Effects of liquidity coverage ratio (LCR) on banks' balance sheet composition

Effects of Liquidity Coverage Ratio (LCR) on Banks’

Balance Sheet Composition

Iván Antonio Flores Martínez 11691697 Abstract University of Amsterdam, Amsterdam Business School Master in Finance: Banking and Regulation This paper analyzes the effect of Liquidity Coverage Ratio (LCR) on Banks’ Balance Sheet composition by calculating the Average Treatment Effect (ATE), one by one for each component of the balance sheet (assets/liabilities). Results show that Banks with High liquidity condition compared to the requirement (LCR>100%) tend to stop assessing their real liquidity needs, reallocating the extra liquidity holdings to a more profitable asset dimension. Adjustments take place primarily through the asset side. Conversely, Banks with a Low liquidity condition compared to the requirement (LCR<100%) move to a more stable funding source and reduce their lending to the economy. Adjustments in this group take place predominantly through liabilities side. Liquidity regulation that relies only on accounting ratios might be distortionary for the former group and binding with negative externalities, for the latter. To the best of my knowledge, this is the first study that measures the impact of LCR as proposed in Basel III in Banks’ balance sheet composition for two different treatment groups, characterized by liquidity conditions.

Statement of Originality This document is written by student Iván Antonio Flores Martínez who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Table of Contents ……… 3 I. Introduction ……… 4 I.1. Liquidity and the Banking Sector……….... 4 I.2. Liquidity during the financial crisis of 2008 ……… 5 I.3. Basel Committee on Banking Supervision (BCBS) ……… 6 I.4. Principal objective and research question ……… 7 II. Literature review ………. 8 II.1. The long-term optimal liquidity level ……… 9 II.2. Determinants of Banks’ liquidity holdings ……… 10 II.3. Individual Liquidity Guidance: An approach to LCR ……… 11 II.4. An alternative to accounting ratios ……….. 13 II.5. Hypothesis ……… 14 III. Data ……….. 15 III.1. Variables of interest……….. 16 III.2. Implementation of liquidity regulation: Mexican case ……….. 17 III.3. Treatments and control groups ………. 18 III.4. A first look at the data: LCR approximation ……… 19 III.5. Descriptive statistics ………. 22 IV. Methodology ………. 24 IV.1. Program evaluation ……….. 24 IV.2. Local projection method and Empirical strategy ……….. 24 IV.3. Estimation model ……… 25 IV.4. Selection of control variables ………. 27 V. Baseline results ………. 30 V.1. High liquidity treatment group ……… 31 V.2. Low liquidity treatment group ……… 33 VI. Robustness check ……… 35 VII. Conclusions ………. 37 VIII. Shortcomings of the study……….. 39 IX. Appendix ……… 40 IX.A Impact of Liquidity Coverage Ratio on Balance Sheet: High Liquidity Group ……… 40 IX.B Impact of Liquidity Coverage Ratio on Balance Sheet: Low Liquidity Group ………. 40 IX.C Impact of Liquidity Coverage Ratio on Balance Sheet: High Liquidity Group (RC) …… 41 IX.D Impact of Liquidity Coverage Ratio on Balance Sheet: Low Liquidity Group (RC) ……. 41 X. References ……… 42

I. Introduction

I.1. Liquidity and the Banking sector

What is liquidity? The answer to this question seems very obvious because the word “liquidity” is used to refer different dimensions of the same concept all over the news and common conversations without loss of generality. However, Foucault, Pagano and Röell (2013) divide this general idea into three facets which I retake with the objective to point out different implications derived from it. Market liquidity is the first feature, and it refers to the gap between the fair value of a certain asset and the price to which can be sold in the market at some point in time. Funding liquidity as a second feature indicates the ability of a bank or corporation to meet its current obligations without became financially constrain, that is to possess enough cash or fast access to cheap credit lines. Finally, Monetary liquidity represents an aggregation of assets based on its degree of market liquidity, and is widely used in macroeconomics. The three dimensions are deeply interconnected, yet each of them can be affected through diverse channels: securities market regulation, banking regulation, monetary and fiscal policies are some examples. Throughout this document, “liquidity” mainly refers to the second dimension and banking regulation is the channel examined.

