not pledged with the ECB). Our results remain robust to this alternative specication.

24In Internet Appendix Figure H1 we also illustrate the cross-sectional variation of the haircut subsidy at the bank level as of September 2011.

ECB Working Paper Series No 2521 / February 2021

groups do not dier across other dimensions such as cash holdings, capitalization, protabil- ity or leverage.25

Alternative measures of exposure. In addition to the haircut subsidy, we also construct three other measures of banks' exposure to the reduction in LOLR policy uncertainty. First, we measure exposure as the sum of all short-term ECB funding taken up by a bank as of September 2011 and normalized to total assets. This measure captures the fact that banks primarily swapped their existing short-term ECB funding into the newly oered long-term funding. Second, we construct a measure of exposure that captures total bank borrowing capacity with the ECB as the value of total banks' security holdings eligible as collateral with the ECB (scaled to total assets). Third, we consider a more rened measure of the latter exposure by focusing only on bank holdings of eligible securities that mature in an horizon between one and three years. By lengthening the maturity of repo operations to three years, the ECB implicitly decreased rollover risk for the funding backed by securities that matured shortly before the vLTRO expiration. In other words, banks did not need to be concerned about the price volatility of these securities at the time of the vLTRO repayment and as a result, they would not need to face re-sale risk due to rollover issues.26 Table 1 contains summary statistics also for these alternative measures of exposure.

3.2 Empirical specication

We use the dierence-in-dierences (DID) framework to compare lending before and after the policy intervention by exploiting the variation in the cross-section of banks' exposure to the LOLR policy uncertainty. We examine the time series evolution of credit at the

25The dierence in security holdings conrms the fact that banks must hold securities to be able to benet from the haircut subsidy. Instead, the dierence in bank's size is directly related to the xed cost of establishing an infrastructure to borrow from the ECB (for example a trading desk). Smaller banks may not nd it benecial to bear this xed cost. For further discussion of dierences in observables see Internet C.

26For more details regarding the construction of these measures see Internet B.

ECB Working Paper Series No 2521 / February 2021

bank-rm-month level following the baseline specication:

log(crediti;j;t) = jt + ij + (HaircutSubsidyi P ostt) + i;j;t;

(1)

where log(crediti;j;t) is log amount of all credit that rm j obtains from bank i at month t. In the main analysis we focus on drawn credit. HaircutSubsidyi denotes our baseline exposure to LOLR policy uncertainty - i.e. the size of haircut subsidy for a bank i. We analyze a 13-month period: June 2011{June 2012. P ostt is a dummy variable that takes the value of one in the post-period (February{June 2012), and zero otherwise. We end our baseline sample period in June 2012 to avoid the overlap with the announcement of the Outright Monetary Operations (i.e., the whatever it takes" speech of the president of the ECB Mario Draghi) in July 2012.

We saturate our model with xed eects to address some of the main empirical challenges. First, a potential bias in estimating the causal eects of the reduction in central bank policy uncertainty can stem from the interaction between credit demand and supply. In line with Khwaja and Mian (2008), we incorporate rm-time xed eects to absorb time-varying rm- specic changes in credit demand and isolate the causal eect of the policy by comparing lending outcomes of the same rm (j) borrowing in the same month (t) from at least two dierently exposed banks.

Second, the bank-rm matching is not random as banks choose their borrowers (and vice-versa). Firm borrowing relationships can also be of a dierent quality across dierent banks (i.e., due to dierent lengths of the relationship, existence and quality of collateral). We address the potential bias related to the non-exogenousbank-rm matching by including bank-rm xed eects that absorb any time-invariantbank-rm variation. Additionally, bank-rm xed eects nest inside bank xed eects and absorb any observable and unobservable time-invariant bank characteristics that could be potentially correlated with our exposure measure. To summarize, our empirical specication relies on a combination of

ECB Working Paper Series No 2521 / February 2021

rm-time and bank-rm xed eects (equation (1)). The main results are presented for non- nancial rms and non-prots which we denote as Private NFCs. We also report results for a larger sample of rms that includes self-employed entrepreneurs and public companies denoted as All rms.

We examine the existence of parallel trends by comparing lending dynamics of exposed and non-exposed banks in the period leading up to the policy intervention using a dynamic dierence-in-dierences specication:

	k6X
log(crediti;j;t) = jt + ij +	k(1i=exposed 1t=k) + i;j;t;	(2)
	=2011m9

where 1i=exposed is one if banks were exposed to the reduction in LOLR policy uncertainty (i.e. they beneted from a haircut subsidy) and zero otherwise. 1t=k is an indicator that equals one in month t, and zero otherwise. To test for the absence of the pre-trend, estimates of k need to be statistically insignicant from zero until the policy announcement.

Loan-levelresults

4.1 Intensive margin: loan quantities

Table 2 presents the main result on the intensive margin using the specication in equation (1). Column (1) shows the results using the full sample of loans (bank-rm pairs). Here, we use bank xed eects (to absorb any time-invariant bank characteristics) and time xed eects. From Column (2) onward, we restrict the loan sample to rms that borrow from at least two banks at each month. We denote this as multiple bank relationships". In Column

(3) we replace the time xed eects with rm-time xed eects to absorb any variation from rm-level (demand) changes in line with Khwaja and Mian (2008). It is plausible that a rm has a dierent relationship with dierent banks. To address this challenge, in Column

ECB Working Paper Series No 2521 / February 2021

we introduce a set of bank-rm control variables. In our preferred specication shown in Column (5) we also saturate the model with bank-rm xed eects to address any potential threat coming from a non-exogenous matching between banks and rms.
The coecient estimate of for all specications is positive and statistically signicant suggesting that the reduction in LOLR policy uncertainty triggered by the vLTRO had a positive impact on bank lending to rms.27 As aggregate credit to rms in Portugal decreased during the period, we interpret the positive coecient as a smaller contraction in lending by more exposed banks. In terms of economic signicance, the coecient estimate of 0.824 (Column (5) of Table 2) implies that a one standard deviation increase in bank exposure to the reduction in LOLR policy uncertainty (3.91 percent of total assets from Table 1) is associated with a 3.22 percent increase in lending on the intensive margin.
Our results suggest that a reduction of LOLR policy uncertainty had a positive eect on lending to rms. In particular, more exposed banks (i.e., those beneting from a larger haircut subsidy) internalized the reduction in policy uncertainty and consequently reduced their lending by less. Our empirical evidence is consistent with the real option channel of policy uncertainty (see e.g. Bernanke, 1983). The option value of delaying banks' illiquid investments is high when uncertainty about the future availability of LOLR funding is high. Due to the lack of commitment by the central bank to provide unlimited liquidity for an extended period of time, banks preferred to wait and be more cautious with their lending. The introduction of the vLTRO reduced this uncertainty, decreased banks' option to wait and as a result incentivized them to no longer postpone their lending.
Comparing the estimates with and without rm-time xed eects (Table 2 Columns (2) and (3)), we nd that not controlling for the overall rm's credit demand overestimates the eect of the policy action on lending. The estimate decreases in magnitude when we introduce the rm-time xed eects but it remains stable, positive and statistically signicant. This

27The Within R-squared" reported in Table 2 suggests that the reduction in LOLR policy uncertainty entailed in the ECB's vLTRO explains around 10% of the within bank-rm credit variation.

ECB Working Paper Series No 2521 / February 2021

suggests that changes in uncertainty simultaneously a ected rm demand and bank supply of credit. Thus, it is important to control for time-varying credit demand to disentangle the two forces and estimate the causal impact of uncertainty through the bank-lending channel.28 Alternative exposure measures. Internet Appendix Table B1 presents the intensive margin results using the other three alternative measures of exposure discussed in Section

3.1. The results are robust to the use of these alternative measures.

Robustness. We conduct a battery of tests to support our identi cation. Internet Appendix Table D1 shows the results of the baseline speci cation in equation (1) rewritten as a collapsed di erence-in-di erences (comparing average bank- rm credit in the pre- and post- periods) to derive estimates with more conservative standard errors.

Internet Appendix Table E1 presents a number of additional results. First, we show robustness to changes in the credit and rm de nitions. Second, our results are unchanged if we only focus on the variation in the cross-section of exposed banks and are also robust when controlling for bank characteristics interacted with the POST dummy. Third, around the time of the vLTRO announcement, four banks were undergoing the stress tests conducted by the European Banking Authority.29 Our results remain unchanged even when we exclude these banks from the sample. Finally, to corroborate the importance of our ex-ante measure of exposure in capturing the e ects of the reduction in LOLR policy uncertainty, we document that estimates using the endogenous vLTRO uptake deliver results that are not signi cantly di erent from zero. This is because, di erently from the haircut subsidy, the endogenous vLTRO uptake is not an ex-ante measure of exposure to the reduction in policy uncertainty. Thus, it could reects a variety of reasons for banks' decisions regarding the newly available long-term liquidity, that are unrelated to the reduction in LOLR policy uncertainty.

28We report estimates using two-way clustered standard errors at bank-time and rm level, as it allows us to address the threat that rm-shocks can be serially correlated and also bank-time shocks (our source of variation) can be correlated across rms. Our results are robust to alternative clustering levels and we report these estimates in the Internet Appendix.

29A number of papers study the eects of the 2011 EBA shock on banks' balance sheets and the real economy (e.g. Blattner et al., 2019; DeGryse et al., 2019; Gropp et al., 2019).