To understand why liquidity is very important for Banks it is vital to know what kind of services they provide to the economy. In simple words, they became the main intermediaries between people with a financial surplus and people looking money for investment opportunities, or a financial deficit. However, this service has many other implications. First, by receiving people’s money in the form of deposits, Banks act as a big safe-deposit box for savings which also means that an important part of the payment system relies on them. Second, Banks perform a liquidity and maturity transformation, that is they transform very liquid and short-term assets into more illiquid and long-term ones. All this can be done by 1) granting depositors instant access to their money at any moment they want, 2) providing long-term funds to borrowers and 3) a certain level of public confidence, preferably at the same time.

Following the idea, liquidity becomes an issue when there is a mismatch between depositors’ demand for their money to satisfy their own liquidity needs and the level of liquidity hold by a certain bank. It is reasonable to think that Banks know their customers and have established internal measures to determine the level of liquidity required or they perform liquidity risk management. I.2. Liquidity during the financial crisis of 2008 The latest financial crisis of 2008 has proven that this is not the case, or at least I would say that the measures established relied on the ceteris paribus premise of a demand of money based on “normal or common” behavior. Without taking into account 1) possible loss of confidence by customers, 2) liquidity problems in other areas of the financial system (contagion), 3) innovations in the way of making banking itself (e.g., rely more on short-term funding and repos rather than traditional deposits) or 4) innovations in the financial system in general (e.g., asset-backed securities (ABS), Credit Default Swaps (CDS), etc.).

During 2008 the financial sector experienced a liquidity crisis of unexpected magnitude. It was unexpected because it started in the subprime mortgages sector, a very small sector compared to the financial system as a whole. Moreover, market liquidity was good enough, and Banks hold healthy capital levels. The key point to understand how the financial system fell into a vicious circle that caused liquidity evaporation and bankruptcies is Uncertainty. It started in the subprime sector as a result of the development of various new financial instruments that play a vital role in spreading panic to other instruments and sectors very quickly. If the market is facing uncertainty and reacting accordingly, the rational action to do as a market participant is to take your money out while you can, causing liquidity vanished. Caballero and Krishnamurthy (2008) compare the case to the musical chairs game. Under normal circumstances, just one player stands at the end of the game, but if the rules are not clear participants goes to panic, creating chaos and causing the end of the game with more than one loser or worst-case scenario no winner.

Since this moment, the role of liquidity became a central topic again for financial authorities, due to its implications for banking system stability. It is not the first time in recent history that regulators realize the importance of financial supervision, regarding liquidity and capital issues. In 1974 an international banking crisis was starring by Herstatt Bank in Germany. At that time, German authorities were in a dilemma between a self-regulated market and government interventions.

I.3. Basel Committee on Banking Supervision

As described by Mourlon-Druol (2015) in his narrative of events, the collapse of Herstatt Bank and the consequences for the banking system in 1974 was the result of a wide reliance on self-regulated markets and a late and ineffective response by the supervisors. The situation is often recognized as the inflection point in liquidity and capital regulation as well as the origin of the Basel Committee on Banking Supervision (BCBS). The Committee had the mandate to establish cooperation among the international community to provide the banking system with minimum standards of behavior and supervision, in terms of liquidity and capital, with the ultimate objective of preventing an international crisis or at least assure its survival with the minimum amount of losses as possible.

Unfortunately, the Committee went from the well-known opening meeting in 1975, with George Blunden as a Charmain stating that “…the Committee’s main objective was to help ensure bank solvency and liquidity” Goodhart (2011) to more than 30 years of supervision focus on capital requirements, failing to incorporate liquidity as an issue that could threaten the stability of the banking systems overall. After healthy banks, in terms of capital, were victims of the liquidity evaporation and faced difficulties accessing funding markets during 2008, the crisis brought to life liquidity as a problem that can have consequences of the same magnitude than capital, a problem that has evolved with the time, while regulation is 30 years behind.

Under these circumstances, the BCBS released the “Principles for Sound Liquidity Risk Management and Supervision” in 2008 and “Basel III: The Liquidity Coverage Ratio and liquidity risk monitoring tools” in 2013. The former as simply guidelines to assess liquidity

risk and the latter as a step forward in the construction of minimum regulatory standards. Basel III stands out for the development of the Liquidity Coverage Ratio (LCR) and the Net Stable Funding Ratio (NSFR). The objective is to assure that any bank has: 1) Enough High-Quality Liquid Assets (HQLA) to cope with a stress scenario of 30 days duration and 2) sources of funding that generates a more stable maturity structure. The implementation was designed gradually. The LCR was planned to start in January 2015 at a 60% level, increasing 10% each year until reach 100% in 2019, while the NSFR was planned to become a minimum standard in January 2018. Taking this into consideration and data availability, the present document only analyzes the establishment and impact of the LCR in bank’s balance sheet composition. Summarizing, it took almost ten years to construct and apply the first attempt of liquidity regulation since the crisis. In my opinion, this delay is due to the financial authorities concern about the mechanisms to comply with the regulation and its effects on Banks’ operation. It is optimist to expect that Banks will adjust the amount of HQLA to the current level of operation, but given the accounting nature of the LCR, they can respond in many ways, adjusting different dimensions of their balance sheets to meet the requirement, having an impact not only for the financial system but probably for the real economy. I.4. Principal objective and research question. My principal concern is that the establishment of accounting tools could cause side effects by “allowing” banks to stop assessing their own liquidity risk. On the one hand banks holding liquidity under the requirement will find one way or another to fully comply, but on the other hand banks over the requirement could feel free to relieve their “high” current levels of liquidity. Just by looking the history of capital requirements as an example (Basel I, Basel II and Basel III) it has become clear that several shortcomings should be expected, even if this idea implies that regulators have to publicly recognize them and reform the law. Some episodes in financial history have demonstrated that regulations (regardless of the good intentions) can generate consequences of big impact for the financial system and the

An example to remember is the Short-selling Bans during the financial crisis of 2008. Regulators imposed bans on short sales with the goal of preventing stock price to keep dropping, particularly Banks’ stock. Beber and Pagano (2013) found that this regulation was damaging for market liquidity (first dimension of liquidity), predominantly for stocks with small capitalization and high volatility, it also slowed price discovery and lastly failed to support prices. The negative effects were so important that Christopher Cox SEC Chairman, in a telephone interview in December 2008, told Reuters “Knowing what we know now, I believe on balance the Commission would not do it again. The costs (of the short-selling ban on financials) appear to outweigh the benefits” Beber and Pagano (2013). Therefore, hypothesis, methodology and further analysis are built to address this concern. I expect to shed some light on these issues following the research question: “What are the effects of the implementation of LCR on Banks’ Balance Sheet composition?” II. Literature Review

It is not surprising that literature regarding the economic and financial implications of liquidity regulation is not conclusive since its implementation is very recent. Nevertheless, authorities, market participants and scholars have drawn several conclusions, which are in some cases complementary or contradictory among them.