ECB Working Paper Series No 2521 / February 2021

Dynamic setup. Equation (1) provides a consistent estimate of under the identifying assumption of parallel trends. In equation (2), we modify the empirical framework into a dynamic setup which allows us to examine the existence of pre-trends. Figure 2 presents the results. Consistent with the parallel trends assumption, the gure clearly displays no relation between haircut subsidy and lending dynamics until November 2011. After the policy announcement (December 2011), we observe statistically signi cant di erences in lending behavior between exposed (treated) and unexposed (control) banks. The positive di erence-in-di erences coecients need to be interpreted in the context of the ongoing sovereign debt crisis and nancial deleveraging of the Portuguese banking sector. While the credit contraction continued for non-exposed banks, banks exposed to the reduction in LOLR policy uncertainty signi cantly slowed down the pace of the deleveraging.

Credit lines. One hypothesis that would lend support to the prompt lending response presented in Figure 2 is that exposed banks allowed rms to draw down on their existing credit lines.30 To test this hypothesis, we extend the speci cation in equation1 with a triple interaction term where CreditLinej takes the value of 1 if a rm had pre-approved potential credit prior to the vLTRO policy announcement and 0 otherwise. Table 3 Panel (a) Column

shows a stronger response to the reduction in LOLR policy uncertainty for rms with prior access to credit lines. Furthermore, Column (2) documents a decrease in potential (pre-approved but unused) credit which is consistent with higher draw-downs. This is also reected in higher utilization rates of credit reported in Column (3).31

Placebo test. Were banks more exposed to the 2011 vLTRO generally more prone to react to any bank-speci c liquidity shocks? Iyer et al. (2014) show that that banks that relied

In the credit registry data, we observe regular (drawn) credit as well as potential (pre-approved but unused) credit. As a rm draws from its credit line, the amount disappears from the potential credit category and appears in the drawn category. This data structure does not allow us to directly examine the amount and change in the credit line limits but it provides a reliable picture about the total utilization rates, amount of unused credit and overall credit dynamics.
To avoid any confounding eect coming from the fact that as a reaction to the vLTRO banks could also increase the limits on credit lines, we compute the utilization rates as a share of total credit drawn in month t to total available credit in September 2011.

ECB Working Paper Series No 2521 / February 2021

more on the interbank borrowing decreased their credit supply by more, following the sudden freeze of the European interbank market in August 2007. Thus, we use the 2007 liquidity freeze as a placebo sample to investigate whether the banks more exposed to the reduction in LOLR policy uncertainty in 2011 were also more sensitive to the 2007 liquidity dry-up. We follow the dynamic setup specication from equation (2) and replace the left-hand-side lending outcomes in 2011{2012 with the lending outcomes in 2007. Internet Appendix Figure F1 shows no evidence that the banks more exposed to the 2011 reduction in LOLR policy uncertainty are generally more sensitive to liquidity shocks.

4.2 Intensive margin: loan maturities

According to the real option channel, in response to a decrease in LOLR policy uncertainty, banks should invest more in illiquid and irreversible investments. In the previous sections, we have already shown that with the introduction of the vLTRO, more exposed banks extended larger quantities of loans. In this section, we test whether the reduction in LOLR uncertainty also made banks choose to grant loans of longer maturity, i.e. more irreversible.

In the credit registry data, we observe the loan maturities reported using maturity baskets (1{90 days , 90{180 days etc.). We construct a continuous measure of loan maturity in two steps. First, we approximate the maturity of the loan as a mid-point of a basket interval. Second, as a rm may have multiple loans outstanding with the same bank, we compute the weighted average of the loan maturity at the bank-rm-time level using loan sizes as weights.

We estimate the eect of the reduction on LOLR policy uncertainty on loan maturities by modifying the equation (1) in a following way:

Loan maturityi;j;t = jt + ij + (HaircutSubsidyi P ostt) + Qi;j;t + i;j;t:

(3)

where Qi;j;t denotes bank-rm and loan controls. Table 3 Panel (b) displays a positive and statistically signicant coecient suggesting that the reduction in LOLR policy uncertainty

ECB Working Paper Series No 2521 / February 2021

had a positive eect on the maturity of banks' loans to rms. Consistently with the real option channel, banks more exposed to LOLR uncertainty were more likely to oer longer maturities on their loans in the post period. Thus, as the uncertainty about availability of the future LOLR funding resolves, exposed banks invest more into long-term and irreversible projects.

4.3 vLTRO and LOLR policy uncertainty

In this section, we provide evidence in support of our argument that the vLTRO was associated with a sudden reduction in LOLR policy uncertainty. The vLTRO allowed banks to access LOLR funding of longer maturity in times of a sizable gap between the ECB and private market security haircuts. We argue that the presence of a high haircut subsidy is a necessary condition for changes in the maturity of LOLR funding to reduce uncertainty for banks and have real eects. In the absence of a high haircut subsidy, uncertainty about the future availability of LOLR liquidity is not expected to have a large eect on the borrowing capacity of banks and thus on their lending behavior. This is because borrowing on the private repo markets is not substantially more costly than borrowing from the central bank. Therefore, we argue that the size of the haircut subsidy is a valid measure of bank exposure to LOLR policy uncertainty.

To test this argument, we study the bank-lending response to a long-term liquidity operation (LTRO) introduced in 2009, in a period of low exposure to the LOLR policy uncer- tainty. Similarly to the vLTRO, the 2009 LTRO oered banks to borrow from the ECB at a long-term(one-year) maturity at the same conditions as if the banks continued to borrow short-term. In 2009, however, the ECB haircuts were closely following the private market haircuts (Figure 1 Panel (d)). The average haircut subsidy of Portuguese banks in 2009 was 0.01% of total assets. This is of a negligible magnitude compared to the average haircut subsidy of 2.48% in 2011. As a result, even in the case of a sudden stop of the LOLR funding

ECB Working Paper Series No 2521 / February 2021

in 2009, banks could switch to borrow from the private market without suering a large drop in their borrowing capacity. Thus, in 2009 banks were substantially less exposed to LOLR policy uncertainty.

We examine the eects of the policy change on bank lending over the period January 2009{Apr 2010. Table 4 summarizes the results. Column (1) shows that the lending eect of 2009 LTRO is not statistically signicant from zero.32 The lengthening of the maturity of the LOLR funding does not have an eect on bank lending in the 2009 period of low haircut subsidy. This suggests that for a lengthening of the maturity of central bank liquidity to stimulate bank lending, it needs to be associated with a large exposure of banks to LOLR policy uncertainty. Thus, a high haircut subsidy is a necessary condition for the policy change to have real eects.

4.4 Intensive margin and rm characteristics

Is the reduction in the LOLR policy uncertainty transmitted equally across rms? To address this question, we exploit the matching of the credit registry data with the rm census data and introduce rm heterogeneity in the baseline specication. Table 5 displays the estimates of the heterogeneous impact of LOLR uncertainty using a triple interaction specication:

log(crediti;j;t) = jt + ij + 1(HaircutSubsidyi P ostt)

+ 2(HaircutSubsidyi P ostt Fj) + i;j;t; (4)

where Fj denotes rm characteristics such as size, length of the bank-rm relationship and ex- ante rm riskiness. Column (1) of Table 5 shows a stronger positive eect of the reduction in LOLR policy uncertainty on small rms. This nding is consistent with the existing empirical evidence that emphasizes that small rms are more aected by shocks propagated

32For comparison with our baseline, we report our main result in Column (2) and show all estimates re-scaled using the respective standard deviations.

ECB Working Paper Series No 2521 / February 2021

through the bank lending channel (Khwaja and Mian, 2008; Iyer et al., 2014).33

Column (2) of Table 5 illustrates a stronger positive eect for rms with a shorter (less than two-year) relationship with a bank. While the stronger results for rms with shorter lending relations mitigates concerns of ever-greening, it does not exclude the possibility that some of the credit went to bad investment projects. Table 5 Column (3) shows stronger lending eects for risky rms. We denote the rm as risky (one) if its z-score is above the median and zero otherwise.34 Section 4.6 further explores bank risk-taking for new credit.

4.5 Extensive margin

Did the reduction in LOLR policy uncertainty entailed in the vLTRO aect banks' decision to terminate fewer loans? Is there any evidence that more exposed banks started to establish new lending relationships with previously unconnected rms? To address these questions, we study the eects of the policy action on the extensive margin.

Loan termination. To study the impact on loan terminations, we consider a collapsed version of the dierence-in-dierences framework where we compare bank-rm pairs in the pre-period (2011m6{2011m10) and post-period (2012m2{2012m6):

EXITi;j = j + HaircutSubsidyi + Bi + i;j:

(5)

The sample includes all loans that were outstanding in the period prior to the vLTRO (2011m6{2011m10). EXIT dummy equals one if the loan only appears in the pre-period and

This result is in line with the idea that banks bene ting from the ECB policy change may nd it more pro table to loosen their credit standards and extend more credit to smaller borrowers, which are generally riskier and pay higher interest rate on their loan. This is consistent with our result on larger credit ow towards riskier rm and more broadly with the evidence on the risk-taking channel of monetary policy (e.g. Jimenez et al., 2014). In addition, with the intent to rebuild lending relationships, exposed banks may extend more credit to rms to which they had previously cut it by more. This rationale is consistent with the nding of Iyer et al. (2014) who show that Portuguese banks experiencing an interbank liquidity crunch at the onset of the global nancial crisis cut lending by more towards small rms compared to large rms.
We utilize the z-score measure for Portuguese rms by Antunes et al. (2016).

ECB Working Paper Series No 2521 / February 2021

it does not exist in the post-period.35

Column (1) of Table 6 uses a combination of bank and rm controls. Column (2) replaces rm controls with rm xed eects to investigate whether the loan exit results continue to hold after we control for rm observable and unobservable characteristics. Both results suggest that a 1 standard deviation increase in the exposure to the vLTRO (3.91 percent) decreases the exit rate by 8.76 percent (3.91(-2.24)). This conrms the hypothesis that in a period of crisis, banks more exposed to a reduction in LOLR policy uncertainty are less prone to terminate relationships with rms which overall slowed down the pace of deleveraging.

New credit approvals. Were more exposed banks also more likely to start lending to new clients? We address this question by augmenting our dataset with the credit consultation data. Banks obtain records of new rms, by accessing the consultation database upon the rm's consent. We analyze all loan consultations after the policy announcement and match them with the actual entries in the credit registry. We construct a dummy variable ENTRYi;j which equals one if a bank-rm consultation entry is matched with a new bank-rm record in the credit registry, and zero otherwise.36

In the main analysis, we focus on consultations made between December 2011 and April 2012 and we match them with the credit registry outcomes in the period December 2011{ June 2012. Roughly 10% of loan consultations are successful and appear in the credit registry as new loans.37 We estimate the eects on the extensive margin using the following specication:

ENTRYi;j = j + HaircutSubsidyi + Bi + i;j:

(6)

Columns (3) and (4) of Table 6 show that the coecient is very stable with or without rm FE. The results suggest that a 1 standard deviation increase in the exposure to the

35We only consider a sample of loans for which the maturity would not naturally end in the post-period.36Our construction of the extensive margin is consistent with Jimenez et al. (2012, 2014) who study loan

approvals in Spain. Spanish credit registry (CIR) has a similar data structure to the Portuguese CRC.

37If approved, the majority of the loans are granted within two months. We also perform robustness tests for changes in the consultation window and the results are not aected.

ECB Working Paper Series No 2521 / February 2021

reduction in LOLR uncertainty increases the probability of a new relationship by 4.65 percent (3.911.19).38

4.6 Extensive margin and rm riskiness

New credit and ex-ante rm riskiness. How does new lending relate to bank risk- taking? Did risky rms benet more from the reduction in LOLR policy uncertainty? To this end, we introduce triple interactions that capture dierential outcomes for ex-ante risky rms:

ENTRYi;j = j + 1HaircutSubsidyi + 2(HaircutSubsidyi RiskyF irmj) + Bi + i;j: (7)

Columns (5) and (6) of Table 6 show that exposed banks are more likely to establish new relationship with riskier rms. This is true whether we proxy for rm riskiness with z-scores

or with rms recent loan delinquencies, measured as loan delinquencies in the past year.

Do these new relationships default more in the following years? To answer this

question, we compare loan defaults in a full dynamic framework:

	k6X
loanDefaulti;j;t+2Y = i+ind;t+reg;i+	k(HaircutSubsidyi 1t=k)+Qi;j+i;j;t: (8)
	=2011q4

For identication, we exploit the cross-sectional variation in banks' haircut subsidy and we compare defaults of observationally equivalent loans issued in dierent quarters. The latter are dened as loans of the same outstanding amount and purpose issued by the same bank to rms of the same size, region and industry. In the baseline, we report loan defaults within two years after the loan origination.

Figure 3 displays the dierences in loan default rates between more and less exposed

38The main results are based on the linear probability models widely used in the literature (see Khwaja and Mian, 2008; Jimenez et al., 2012). Logit and probit specications provide very similar results. Results available upon request.

ECB Working Paper Series No 2521 / February 2021

banks. In 2011, there is no relation between exposure and default rates, consistent with parallel trends. In the two quarters following the policy (Q1{Q2 2012), more exposed banks are more likely to grant new credit to rms that default on these loans more within the next two years. This dierence dies out already in mid-2012, suggesting that more exposed banks loosened their lending standards only temporarily after the policy action. This nding suggests that the risk-taking channel previously documented for monetary policy (e.g. Jimenez et al., 2014; Heider et al., 2019) is also present in the case of the LOLR policy.

Firm-levelresults

5.1 Credit

In what follows, we explore whether the loan-level results presented above translate into economically relevant rm-level credit outcomes. A valid concern is that as rms received more funding from more exposed banks, their borrowing from other less exposed banks could have been reduced by the same proportion (i.e. a zero-sum outcome). This would imply insignicant rm-level credit eects.

Firm-level credit. In order to measure the net rm-level credit eects, we collapse credit registry loan-level data to the rm level and estimate the following equation:

log(yj) = + HaircutSubsidyj + Bj + Qj + Fj + j:

(9)

The outcome variable is the change in the log value of total credit received by a rm j from all the banks it borrowed between the pre- and post- period. For each rm, we compute an indirect exposure to the reduction in LOLR uncertainty. The rm's exposure measure is given by the weighted average of the haircuts subsidy of each bank (to which the rm is connected). The weights are based on the rm' credit with each individual bank in the pre-

ECB Working Paper Series No 2521 / February 2021

period:

		P	i(HaircutSubsidyi crediti;j;t=pre)
HaircutSubsidy =	:
j	i(crediti;j;t=pre)
			P

We repeat the same procedure for computing indirect measures of bank and bank-rm control variables (denoted as Bj and Qj). Finally, equation (9) controls for rm size and industry- district xed eects. The rm-level estimates are consistent with the results reported on the loan level (see Column (1) of Table 7). Firms borrowing more from exposed banks experienced a less sizable contraction in credit than rms more connected to non-exposed banks. In addition, the reduction in uncertainty had strong credit supply eects on the loan level that translated into positive net eects on rm-level credit.39

Are the aggregate credit eects economically relevant? The reduction in LOLR uncertainty reduced the pace of the lending contraction at the aggregate rm-level during the period of the sovereign debt crisis in Portugal. Within the partial equilibrium setting, we plug the bias-corrected credit estimates into equation (9) and compare the predicted aggregate rm-level credit with the policy ( = 0:644) and without it ( = 0). The reduction in LOLR policy uncertainty triggered by the vLTRO contributed to approximately EUR 892 mil in lending to rms in Portugal. By comparing the credit dynamics to a counterfactual world without any policy intervention, we can conclude that although the policy did not stop the ongoing credit contraction, it signicantly reduced its pace. We estimate that without the policy, the credit would have contracted by additional 2.15 percentage points. This means that while the observed credit contraction in the period after the vLTRO was -5.75%, in the absence of the policy it would have been -7.90%.

39Due to the absence of the rm FE, equation (9) does not absorb any rm-specic shocks which can contribute to the bias in estimate. To address this issue, we compute the bias-corrected coecient as in Jimenez et al. (2019). The corrections cause the coecient to drop by 35% but the nal estimate remains positive (0.644), suggesting that the eect of the policy change remains economically relevant also at the rm level.

ECB Working Paper Series No 2521 / February 2021

5.2 Real eects: investment and employment

Are the lending eects to rms substantial enough to impact the real economy? We use the specication in equation (9) to study the capital investment and employment eects.

Investment. We measure investmentj as the annual log change of investment using the rm census data. Column (2) of Table7 shows that the average eect of rms' indirect exposure to reduction in LOLR uncertainty (HaicutSubsidyj) is positive and statistically signicant. Further, Column (3) highlights that the eect on investment is stronger for small rms.

The impact of this policy on investment is economically sizable. While the observed investment between 2011 and 2012 fell by 18.5%, our estimates suggest that without the policy, rms' investment would have contracted by additional 2.2 percentage points.

Employment. Following the literature on employment dynamics (Davis and Halti- wanger, 1999; Chodorow-Reich,2014), we aggregate the establishment-level data from the employee-employer database to the rm level and create a measure of symmetric growth rate in employment between 2011 and 2012:

employment =	(employment2012;j	employment2011;j)	:
	(employment
j	1	+ employment	2011;j	)
	2	2012;j

Column (4) of Table 7 shows that the eect on the employment of an average rm is statistically insignicant. As we however dierentiate rms by size in Column (5), we nd that the policy had positive eect on the employment in small rms.

In the data, we observe that the year-on-year aggregate employment contacted by 9.7%. Our estimates show that in the absence of the reduction in uncertainty, rms would have reduced their employment by additional 2 percentage points. The real eects on employment are entirely driven by the impact of the policy change on SMEs. Thus, thanks to the low reporting thresholds on loans in the CRC and the granularity of the Portuguese data, we are

ECB Working Paper Series No 2521 / February 2021

able to uncover the real eects of a reduction in central bank policy uncertainty in terms of

aggregate employment.40

Conclusion

This paper provides new empirical evidence in support of the lending and real eects of the sudden reduction in uncertainty regarding central bank liquidity policy. We exploit the ECB's vLTRO as a quasi-natural experiment of a sudden decrease in LOLR policy uncertainty and a novel granular dataset that perfectly matches the ECB monetary policy and market operations data, private repo market haircuts data, rm credit registry and banks' security holdings in Portugal. We nd that banks more exposed to the reduction in LOLR policy uncertainty deleveraged at a slower pace. The reduction in policy uncertainty had a positive and economically sizable impact not only on lending but also on real outcomes.

Our results have several interesting policy implications. First, we show that policy aimed at reducing policy uncertainty in times of crisis can be eective in reviving credit and real economy. Second, we highlight the importance of central bank commitment and long-horizon guidance concerning the future course of its policy actions in the context of the LOLR policy. Third, we provide new insights into the transmission channel of LOLR policy to bank lending and real economic outcomes.

authoryear

40Our results are subject to two caveats. First, our estimates are based on an aggregation of the eects computed at the rm level in the partial equilibrium setup. Consequently, they represent a lower bound on the actual eects. Second, the results may also underestimate the true eects of the reduction in LOLR policy uncertainty as we only consider a period of 1 year after the vLTRO. We used this limited time span due to other policy actions by the ECB post 2012. While it is plausible that the real eects might take more time to materialize, it would be challenging to compute long-term eects that abstract from other confounding factors.

ECB Working Paper Series No 2521 / February 2021

References

Altavilla, Carlo, Miguel Boucinha, Jose-Luis Peydro, and Frank Smets, 2018, Banking su- pervision, monetary policy and risk-taking: Big data evidence from 15 credit registers, ECB Working paper .