At first glance, can be assumed that asset side adjustment is the reasonable response to a liquidity regulation as the LCR. According to Basel III (2013) a bank must hold a sufficient stock of HQLAs, that can be converted into cash in private markets fast and without loss of value. The ultimate objective is to meet its obligations not in normal conditions but in a stress scenario of 30 calendar days. 𝐿𝐶𝑅 = 𝑆𝑡𝑜𝑐𝑘 𝑜𝑓 𝐻𝑄𝐿𝐴𝑠 𝑇𝑜𝑡𝑎𝑙 𝑁𝑒𝑡 𝐶𝑎𝑠ℎ𝑓𝑙𝑜𝑤𝑠 𝑜𝑣𝑒𝑟 𝑡ℎ𝑒 𝑛𝑒𝑥𝑡 30 𝑐𝑎𝑙𝑒𝑛𝑑𝑎𝑟 𝑑𝑎𝑦𝑠 ≥ 100% However, considering my concerns about the implications of the accounting nature of the LCR for banks’ balance sheet composition, I will focus mainly on research which

considered in equal conditions assets and liabilities as possible ways to comply with the liquidity regulation. II.1. The long-term optimal liquidity level Duijm and Wierts (2014) investigate the effect of liquidity regulation on bank liquid assets and liabilities. They test using a database for Dutch banks, which have been subject to a liquidity regulation comparable to the LCR proposed in Basel III since 2003. In this line, Dutch banking system established liquidity requirements even before the financial crisis of 2007. Their results are quite important because they find that asset adjustment is not the principal action that banks take to comply with the required liquidity, but towards liabilities side. This adjustment implies that liquidity ratios should be analyzed carefully, in the Dutch case for example, outflows of market funding are replaced by deposits, increasing competition for this funding source, which could lower Banks’ profits and finally impact the banking system. They also construct the idea of a long-term optimal liquidity level that represents the final outcome of Banks’ willingness to hold liquid assets as a tradeoff between profits and risk. It is reasonable to believe that banks will find an optimal state where they hold enough liquidity to bear the risk present in the market, affecting their profits as little as possible. Based on this construction they find that liquidity buffers tend to increase when liquidity levels are above the long-term optimal level, and the opposite occurred when liquidity levels are under this level. I incorporated this idea in my research because I believe that the actions taken by a specific bank depend on its current liquidity levels. However, instead of constructing a long-term optimal liquidity level I used the LCR binding ratio as the benchmark. In part because after the approval it became clear that not complying with the requirements can have an impact concerning reputational risk and possible fines but also because I strongly believe that the LCR became a substitute for liquidity assessment. That is, banks will stop assessing their liquidity risk based on their internal models but instead, simply following the ratio established by the financial authorities. This is a key point for the hypothesis I present later.

The paper also provides some insight into the role of the financial cycle in the behavior of banks regarding their liquidity levels. They conclude that liquidity buffers are procyclical, increasing during stable times of growth and economic boom and decreasing during bad times. Policy implications are derived from this conclusion because it means that liquidity regulation with these characteristics should be recognized as limited. Micro-prudential regulation, understood, as a regulation based on accounting ratios and individual behavior may be effective in a world where liquidity has not become a problem. However, they propose that should be complemented with macro-prudential regulation than can reach liquidity assessment in a system level. In words of the authors “…monitoring aggregate liquidity developments, both in assets and liabilities, and not to rely only in ratios.”

II.2. Determinants of Banks’ liquidity holdings

Bonner, Lelyveld and Zymek (2014) construct another idea that became central to the structure of my study. The authors mainly focus on the asset side of the balance sheet, because its objective is to discern liquidity holdings determinants for Banks. They find that in the absence of liquidity regulation, banks actively measure their liquidity risk and the models to achieve it vary depending on the financial activities that each bank specialized on, as well as the country where it does business. They test data from 7000 banks from 25 countries belonging to the Organization for Economic Co-operation and Development (OECD) with Generalized Methods of Moments (GMM).