Altavilla, Carlo, Lorenzo Burlon, Mariassunta Giannetti, and Sarah Holton, 2019, Is there a zero lower bound? The eects of negative policy rates on banks and rms, ECB Working paper .

Alves, Nuno, Diana Bonm, and Carla Soares, 2016, Surviving the perfect storm: The role of the lender of last resort, Working Paper .

Amiti, Mary, and David E. Weinstein, 2018, How much do idiosyncratic bank shocks aect investment? Evidence from matched bank-rm loan data, Journal of Political Economy 126, 525{587.

Andrade, Philippe, Christophe Cahn, Henri Fraisse, and Jean-Stephane Mesonnier, 2018, Can the provision of long-term liquidity help to avoid a credit crunch? Evidence from the Eurosystem's LTRO, Journal of the European Economic Association .

Antunes, Antonio, Homero Goncalves, and Pedro Prego, 2016, Firm default probabilities revisited .

Bagehot, Walter, 1873, Lombard Street: A Description of the Money Market (McMaster University Archive for the History of Economic Thought).

Baker, Scott R, Nicholas Bloom, and Steven J Davis, 2016, Measuring economic policy uncertainty, The Quarterly Journal of Economics 131, 1593{1636.

Berger, Allen N., Lamont K. Black, Christa H.S. Bouwman, and Jennifer Dlugosz, 2017, Bank loan supply responses to federal reserve emergency liquidity facilities, Journal of Financial Intermediation .

Bernanke, Ben, 1983, Irreversibility, uncertainty, and cyclical investment, Quarterly Journal of Economics 98, 85{106.

Bernanke, Ben, and Alan Blinder, 1992, The federal funds rate and the channels of monetary transmission, American Economic Review 82, 901{21.

Blattner, Laura, Luisa Farinha, and Francisca Rebelo, 2019, When losses turn into loans: The cost of undercapitalized banks, Working Paper .

ECB Working Paper Series No 2521 / February 2021

Bottero, Margherita, Camelia Minoiu, Jose-LuisPeydro, Andrea Polo, Andrea F. Presbitero, and Enrico Sette, 2019, Negative monetary policy rates and portfolio rebalancing: Evidence from credit register data, IMF Working Paper No. 19/44 .

Cahn, Christophe, Anne Duquerroy, and William Mullins, 2018, Unconventional monetary policy and bank lending relationships, Working Paper .

Caldara, Dario, Matteo Iacoviello, Patrik Molligo, Andrea Prestipino, and Andrea Ra o, 2020, The economic e ects of trade policy uncertainty, Journal of Monetary Economics 109, 38{59.

Campbell, Je rey R., Charles L. Evans, Jonas D.M. Fisher, and Alejandro Justiniano, 2012, Macroeconomic e ects of federal reserve forward guidance, Brookings Papers on Economic Activity Spring, 1{80.

Carpinelli, Luisa, and Matteo Crosignani, 2018, The eect of central bank liquidity injections on bank credit supply, Working Paper .

Cassola, Nuno, Ali Hortacsu, and Jakub Kastl, 2013, The 2007 subprime market crisis through the lens of European Central Bank auctions for short-term funds, Econometrica 81, 1309{1345.

Chakraborty, Indraneel, Itay Goldstein, and Andrew MacKinlay, 2018, Housing price booms and crowding-outeects in bank lending, The Review of Financial Studies 31, 2806{2853.

Chodorow-Reich,Gabriel, 2014, The employment eects of credit market disruptions: Firm- level evidence from the 2008-09 nancial crisis, Quarterly Journal of Economics 129, 1{59.

Davis, Steven J., and John Haltiwanger, 1999, Gross job ows, Handbook of Labor Economics 3, 2711{2805.

DeGryse, Hans, Artashes Karapetyan, and Sudipto Karmakar, 2019, To ask or not to ask: bank capital requirements and loan collateralization, Working Paper .

Drechsler, Itamar, Thomas Drechsel, David Marquez-Ibanez, and Philipp Schnabl, 2016, Who borrows from the lender of last resort?, The Journal of Finance 71, 1933{1974.

Fernandez-Villaverde, Jesus, Pablo Guerron -Quintana, Keith Kuester, and Juan Rubio- Ramrez, 2015, Fiscal volatility shocks and economic activity, American Economic Review 105, 3352{84.

Friedman, Milton, and Anna Schwartz, 1963, A Monetary History of the United States, 1867{1960 (National Bureau of Economic Research, Inc).

ECB Working Paper Series No 2521 / February 2021

Gropp, Reint, Thomas Mosk, Steven Ongena, and Carlo Wix, 2019, Bank response to higher capital requirements: Evidence from a quasi-natural experiment, The Review of Financial Studies 32.

Hauser, Andrew, 2014, Lender of last resort operations during the nancial crisis: seven practical lessons from the United Kingdom, BIS papers 69, 81{92.

Heider, Florian, Farzad Saidi, and Glenn Schepens, 2019, Life below Zero: Bank Lending under Negative Policy Rates, The Review of Financial Studies .

Huber, Kilian, 2018, Disentangling the eects of a banking crisis: Evidence from German rms and counties, American Economic Review 108, 868{98.

Husted, Lucas, John Rogers, and Bo Sun, 2019, Monetary policy uncertainty, Journal of Monetary Economics .

Iyer, Rajkamal, Jose-Luis Peydro, Samuel da Rocha-Lopes, and Antoinette Schoar, 2014, Interbank liquidity crunch and the rm credit crunch: Evidence from the 2007{2009 crisis, The Review of Financial Studies .

Jens, Candace E, 2017, Political uncertainty and investment: Causal evidence from US gubernatorial elections, Journal of Financial Economics 124, 563{579.

Jimenez, Gabriel, Steven Ongena, Jose-Luis Peydro, and Jesus Saurina, 2012, Credit supply and monetary policy: Identifying the bank balance-sheet channel with loan applications, American Economic Review 102, 2301{26.

Jimenez, Gabriel, Steven Ongena, Jose-Luis Peydro, and Jesus Saurina, 2014, Hazardous times for monetary policy: What do twenty-three million bank loans say about the eects of monetary policy on credit risk-taking?, Econometrica 82, 463{505.

Jimenez, Gabriel, Atif Mian, Jose-Luis Peydro, and Jesus Saurina, 2019, The real eects of the bank lending channel, Journal of Monetary Economics .

Kashyap, Anil, and Jeremy C. Stein, 2000, What do a million observations on banks say about the transmission of monetary policy?, American Economic Review 90, 407{428.

Kelly, Bryan, Lubos Pastor, and Pietro Veronesi, 2016, The price of political uncertainty: Theory and evidence from the option market, The Journal of Finance 71, 2417{2480.

Khwaja, Asim Ijaz, and Atif Mian, 2008, Tracing the impact of bank liquidity shocks: Evidence from an emerging market, American Economic Review 98, 1413{42.

ECB Working Paper Series No 2521 / February 2021

Krishnamurthy, Arvind, Stefan Nagel, and Annette Vissing-Jorgensen, 2017, ECB policies involving government bond purchases: impact and channels, Review of Finance 22, 1{44.

Luck, Stephan, and Tom Zimmermann, 2020, Employment eects of unconventional monetary policy: Evidence from QE, Journal of Financial Economics 135, 678{703.

Pastor, Lubos, and Pietro Veronesi, 2012, Uncertainty about government policy and stock prices, The Journal of Finance 67, 1219{1264.

Rodnyansky, Alexander, and Olivier Darmouni, 2017, The eects of quantitative easing on bank lending behavior, The Review of Financial Studies 30, 3858{3887.

Rodrik, Dani, 1991, Policy uncertainty and private investment in developing countries, Journal of Development Economics 36, 229{242.

Swanson, Eric T., 2017, Measuring the eects of federal reserve forward guidance and asset purchases on nancial markets, NBER Working Paper .

van Bekkum, Sjoerd, Marc Gabarro, and Rustom M. Irani, 2018, Does a larger menu increase appetite? Collateral eligibility and bank risk-taking, The Review of Financial Studies 31, 943{979.

Wright, Jonathan, 2012, What does monetary policy do to long-term interest rates at the zero lower bound?, Economic Journal 122, 447{466.

ECB Working Paper Series No 2521 / February 2021

Figure 1: ECB liquidity provisions

(a) Liquidity provisions (amount)

(b) Liquidity provisions (maturity)

Amount(billion EUR) 20 40

Average maturity (months) 10 20 30

01jan2007

01jul2009

01jan2012

01jul2014

01jan2017

01jan2008

01jan2010

01jan2012

01jan2014

01jan2016

other

vLTRO

Average maturity

(d) Security haircuts

100						80

Amount (billion EUR) 50					Average Haircut (%)	40 60
						20
0
2011m1	2011m4	2011m7	2011m10	2012m1	2012m4	0

Bank−issued securities		Government−issued securities

Securitized assets (ABS, MBS)		Other marketable securities
Non−marketable securities

2008m1	2010m1	2012m1	2014m1	2016m1
	ECB		Private Market

Panel (a) shows the ECB liquidity by maturity. vLTRO includes the vLTRO (2011{2014) and T-LTRO (from 2014 onward), other denotes liquidity operations with a maturity below 3 years. Panel (b) plots the average maturity of banks' liabilities with the ECB. Panel (c) plots the evolution security pledging with the ECB split by types. All marketable securities are reported in book values. Non-marketable securities use internal ECB valuation. Panel (d) shows the evolution of the average private market and ECB haircuts for securities pledged by Portuguese banks with the ECB. Vertical red (black) dashed lines denote the 2011 vLTRO (2009 LTRO) period.