The main conclusion is once the regulation became binding for Banks, the only variables that remain significant are those that represent banking system concentration and disclosure requirements, consequently, country and bank-specific variables that before were significant, do not play a role as determinants for holding liquidity anymore. In my opinion, results are reasonable since competence for liquidity is at play the very moment that a certain level is dictated by law. Concentration and disclosure became the compass that marks how difficult and expensive (or cheap) will be to access liquid assets at any point

in time. Contrary, other indicators about optimal levels of liquidity are no longer relevant since the benchmark is established in a legal document. To exemplify the reasoning behind this idea I will discuss some details about the Mexican case. During the period of study, Mexican authorities allowed banks to provide information only at the end of every month and made information available to the general public every quarter. The implication is that higher demand for liquidity is expected at the end of every month and even higher at the end of every quarter. In a scenario like this, what matters is how much liquidity is available in the market and how much liquidity other Banks hold compare to the required level (i.e., disclosure and concentration). It was not until the beginning of 2018 that the information provided to the supervisor is on a daily basis, but even under these circumstances the same rule applies. Last but not least Bonner et al. (2014) have other interesting findings. Concerning policy implications, they conclude that an environment characterized by the presence of liquidity regulation is not correlated with less financial crisis or conducts to a general increase in liquidity holdings. Nonetheless, liquidity regulations might prevent volatility and allow banks to stay stable during stress scenarios. This is probably the case because the presence of the regulation itself grants stability to the financial system, even though this might mean tradeoff with the real economy stability. It can be argued that regulation will always have shortcomings, but at least the rules of the game are established. II.3. Individual Liquidity Guidance: An approach to LCR Another central paper in the construction of my work is the research conducted by Banerjee and Mio (2014). They study the introduction of the Individual Liquidity Guidance (ILG) in 2010 within the United Kingdom, as a proxy to the internationally established LCR. Results proved that UK Banks increase the portion of HQLA and moved from short-term intrafinancial loans and wholesale funding to a more stable UK non-financial deposits, which also means that there is no impact on the balance sheets’ size or a decline on the lending to the non-financial sector. This implies that the liquidity regulation operated just exactly as

expected and did not bring along any inconvenient to the real economy, which contradicts the opinion among other studies.

That is the main reason that I decide to follow them step by step, technically speaking. The second reason is that the data I selected fits perfectly to perform their methodology, since a “natural” control group is key in their approach and it is derived from the fact that the regulation adopted in the UK did not follow a homogenous pattern, that is, certain type of banks was waived at the time. The same happened in the Mexican banking system case, but I will discuss this issue deeper in the methodology section. Finally, this approach considers adjustments on both sides of the balance sheet, which I believe is the appropriate way to look at this concern. However, I anticipate that combining my hypothesis with their technical approach would generate half of the good news that their estimation shows. The technical implication of my proposition is that instead of having one treatment group, it is required to create two treatment groups. I aim the division of the treatment group into High liquidity and Low liquidity banks, which ultimately would lead to opposite reactions than the simple case when just one treatment group is considered.

It is understandable that a study conducted and published for the Bank of England and the Bank for International Settlements (BIS), is expected to generate positive externalities regarding the impact of the regulation in the market, especially if it is believed that the opposite could also occur. The same reasoning applies when Banerjee and Mio (2014) explain how the financial industry has roughly opposed to the regulation claiming that the effects would generate higher cost for liquidity funding that would be transferred to the real economy through higher interest rates or by reducing the amount of lending. Either case, the reader must keep in mind that Banerjee and Mio (2014) recognized that the period used to assess the impact of the regulation on bank behavior might be short. I would also add that similarly to the Dutch case; the ILG ratio represents a national effort while the LCR stands for an international effort to harmonized liquidity regulation. Therefore, implications in terms of risk assessment, profits, concentration and disclosure play different roles. Consequently, it is reasonable to suppose that the results might be

different even if the methodology is the same since the perspective and circumstances are also different.

II.4. An alternative to accounting ratios

Perotti and Suarez (2011) design an analysis that goes one step forward, since they propose already an alternative to quantity-based ratios that can be ineffective and distortionary under certain circumstances. They compare the efficacy of liquidity ratios as the LCR and NSFR against a tax on short-term funding, what they call a Pigovian tax. This research is key in my study since constitute an indication that not only shortcomings in the regulation as established in Basel III are expected, but replacements or additions are being anticipated.