ECB Working Paper Series No 2521 / February 2021

Figure 2: Lending outcomes using dynamic dierence in dierences

.3
.2
.1
0
-.1
-.2
2011m6	2011m12	2012m6	2012m12
	date

This gure presents coecient estimates of k for each month from equation (2). Vertical bands represent +/- 1.96 times standard error of each point estimate. Dashed lines indicate the vLTRO period. Standard errors are two-way clustered at the bank-time and rm level.

ECB Working Paper Series No 2521 / February 2021

Figure 3: Defaults of new loans

-.5

2011q1

2011q3

2012q1

2012q3

2013q1

quarter

This gure presents coecient estimates of k for each quarter from equation (8). Estimates in red color refer to the periods of high defaults of new loans following the reduction in LOLR policy uncertainty induced by the vLTRO. Vertical bands represent +/- 1.96 times standard error of each point estimate. The dashed line shows the policy announcement. Standard errors are clustered at the industry level.

ECB Working Paper Series No 2521 / February 2021

Table 1: Summary statistics of bank characteristics

		N	Mean	S.D.	p25	p50	p75

Main exposure measure:
Haircut subsidy 2011	% Assets	30	2.48	3.91	0	0.05	3.78
Haircut subsidy 2009	% Assets	33	0.01	0.05	0	0	0.001
Alternative exposure measures:
Total ECB liquidity	% Assets	30	5.9	8.4	0.0	1.1	9.2
Eligible securities	% Assets	30	9.2	11.5	0.0	5.1	13.9
Eligible securities (1-3Y)	% Assets	30	2.2	3.6	0.0	0.1	2.8
Other bank observables (as of September 2011):
Total Assets	bn EUR	30	16.9	32.7	0.6	1.9	11.1
Cash reserves	% Assets	30	0.7	0.9	0.0	0.5	1.1
Loans	% Assets	30	47.4	27.6	25.5	43.0	69.2
Deposits	% Assets	30	33.3	29.5	4.4	33.6	50.4
Leverage	Liab/Equity	30	13.6	12.1	6.6	11.0	14.6
Capital ratio	Capital/RWA	30	11.2	16.3	8.8	10.3	14.4
ROA	Prot/Assets	30	0.4	2.6	-0.3	0.0	0.4
Equity	% Assets	30	13.8	15.8	6.4	8.4	13.2

ECB Working Paper Series No 2521 / February 2021

Table 2: Intensive margin: baseline version

				log(crediti;j;t)
	(1)	(2)	(3)	(4)	(5)

HaircutSubsidyi Postt	1.078	1.188	0.874	0.820	0.824
	(0.202)	(0.194)	(0.179)	(0.182)	(0.154)
Time FE	Yes		Yes
Bank FE	Yes		Yes	Yes	Yes
Firm-Time FE				Yes	Yes	Yes
Bank-Firm controls					Yes
Bank-Firm FE						Yes

Observations	2,914,218	1,487,089	1,487,089	1,487,089	1,487,089
Overall R2	0.0674	0.123	0.737	0.916	0.996
Within R2	0.001	0.001	0.0615	0.609	0.101
Loan Sample	Full			Multiple bank relationships

This table presents coecients from regressions related to loan-level intensive margin, as described in equation (1). The dependent variable is log credit drawn by a non-nancial rm j from bank i in month t. Bank-rm controls include the length of bank-rm relationship, indicator whether the loan is secured by collateral and share of loan size to total rm credit which controls for the importance of the lending relationship. Standard errors are two-way clustered at the bank-time and rm level in parentheses. ***

p<0.01, ** p<0.05, * p<0.1.

ECB Working Paper Series No 2521 / February 2021

Table 3: Intensive margin, credit lines and loan maturities

(a) Credit lines

	Drawn credit	Potential credit	Utilization rates
	(1)	(2)	(3)

HaircutSubsidyi Postt	0.663	-1.114	0.694
HaircutSubsidyi Postt CreditLinej	(0.118)	(0.500)	(0.123)
0.215
	(0.120)
Firm-Time FE	Yes	Yes	Yes
Bank-Firm FE	Yes	Yes	Yes

Observations	1,487,089	1,487,089	1,485,260
Overall R2	0.996	0.959	0.829
Within R2	0.103	0.053	0.084

(b) Loan Maturities

				Loan maturityi;j;t
	(1)	(2)	(3)	(4)	(5)

HaircutSubsidyi Postt	1.061	1.561	1.005	1.294	2.196
	(0.184)	(0.0889)	(0.403)	(0.378)	(0.299)
Loan controls	Yes		Yes	Yes	Yes	Yes
Time FE	Yes		Yes
Bank FE	Yes		Yes	Yes	Yes
Firm-Time FE				Yes	Yes	Yes
Bank-Firm controls					Yes
Bank-Firm FE						Yes

Observations	2.056,111	968,463	968,463	968,463	967,555
Overall R2	0.114	0.114	0.556	0.568	0.970
Within R2	0.062	0.047	0.084	0.122	0.098
Loan Sample	Full			Multiple bank relationships

Table (a) examines the role of credit lines. The dependent variable Drawn credit refers to the main outcome variable in the previous analysis - log(crediti;j;t) - the log credit drawn by a non-nancial rm j from bank

in month t. Potential credit is dened as (log(potential crediti;j;t + 1) where potential credit denotes a sum of all unused credit lines between rm j and bank i at month t. Utilization rate is a share of drawn

credit to total approved credit (i.e. the sum of potential credit and drawn credit). Total credit in the denominator is expressed as of September 2011. Table (b) presents coecients from regressions described in equation (3). The dependent variable is the maturity of drawn loans. Loan controls include bank-rm log credit. Bank-rm controls include the length of bank-rm relationship, dummy whether the loan is

secured by collateral, and share of loan size to total rm credit. Standard errors are two-way clustered at the bank-time and rm level in parentheses. p < 0:10, p < 0:05, p < 0:01.

ECB Working Paper Series No 2521 / February 2021

Table 4: Comparison of lending outcomes in the periods of low and high haircut subsidy

	log(crediti;j;t)
	Low-haircut subsidy enviroment	Main sample
	2009 LTRO	2011 vLTRO

	(1)			(2)

2009 HaircutSubsidyi Postt	8.795
2011 HaircutSubsidyi Postt	(7.868)
			0.824
				(0.154)

Eect per one std. deviation	0.440			3.222
Firm-Time FE	Yes	Yes
Bank-Firm FE	Yes	Yes

N	3,414,089			1,487,089
Overall R2	0.977			0.996
Within R2	0.006			0.101

This table presents coecients from equation (1). The dependent variable is log credit granted to private non-nancial rms in Portugal around the LTRO periods. Standard errors are two-way clustered at the bank-time and rm level in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

ECB Working Paper Series No 2521 / February 2021

Table 5: Intensive margin: heterogeneous outcomes by rm type

		log(crediti;j;t)
	(1)	(2)	(3)

HaircutSubsidyi Postt	1.077	1.309	0.631
	(0.192)	(0.236)	(0.183)
HaircutSubsidyi Postt FirmSizej	-0.283
	(0.131)
HaircutSubsidyi Postt RelationFirmj		-0.481
		(0.199)
HaircutSubsidyi Postt RiskyFirmj			0.354
			(0.107)
Firm-Time FE	Yes	Yes	Yes
Bank-Firm FE	Yes	Yes	Yes

Observations	1,166,299	1,166,299	896,268
Overall R2	0.995	0.995	0.994
Within R2	0.120	0.119	0.124

This table presents coecients from regressions related to loan-level intensive margin and rm hetero- geneities, as described in equation (4). All rm heterogeneities are dummy variables: F imsSizej is one if the rm's size is larger than a median rm, RelationF irmj takes a value of one if the bank-rm pair exists for at least 24 months prior to the vLTRO, RiskyF irmj is one if a rm's z-score is above median, Defaultj is one if a rm had had any delinquent loan with any bank in the part year, and zero otherwise. Standard errors are two-way clustered at the bank-time and rm level in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

ECB Working Paper Series No 2521 / February 2021

Table 6: Extensive margin: exit and entry

	EXITi;j			ENTRYi;j
	(1)	(2)	(3)	(4)	(5)	(6)

HaircutSubsidyi	-2.45***	-2.24***	1.57*	1.19**	0.91	1.36
	(0.532)	(0.497)	(0.865)	(0.538)	(0.910)	(0.864)
HaircutSubsidyi RiskyFirmj						1.05
						(0.299)
HaircutSubsidyi DefLoanj							1.77
							(0.617)
Bank controls	Yes	Yes		Yes	Yes	Yes	Yes
Firm controls	Yes			Yes		Yes	Yes
Firm FE		Yes			Yes

Observations	284,179	274,605	49,361	31,873	37,612	49,287
Overall R2	0.25	0.53	0.06	0.47	0.07	0.07
Within R2	-	0.24	-	0.02	-	-

This table presents coecients from regressions related to loan-level extensive margin, as described in equations (5), (6) and (7). For a given loan, EXIT is classied as one if the loan is not renewed and the bank-rm relationship ceases to exist in the post-period. ENTRY equals one if a bank-rm loan consultation entry is matched with the new bank-rm loan in credit registry, and zero otherwise. RiskyF irm takes the value of 1 if the rm's z-score is above the median z-score, and 0 otherwise. DefLoan takes the value of 1 if the rms as defaulted on any loan in the past year, and 0 otherwise. Bank controls include ln(T A), CapitalRatio, LiqRatio, Equity=T A and Loans=T A. Firm controls include log of rm's total assets and industry-district xed eects. Standard errors clustered at the bank level in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

ECB Working Paper Series No 2521 / February 2021

Table 7: Firm-level credit and real outcomes

						log(creditj)		investmentj		employmentj
						(1)	(2)	(3)	(4)	(5)

						1.031	0.791			-6.230
HaircutSubsidyj	0.429	-1.365
						(0.114)	(0.320)	(0.393)	(3.458)	(3.353)
									0.774		11.31
HaircutSubsidyj SmallFirmj
									(0.394)		(5.032)

Bias corr. HaircutSubsidyj	0.644
								Yes	Yes
Bank controls	Yes		Yes	Yes

Bank-rm controls	Yes		Yes	Yes		Yes	Yes
Firm controls	Yes		Yes	Yes		Yes	Yes

Observations	138,225	73,493	73,493	89,176	89,176
R2	0.0273	0.0291	0.0291	0.0129	0.0129

This table presents coecients from regressions related to rm-level intensive margin, as described in equation (9). The dependent variable in Column (1) log(creditj) is a log change in total bank lending on the rm level between the pre- and post- period. Following Jimenez et al. (2019), we compute the bias-correct coecient as :

					var(HaicutSubsidy )	0:03912
bj = bj;OLS
(OLS	i	= 1:031	(1:134 1:029) 0:02042	= 0:644
F E) var(HaicutSubsidyj)
			b	b

Dependent variable in Columns (2)-(3) is an annual log change in rm investment and in Columns (4)-(5) the annual growth rate in employment. Bank controls denote indirect measures of log(T A), CapitalRatio, LiqRatio, Equity=T A, log(pre LT ROpledging) and Loans=T A. Firm controls include log(T Aj) and industry-district xed eects. Standard errors are clustered at the industry-district level in parentheses.

*** p<0.01, ** p<0.05, * p<0.1.

ECB Working Paper Series No 2521 / February 2021

Supplementary Material { Internet Appendix

Policy Uncertainty, Lender of Last Resort

and the Real Economy

ECB and Portuguese bank announcements

A.1 ECB speeches and policy announcements related to the uncertain future of xed rate full allotment

One of the critical institutional aspects why the vLTRO is associated with the reduction in funding uncertainty is that the ECB did not credibly commit to keep the xed-rate full allotment (FRFA) for longer than eight months in the period between 2008{mid 2012. This list of the ECB speeches and policy announcements shows how the regulators slowly moved the FRFA time horizons. Around the time of the vLTRO announcement, the ECB has only committed to keep the FRFA in place until at least the rst half of 2012. The rst long- term commitment to keep the FRFA in place for more than eight months occurred in July 2012 (half a year after the vLTRO) around the same period as Draghi's Whatever it takes" speech related to the Outright Monetary Transactions (OMT).

The Governing Council decided to conduct them in the second quarter of 2011 on the same conditions as in the rst quarter of 2011. This means that we will continue to apply xed-rate tender procedures with full allotment in all our renancing operations at least until mid-July."(Jean-Claude Trichet, March, 21, 2011)
On 6th October the Governing Council, in response to a worsening of liquidity tensions in the market, has committed to maintaining the FRFA policy until the middle of July 2012." (Jose Manuel Gonzalez-Paramo, October 21, 2011)
These measures address the risk that persistent nancial markets tensions could aect the capacity of euro area banks to obtain renancing over longer horizons. [...] in all renancing operations until at least the rst half of 2012 all liquidity demand by banks would be fully allotted at xed rate." (Mario Draghi, December 19, 2011)
The Eurosystem decided to maintain the FRFA procedure in all renancing operations until at least the end of June 2012." (Vitor Constancio, April 25, 2012)

ECB Working Paper Series No 2521 / February 2021

The Governing Council decided in June to continue conducting all renancing opera- tions as xed-rate tender procedures with full allotment, at least until mid-January 2013." (Mario Draghi, July 9, 2012)
The Governing Council decided in December to continue conducting all renancing operations as xed rate tender procedures with full allotment, at least until July 2013." (Mario Draghi, December 17, 2012)
The Governing Council decided in May to continue conducting all renancing oper- ations as xed-rate tender procedures with full allotment at least until mid-July 2014." (Mario Draghi, July 8, 2013)

ECB Working Paper Series No 2521 / February 2021

A.2 Announcement of Portuguese banks related to the vLTRO

We also took the opportunity to borrow from the ECB at three years, which made funding more stable and took pressure o the use of weekly borrowing operations." (Caixa Economica Montepio Geral, Annual Report and Accounts, 2011)
By transforming the short-term nancing with the ECB into 3 years, the Bank not only maintained a very comfortable position regarding permanent liquidity but also guaranteed the same position for the coming 2 years." (Banco Carregosa, Report & Accounts, 2011)
New ECB nancing signicantly increased the liquidity buer and improved nanc- ing structure by replacing short-term maturities by long term funding." (Santander, Annual Report, 2011)
... structural improvement in the prole of maturities, substituting a part of its short term renancing requirements by resources with a maturity of 3 years." (Caixa Geral de Depositos, Annual Report, 2012)
Banco Popolare's adhesion to ECB's three-year LTRO auctions enabled the Group to stabilise its structural liquidity prole." (Banco Popolare, Annual Report, 2012)
This enabled the Group a very respectable liquidity prole able to withstand the most severe stress tests, and made it possible for the LCR (Liquidity Coverage Ratio), envisaged by Basel III for 2015, to record a percentage of over 100%." (Banco Popolare, Annual Report, 2012)
This [LTRO] will allow the cost of nancing to be cut by improving its struc- ture." (CaixaBank, Management Report and Annual Financial Statements, 2011)

ECB Working Paper Series No 2521 / February 2021

total assetsi;Sep2011

Details about the alternative measures of exposure

The paper exploits the ex-ante variation in banks' exposure to the reduction in LOLR central bank uncertainty. In addition the the baseline exposure to the uncertainty proxied by the size of the haircut subsidy on securities pledged by Portuguese banks with the ECB, we also consider three alternative measures of exposure. Finally, we will also perform a robustness check based on an alternative haircut subsidy measure that takes into account all eligible securities held by banks.

First, we construct a liability-side measure dened as the sum of all the short-term ECB funding taken up by a bank as of September 2011 and normalized to its total assets:

Existing ECB liquidityi = total secured ECB borrowingi;Sep2011 :

This measure captures the fact that the policy allowed banks to swap the existing short- term funding provided by the ECB into the newly available very long-term(three-year) funding while keeping all other margins unchanged (e.g., same eligible collateral, haircuts, interest rates). Thus, banks were able to costlessly increase the maturity of their liabilities and hence lower their uncertainty regarding the future stance of the LOLR policy. In the data, we observe that banks swapped on average 86% of their short-term funding into the three-year funding.

Second, we consider an asset-side measure that captures total bank's holdings of securities eligible as a collateral:

EligibleSec Hold.i = holdings of eligible securities (ECB haircut-adjusted value)i;Sep2011 :

As banks need to pledge eligible collateral to participate in the LOLR funding, this measure directly captures the total borrowing capacity of a bank.41 Exploiting the matching of the ECB data on collateral and haircuts with the Portuguese data on banks' security holdings, we derive the haircut-adjusted value of all eligible securities in bank's portfolio, regardless of whether they had been pledged with the central bank before or not.42

Third, while banks can pledge any eligible security as a collateral, the main benet of the vLTRO is captured by securities that mature in the horizon of one to three years after the policy announcement. By lengthening the maturity of repo operations to three years, ECB implicitly decreased rollover risk for the funding backed by the securities that mature

41This approach is related to Rodnyansky2017 who study the lending impact of QE in the US.

42We use ISIN-specic haircuts applied by the ECB and compute the haircut-adjusted value of all eligible securities. In the data, we clearly observe over-collateralization. That is, bank utilize around 75% of their borrowing capacity and leave the rest as a buer against sudden changes in asset prices and related margin calls. Importantly, we do not nd any change in the over-collateralization rates around the policy dates.

ECB Working Paper Series No 2521 / February 2021

total assetsi;Sep2011

shortly before the vLTRO expiration. In other words, banks do not need to be concerned about the price volatility of these securities at the time of the vLTRO repayment and as a result, they would not need to face re-sale risk due to rollover issues.43 As a result, we narrow down the previous exposure measure to the most relevant group of eligible securities:

EligibleSec Hold. (1Y-3Y)i = holdings of eligible securities, maturity 2 (1Y; 3Y )i;Sep2011 :

Table B1 presents the loan-level intensive margin results using three alternative measures of exposure. Figure B1 presents coecient estimates from the dynamic dierences-in- dierences described in equation (2) for three alternative exposure measures.

Table B1: Intensive margin: alternative exposure measures

			log(crediti;j;t)
Exposure denitions	Haircut		Existing ECB	EligSec Hold	EligSec Hold
	Subsidy		Liquidity	(All matur.)	(1Y-3Y)

	Baseline			Robustness
	(1)	(2)	(3)	(4)

Exposurei P ostt	0.824***	0.631***	0.521***	0.530***
	(0.154)	(0.063)	(0.053)	(0.147)
Firm-Time FE	Yes		Yes	Yes	Yes
Bank-Firm FE	Yes		Yes	Yes	Yes

Observations	1,487,089	1,487,089	1,487,089	1,487,089
Overall R2	0.996	0.996	0.996	0.996
Within R2	0.101	0.106	0.095	0.087

This table presents coecients from regressions related to loan-level intensive margin, as described in equation (1) for four measures of exposure. The dependent variable is log credit granted to private non- nancial corporations in Portugal. Standard errors are two-way clustered at the bank-time and rm level in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

43There are three reasons why we focus on securities with remaining maturity of one to three years: (i) until the announcement the longest ECB operation had a maturity of one year, (ii) banks can repay vLTRO early but no sooner than one year after the allotment, and (iii) the full vLTRO maturity is three years.