Results presented for the authors demonstrate that reaction to regulations rely primarily on bank level characteristics, arguing that effectiveness of the two approaches to liquidity regulation (including accounting ratios) is subject to particular differences between Banks. More specific, if differences are related to credit opportunities, they find that a Pigovian tax is more effective while accounting ratios are distortionary, but if they are related to risk-taking incentives, liquidity risk is better managed with net funding and liquidity ratios (quantity-based ratios). The strongest implication is that an optimal policy could be either, Pigovian taxes, quantity-based ratios but most likely a combination of both. I capture the idea that reactions among banks to the same type of regulation depend on bank individual characteristics. The authors explore credit opportunities and gambling incentives and how they impact banks behavior to liquidity regulation. In this study, is explored the idea that liquidity condition itself as an individual characteristic, can cause different responses to regulation among Banks. A simple approximation to this proposition is that LCR is truly binding for banks when their liquidity levels are under the requirement. For those above, reactions can be many as long as the level stay above. Thus, it is expected LCR to be effective only for some part of the Banks population, which is aligned with the idea that optimal policy might be reached only through a combination of different nature regulations that considers bank-specific characteristics.

II.5. Hypothesis

All the findings above put together in the same proposition is to make one step forward into the understanding of the effects of the liquidity regulation on Banks behavior, which is the core objective of this study.

Duijm and Wierts (2014) conclude that exists a long-term optimal liquidity level result of opposite forces (risk-profits) where liquidity reaches its optimal, and this level depends on if banks are above or below this equilibrium. Moreover, the adjustment in the balance sheet is made through liabilities side. Bonner Lelyveld and Zymek (2014) conclude that under a liquidity regulation environment, bank-specific and country variables are no determinants of liquidity anymore, just concentration and disclosure. The adjustment in the balance sheet is made through assets side. However, I propose:

General Hypothesis. Under a liquidity regulation scenario, the raised long-term optimal level is substituted by the liquidity level established by law. This means that risk and profits are no longer determinants of the optimal level; however, they remain determinants of the actual liquidity level along with concentration and disclosure. Moreover, the adjustment can be achieved through both sides of the balance sheet, and it will depend if a particular bank is under or above the required level. This general hypothesis can be expressed in the following statements.

Hypothesis 1. The effects of LCR on Bank’s balance sheet composition depends on the current level of liquidity at every moment in time. The reaction to the new circumstances can be made through assets and liabilities on equal terms, given the accounting nature of LCR ratio.

Hypothesis 2. Banks with a liquidity level above (enough) the mandatory level will react driven by profits and risk forces. Relying on the idea that it is possible to take advantage of the new level established trading their “liquidity excess” to something more profitable. I will refer this group of banks as High Liquidity group from now on forwards.

Hypothesis 3. Banks with a liquidity level under (or very close) to the mandatory level, will react driven by system concentration and disclosure forces (market forces) adapting their balance sheet to fully comply with the required level or force to it. I will refer this group of Banks as Low Liquidity group from now on forwards. III. Data I based this study on data from the Mexican banking system mainly for two reasons. First, regulators opted for a heterogeneous implementation of the Basel III liquidity regulation which provides a natural control group. As stated before, this allows following the technical approach/methodology presented in the Bank for International Settlements (BIS) Working paper No. 470 proposed by Benerjee and Mio (2014), which ultimately lead me to answer the research question proposed. Second, it is reasonable to assume that results from this study can be generalized to other countries assessed as compliant with the minimum Basel liquidity standards by the BCBS. According to Regulatory Consistency Assessment Programme (RCAP) Assessment of Basel III LCR regulations (RCAP- LCR) – Mexico (March 2015), the Mexican framework for LCR requirements was issued in December 2014 through the General Provisions on Liquidity Requirements for Commercial Banks. This resolution ensures that all components of the LCR framework deployed by the Mexican government are aligned to the minimum Basel liquidity standard. In other words, Mexican banking regulators were prepared to implement a financial stability regulation according to international standards.

All data collected to conduct the present analysis was made available by the National Banking and Securities Commission (CNBV by its Spanish acronym) which along with the Central Bank, Bank of Mexico (BANXICO by its Spanish acronym) constitute the most important authorities for Mexican banking supervision. The CNBV is a decentralized organism with the objective of supervising all the entities that integrated the Mexican financial system, in order to ensure its stability and proper functioning, as well as pursue

the healthy and balanced development of the system as a whole, having public interest as the first and main priority CNBV(2018).

The statistical information accessible through the CNBV website offers a collection of the financial statements for the Mexican banking system, that can be examined to construct the evolution of each dimension of the balance sheet as well as Banks’ reaction to the introduction on the LCR. I compile a database that contains monthly information of 76 components of Balance Sheet and Results from 2007 to 2017, a 10 years period that permits an historical analysis that includes the financial crisis (2008), the first attempts to harmonize liquidity regulation internationally (2013) and its legal introduction in the Mexican banking system (2015).

III.1. Variables of interest

Understanding the variables selected to conduct the present study is key to understand the potential scope that can be achieved since it depends on the aggregation level of each variable. A standard (completely disaggregated) balance sheet can cover hundreds and sometimes thousands of accounting items and most of them are not public information. From an academic point of view, information at such level could lead researchers to a better understanding of banking behavior. Aggregation level became central because the research question can always be taken one step forward (while the information remains available). For example, if the conclusion is that Bank Debt has reacted to the liquidity regulation, the remain question is what component of Bank Debt is reacting? From the 76 variables collected, the scope of this study is limited to the following 15 general components (bold letters) that can be used to comply given the accounting nature of the LCR regulation, eight of them belong to assets and the rest represent liabilities. Table 1 presents them in the form of a balance sheet to better capture its role and its level of aggregation.

Table 1. Variables of Interest: Balance Sheet Dimensions 1000000 Assets 2000000 Liabilities 1100000 Cash and equivalents 2100000 Retail Deposits 1110000 Cash 2110000 Retail Deposits (Insured) 1120000 Cash in other Banks 2112000 Retail Deposits (Short-term) 1121000 Cash in Central Bank 2210000 Wholesale Deposits 1122000 Cash in other Financial Institutions 2220000 Bank Debt 1123000 Foreign Currency 2221000 Bank Bonds 1130000 Others 2222000 Bank Certificates 1140000 Cash restricted or given as collateral 2223000 Another bank debt 1300000 Securities 2230000 Intrafinancial Borrowing 1310000 Securities for trading 2231000 Deposits from other Banks 1320000 Securities available for sale 2232000 Intrafinancial (Short-term) 1330000 Securities held to maturity 2233000 Intrafinancial (Long-term) 1710000 Total Portfolio (Loans) 2700000 Subordinate Debt issued 1711100 Productive sector 2710000 Contingent convertibles 1711200 Financial Institutions 2720000 Convertible Debt by holder decision 1711300 Government 2730000 Convertible Debt by issuer decision 1712000 Consumption 2740000 No convertible 1713000 Mortgages 3000000 Equity III.2. Implementation of Liquidity regulation: Mexican case Mexico opted for a heterogeneous implementation of the Basel III, publishing the following binding requirements on December 31st_{, 2014 through the General Provisions on Liquidity} Requirements for Commercial Banks. The framework stated the legal levels required as well as the waivers, according to Table 2. The implementation of the LCR according to Mexican regulation provides a clear control group for the year 2015. For the year 2016, a control group is also generated based on Article 5, section first from General Provisions on Liquidity Requirements for Commercial Banks, which states that banks founded after January 1st_{, 2015 with a total credit portfolio} lower to 30 million UDIS, will not be subject to LCR during the first 12 months since starting operations date.