ECB Working Paper Series No 2521 / February 2021

Figure B1: Lending outcomes: robustness to other measures

(a) ECB repo funding

.4
.2
0
-.2
2011m6	2011m12	2012m6	2012m12
	date

(b) All eligible securities

.3
.2
.1
0
-.1
2011m6	2011m12	2012m6	2012m12
	date

.3
.2
.1
0
-.1
2011m6	2011m12	2012m6	2012m12
	date

This gure presents coecient estimates of k for each month from equation (2). We divide banks into two groups: exposed and non-exposed. Panel (a) denotes a bank as exposed if it was borrowing from the central bank in repo operations prior to the vLTRO announcement, and zero otherwise. Panel (b) denes exposed as one if a bank i was holding securities eligible as collateral prior to the policy announcement, and zero otherwise. Panel (c) set exposed to one if a bank i was holding securities with that it will not have to re sell at the repayment date due to rollover issues (the remaining maturity of securities is between 1 and 3 years), and zero otherwise. Vertical bands represent +/- 1.96 times standard error of each point estimate. Dashed lines separate the vLTRO period. Standard errors are two-way clustered at the bank-time and rm level.

ECB Working Paper Series No 2521 / February 2021

Pledged vs Eligible Securities. We also examine an alternative measure of haircut subsidy. While our baseline measure of haircut subsidy was computed for only securities pledged with the ECB, as a robustness we consider a haircut subsidy based on all eligible securities held by a bank (both pledged and not pledged with the ECB). Table B2 presents coecients from the baseline regressions related to loan-level intensive margin, as described in equation (1), for two alternative measures of haircut subsidy. The results remain robust to the alternative specication of the haircut subsidy.

Table B2: Intensive margin: alternative denition of haircut subsidy

	log(crediti;j;t)
Exposure denitions	Haircut Subsidy		Haircut Subsidy
	Securities Pledged with ECB		Eligible Securities Held

	Baseline		Robustness
	(1)	(2)

Exposurei P ostt	0.824***	0.871***
	(0.154)	(0.158)
Firm-Time FE	Yes		Yes
Bank-Firm FE	Yes		Yes

Observations	1,487,089	1,487,089
Overall R2	0.996	0.996
Within R2	0.101	0.100

This table presents coecients from regressions related to loan-level intensive margin, as described in equation (1) for two alternative measures of haircut subsidy. Column (1) corresponds to the haircut subsidy computed for securities held by a bank and pledged with the ECB. Column (2) corresponds to the haircut subsidy computed for all eligible securities held by a bank (both pledged and not pledged with the ECB). Standard errors are two-way clustered at the bank-time and rm level in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

ECB Working Paper Series No 2521 / February 2021

Dierences in bank observables

Table C1: Comparison exposed and non-exposed banks

		Exposed banks		Non-exposed banks		Di.

		Mean	S.D.		Mean	S.D.

Total Assets	bn EUR	28.5	40.0	1.6	2.0	26.87**
Cash reserves	% Assets	0.7	0.5	0.7	1.3	0.0
Capital ratio	Capital/RWA	14.3	7.5	7.2	23.2	7.1
ROA	Prot/Assets	0.0	1.2	0.9	3.8	-0.9
Loans	% Assets	41.4	19.9	55.3	34.5	-13.9
Deposits	% Assets	30.7	17.7	36.8	40.8	-6.1
Leverage	Liab/Equity	13.0	9.4	14.4	15.3	-1.4
Equity	% Assets	11.6	10.8	16.7	20.8	-5.1
Security holdings	% Assets	28.0	15.4	8.1	12.4	19.9***

This table shows the means of the respective variables for the group of exposed and the group of non- exposed banks. We dene exposed banks as ones which participated in the three-year vLTRO funding and non-exposed otherwise. All variables expect for are reported as of September 2011. Stars denote p-values for pairwise t-tests that test whether the mean is the same for the two groups of banks. *** p<0.01, ** p<0.05, * p<0.1.

Table C1 compares averages of bank's observable characteristics by splitting banks into two groups: exposed and non-exposed. We nd that exposed banks are on average larger and hold more securities while they do not dier across other features such as cash holdings, capitalization, protability or leverage.

The dierence in size highlights that all non-exposed banks are relatively small (mean size is EUR 1.6bn with a standard deviation of EUR 2bn) while exposed banks are very heterogeneous (mean size is EUR 28.5bn with a signicant standard deviation equal to EUR 40bn). The size dierence is directly related to the xed cost of establishing an infrastructure to borrow from the ECB (for example a trading desk). Smaller banks may not nd it benecial to bear this xed cost. Furthermore, it is important to point out that while all large Portuguese banks fall into the exposed category (i.e., the dummy variable of exposed is equal to 1), the actual magnitude of the haircut subsidy (i.e., continuous measure of exposure) does not correlate with bank size.

Table C1 also shows that exposed banks on average hold more securities. The dierence in terms of security holdings conrms the fact that banks must hold securities to be able to benet from the haircut subsidy. In face, we exploit holdings of eligible securities as one of the alternative measure of exposure presented in Internet B and in the robustness exercise in the Internet Appendix Table B1.44

44Majority of securities holding of banks are in fact securities that are eligible for pledging with the ECB.

ECB Working Paper Series No 2521 / February 2021

Collapsed dierence-in-dierences

We collapse the time series information from equation (1) into a pre- (June{October 2011) and post- (February{June 2012) periods to derive more conservative standard errors :

log(crediti;j) = j + HaircutSubsidyi + Bi + Qi;j + i;j;

(10)

where log(crediti;j) denotes the change in average bank-rm credit between the two collapsed periods. Unlike the baseline specication used in the main paper, the collapsed DID does not allow us to implement the rich set of xed eects as in equation (1) and we therefore include a set of bank controls, Bi, (total assets, capital ratio, liquidity rate, equity/TA, and loans/TA) and bank-rm controls, Qi;j, (the length of the bank-rm relationship and information about previous loan delinquencies in this relationship). We use rm xed eects, j, to disentangle credit demand from credit supply and compare change in lending outcomes for a rm borrowing from at least two banks.

Table D1: Intensive margin: collapsed version

			log(crediti;j)
	(1)	(2)	(3)	(4)	(5)

HaircutSubsidyi	1.202	1.235	1.095	1.134	1.029
	(0.499)	(0.480)	(0.506)	(0.477)	(0.348)
Bank Controls	Yes	Yes		Yes	Yes	Yes
Bank-Firm Controls	Yes	Yes		Yes	Yes	Yes
Firm Controls		Yes			Yes
Firm FE						Yes

Observations	203,018	202,920	114,116	114,116	114,116
Overall R2	0.0283	0.0460	0.0445	0.0650	0.467
Within R2	-	-	-	-	0.0737
Sample	Full sample		Multiple bank relationships

This table presents coecients from collapsed regressions related to loan-level intensive margin, as described in equation (10). The dependent variable is the change in average log credit granted to private non-nancial corporations before and after the vLTRO. Bank controls include ln(T A), CapitalRatio, LiqRatio, Equity=T A and Loans=T A. Firm controls include log of total assets of a rm j and industry- district xed eects. Bank-rm controls include the length of the bank-rm pair and information about previous loan delinquencies in the bank-rm relationship. Standard errors clustered at the bank level in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Table D1 shows that our ndings are robust to the collapsed DID which compares the average lending before and after the policy implementation following equation (10). While the

As are result there is very little dierence between all security holding and eligible security holdings.

ECB Working Paper Series No 2521 / February 2021

eects are quantitatively similar, we prefer to use the time-series specication as our baseline estimation since it allows us to absorb any observable and unobservable time-invariantbank-rm characteristics into xed eects.

Alternative credit, rms and bank variables

Table E1 shows that our ndings are also robust to changes in the credit and rm denitions. Columns (2) and (3) show variations to the baseline. Column (2) reports results for the dependent variable measured as the log of total credit. We dene total credit as a sum of drawn credit and potential credit (i.e., unused credit lines which are reported o-balance sheet). Our results are also robust to dierent denitions of rms. While the baseline results are reported for private non-nancial corporations, we also extend the denition of rms to include public rms and individual entrepreneurs which leads to an increase of the sample by 900,000 additional observations (denoted as All rms in Column(3)).

A possible concern is that the banks that did not participate in the ECB's open market operations could be signicantly dierent from the exposed banks. As exposed banks are on average larger and hold more securities.45, In Column (4) of Table E1 we only focus on the variation in the cross-section of exposed banks and the results remain consistent.

Finally, Column (5) shows that results hold robust also when controlling for bank characteristics (i.e., log of total assets, capital ratio, liquidity ratio, equity ratio and loan-to-assets ratio) interacted with the POST dummy.

E.1 EBA stress tests

A potential threat to our identication strategy comes from other concurrent policy actions. Around the time of the vLTRO announcement, four banks were undergoing the stress tests conducted by the European Banking Authority. To examine if our results are robust to this potential confounding factor, we exclude these banks from our sample as a part of the robustness exercise. In Column (6) of Table E1 we drop these banks from our sample and the results remain robust.

45See Internet C for more details.

ECB Working Paper Series No 2521 / February 2021

2021 February / 2521 No Series Paper Working ECB

Table E1: Intensive margin: robustness

				Baseline		Total credit		All rms		Exposed	Bank	EBA shock		Endogenous
										banks	Controls			uptake

		P ost		(1)	(2)	(3)	(4)	(5)	(6)	(7)
HaircutSubsidy	i	t	0.824	0.885	0.994	0.610	0.911	2.018
vLTROuptakei P ostt		(0.154)	(0.146)	(0.130)	(0.222)	(0.258)	(0.243)
											0.167
													(0.411)

Firm-Time FE				Yes		Yes		Yes		Yes	Yes	Yes		Yes
Bank-Firm FE				Yes		Yes		Yes		Yes	Yes	Yes		Yes
Observations				1,487,089	1,673,619	2,387,616	1,012,225	1,487,089	447,520	1,487,089
Overall R2				0.996	0.996	0.996	0.996	0.996	0.991	0.936

This table presents cientscoe from regressions related to loan-level intensive margin, as described in equation (1). Column (1) represents the baseline cationspeci as shown in the last column of Table 2. The dependent variable is log credit granted to private non-nancial corporations in Portugal. The exposure is nedde as a sum of all available borrowing from the ECB as a share of total assets prior to the vLTRO. Columns (2){(6) present variations to the baseline. Column (2) shows results for the dependent variable measured as a log of total credit (a sum of credit granted and potential credit). Column

extends the sample to all rms in Portugal, including not only private NFC but also independent entrepreneurs and publicly-owned companies. Column
focuses only on the variation across banks that were exposed to the ECB open market operations prior to the vLTRO (in September 2011). Column (5) introduces bank controls interacted with the POST dummy. Column (6) drops four largest banks which were subject to the stress test exercise conducted by the European Banking Association (EBA) around the same time as the vLTRO. Finally, Column (7) illustrates the null ecte when using the measure of endogenous uptake of the vLTRO as a source of cross-sectional variation. Standard errors are two-way clustered at the bank-time and rm level in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Placebo test

Figure F1: Placebo test: freeze of the European interbank market in August 2007

2
1
0
-1
2007m1	2007m7	2008m1	2008m7
		date

The gure uses an unexpected freeze of the European interbank market in August 2007 as a placebo test. The negative eects of the interbank liquidity crunch on lending in Portugal were previously shown by Iyer2014. This gure presents coecient estimates of k for each month from equation (2). There is no evidence that banks more exposed to the vLTRO in 2011 were generally more sensitive to the liquidity dry-up in 2007 (the plotted estimates of k are not signicantly dierent from zero). Vertical bands represent +/- 1.96 times standard error of each point estimate. Dashed line shows the European interbank market in August 2007. Standard errors are two-way clustered at the bank-time and rm level.

Internet Appendix Figure F1 uses the 2007 liquidity freeze as a placebo sample to investigate whether the banks more exposed to the reduction in funding uncertainty induced by the vLTRO policy in 2011 were also more sensitive to the 2007 liquidity dry-up. We follow the dynamic setup specication from equation (2) and replace the left-hand-side lending outcomes in 2011{2012 with the lending outcomes in 2007. If the 2007 and 2011 exposures were spuriously correlated, we would expect negative and statistically signicant coecients k after August 2007. Instead, we nd that the plotted estimates of k are not statistically dierent from zero throughout 2007. We can conclude that there is no evidence that the banks more exposed to the 2011 reduction in bank funding uncertainty are generally more sensitive to liquidity shocks.

ECB Working Paper Series No 2521 / February 2021

Standard error clustering

Table G1 documents that our results are robust to alternative clustering either at the bank- time (i.e. unit of variations), bank and time and bank and rm level.

Table G1: Intensive margin: robustness to standard error clustering

			log(crediti;j;t)
S.E. Clustering	Bank-Time		Bank-Time		Bank and Time	Bank and Firm
	and Firm

	(1)	(2)	(3)	(4)

HaircutSubsidyi Postt	0.824	0.824	0.824	0.824
	(0.202)	(0.202)	(0.462)	(0.457)
Bank-Firm FE	Yes		Yes		Yes	Yes
Firm-Time FE	Yes		Yes		Yes	Yes

Observations	1,487,089	1,487,089	1,487,089	1,487,089
Overall R2	0.996	0.996	0.996	0.996

This table presents coecients from regressions relating to loan-level intensive margin, as described in equation (1). The dependent variable is log credit granted to private non-nancial corporations in Portugal. Clustered standard errors in parentheses. Level of standard error clustering is bank-time and rm, bank- time, bank and time, and and bank and rm, respectively. *** p<0.01, ** p<0.05, * p<0.1.

ECB Working Paper Series No 2521 / February 2021

Additional Figures and Tables

Figure H1: Cross-sectional variation in haircut subsidy at the bank level

.6
.4
Fraction
.2
0	.05	.1	.15
0
	Bank Haircut Subsidy / TA

This histogram shows the variation of haircut subsidy at the cross-section of banks in September 2011.

Figure H2: Cross-sectional variation in pledging of securitized assets

pre

post

.4
Fraction
.2
0
0	.2	.4	.6	.8	1 0	.2	.4	.6	.8	1
Graphs by post		Securitized Assets to Total Pledging

These histograms show the variation of securitized assets (as a share of total pledging) at the cross-section of exposed banks. 'Pre' denotes average pledging of each bank in the period prior to the vLTRO, i.e. Jun{Oct 2011 and 'Post' denotes average pledging of each bank in the period after to the vLTRO, i.e. Feb{Jun 2012.

ECB Working Paper Series No 2521 / February 2021

Table H1: Summary statistics of loan characteristics

Mean

S.D.

p25

p50

p75

Baseline sample (Firms with multiple bank relationships)

Drawn credit (baseline)	1,487,089	452,218	3,878,045	15,008	49,000	186,308
All credit	1,673,619	534,242	4,985,075	13,802	48,214	193,284

			Full sample of all rms

Drawn credit	2,914,218	349,502	3,413,031	10,143	30,761	118,783
All credit	3,276,700	383,756	4,051,314	8,559	29,062	116,491

This table reports the summary statistics of monthly loan-level credit data for the period June 2011{June 2012 in EUR. Minimum reporting threshold is EUR 50. Drawn credit represents performing regular and renegotiated credit. All credit represents the sum of drawn credit and potential credit (i.e., unused credit lines).

ECB Working Paper Series No 2521 / February 2021

Acknowledgements

We thank Alin Andries, Michele Boldrin, Diana Bonfim, Markus Brunnermeier, Francesco D'Acunto, Itamar Drechsler, Mariassunta Giannetti, Erik Gilje, Michael Gofman, Itay Goldstein, João Gomes, Florian Heider, Jakub Kastl, Constantine Yannelis, Luc Laeven, Atif Mian, Daniel Paravisini, José-Luis Peydró, Kasper Roszbach, Philipp Schnabl, Amit Seru, Enrico Sette, Janis Skrastins, Luke Taylor, Stijn Van Nieuwerburgh and Wei Xiong, and seminar and conference participants at the Bank of Portugal, ECB, Princeton University, Wharton School, Norges Bank, BI Norwegian Business School, Indiana (Kelley), Temple University, Barnard College, Bank of Canada, Bank of Spain, Fed Board, CEPR 3rd Annual Spring Symposium in Financial Economics, IHW Halle, SED 2018, the 7th MoFiR Workshop on Banking, Columbia, FRB NY, Williams College, 10th EBC Network Conference, Lenzerheide, FIRS 2019, EFA Day Ahead, NFA, FMA, and EFA for helpful comments and suggestions. We are grateful to Paulo Guimaraes, Ettore Panetti, Pedro Prospero, Fatima Teodoro, Maria Lucena Vieira and the staff of the "Laboratorio de Investigacao com Microdado" at the Bank of Portugal (BPLim) for their help with data collection and management and colleagues in the Department of Market Operations of the European Central Bank and the Bank of Portugal for useful comments. This paper started when Mendicino was an economist in the Department of Economic Studies of the Bank of Portugal. Jasova gratefully acknowledges support from the Czech Science Foundation (Project No. GA 18-05244S) and the European Union's Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No. 681228. We thank the Bank of Portugal and the European Central Bank for the hospitality during our research visits. The opinions expressed herein are those of the authors and do not necessarily reflect those of the ECB or the Eurosystem. All errors are our own.

Martina Jasova

Barnard College, Columbia University, New York, United States; email: mjasova@barnard.edu

Caterina Mendicino

European Central Bank, Frankfurt am Main, Germany; Bank of Portugal, Lisbon, Portugal; email: caterina.mendicino1@ecb.europa.eu

Dominik Supera

Wharton School, University of Pennsylvania, Philadelphia, United States; email: superad@wharton.upenn.edu

Postal address 60640 Frankfurt am Main, Germany

Telephone +49 69 1344 0

Website www.ecb.europa.eu

All rights reserved. Any reproduction, publication and reprint in the form of a different publication, whether printed or produced electronically, in whole or in part, is permitted only with the explicit written authorisation of the ECB or the authors.

This paper can be downloaded without charge from www.ecb.europa.eu, from the Social Science Research Network electronic library or from RePEc: Research Papers in Economics. Information on all of the papers published in the ECB Working Paper Series can be found on the ECB's website.

PDF

ISBN 978-92-899-4521-9

ISSN 1725-2806

doi:10.2866/442444

QB-AR-21-012-EN-N

This is an excerpt of the original content. To continue reading it, access the original document here.

Attachments

Original document
Permalink

Disclaimer

ECB - European Central Bank published this content on 08 February 2021 and is solely responsible for the information contained therein. Distributed by Public, unedited and unaltered, on 08 February 2021 14:58:09 UTC.

GSK : UBS gives a Buy rating	07:50am	ZD
BBVA : RBC maintains a Buy rating	07:49am	ZD
AMAZON COM INC : Gets a Buy rating from JP Morgan	07:48am	ZD
HSBC HLDGS : JP Morgan reiterates its Neutral rating	07:47am	ZD
India's Jindal Stainless to invest $647 million for expansion, acquisition plans	07:43am	RE
Johnson Controls rides on residential construction demand to beat profit estimates	07:43am	RE
	07:36am	ZM
	07:36am	ZM
	07:35am	ZM
	07:35am	ZM
	07:35am	ZM
	07:35am	ZM
	07:35am	ZM
	07:35am	ZM
Ethiopian Airlines sees 30% surge in passengers this year, CEO says	07:35am	RE

Stock Market News in real time

Policy uncertainty, lender of last resort and the real economy

All News: More news

Headlines

Most Read News