Table 2. Heterogeneous Implementation of the liquidity regulation in Mexico Classification 1Q15 2Q15 3Q15 4Q15 1Q16 2Q16 3Q16 4Q16 1Q17 2Q17 3Q17 Total Portfolio greater or equal to 30 million UDIS* 60% 60% 70% 70% 70% 70% 70% 70% 80% 80% 80%

Total Portfolio lower to 30 million UDIS and 5 or more operation years

NA NA NA NA 60% 60% 70% 70% 70% 70% 80%

Total Portfolio lower to 30 million UDIS and 5 or less operation years

NA NA NA NA NA NA 60% 60% 70% 70% 70%

*UDIS: Unidades de Inversión. Unit of value that takes into account inflation. However, the scope of this study will consider the period 2014/M12 as the PRE-LCR period t (benchmark period), since the liquidity framework was released in December 2014. The final POST-LCR period is H=2015/M06. This is the most important period because it captures the behavior showed by the treatment groups and control group right after the introduction of the LCR. The period that goes between 2007/M01 and 2014/M12 is analyzed through descriptive statistics while the first half of 2015 is the period used to calculate the Average Treatment Effect. III.3. Treatments and Control Group To divide the treatment group between “High Liquidity” and “Low Liquidity”, I calculated the difference between the actual level of liquidity and the legal requirement level at the end of every quarter. I also took into consideration the fact that the level required is lower than the actual 100% stated in Basel III (Table2). In my opinion, the requirements were set in a way all banks already satisfy them, which means some banks were below the 100% proposed in Basel III. a) High liquidity: If the difference between the actual and required level of liquidity is more than 50 percentage points. If the LCR required is 60%, a certain bank will be considered to have high liquidity if its LCR is above 110%.

b) Low liquidity: If the difference between the actual and required level of liquidity is less than 50 percentage points. If the LCR required is 60% a certain bank will be considered to have low liquidity if its LCR is under 110%.

Please note, that the 50 percentage points were chosen taking into account the 100% level of LCR that Banks will eventually reach at some point in the future. The difference can be changed in order to be more or less strict in the classification of “high” and “low” liquidity. The results of this analysis can be summarized in Table 3, that indicates the number of banks for each category. Table 3. Treatment /control group division: Liquidity Level PERIOD 1M15 2M15 3M15 4M15 5M15 6M15 Control Group 38 Low Liquidity 3 High Liquidity 4 Total Banks 45 III.4. A first look at the data: LCR approximation

The first attempt to analyze the reaction of High and Low liquidity groups to liquidity regulation was to measure the HQLA holdings against the required liquidity based on the variables mentioned above, in simple words, an approximation of the LCR. It is reasonable to believe that observing the same measure from 2015 when it was implemented backwards, will provide some conclusions about Banks’ behavior after 2015. To achieve this, I constructed a broad LCR, according to the following: 𝐿𝐶𝑅𝑎𝑝𝑟𝑜𝑥𝑖𝑚𝑎𝑡𝑖𝑜𝑛 = 𝐶𝑎𝑠ℎ & 𝐸𝑞𝑢𝑖𝑣𝑎𝑙𝑒𝑛𝑡𝑠 (𝐻𝑄𝐿𝐴) (𝑅𝑒𝑡𝑎𝑖𝑙 𝐷𝑒𝑝𝑜𝑠𝑖𝑡𝑠 5% + 𝑊ℎ𝑜𝑙𝑒𝑠𝑎𝑙𝑒 𝐷𝑒𝑝𝑜𝑠𝑖𝑡𝑠 75% + 𝐵𝑎𝑛𝑘 𝐷𝑒𝑏𝑡 75% + 𝐼𝑛𝑡𝑟𝑎𝑓𝑖𝑛 𝐵𝑜𝑟𝑟𝑜𝑤𝑖𝑛𝑔 75%)

.6 .8 1 1.2 1.4 1.6 L C R a p ro xi ma tio n 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

Graph 1. LCR approximation: High Liquidity Group

1 1.4 1.8 2.2 L C R a p ro xi ma ti o n 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

High liquidity Banks, after 2008 experienced a constant drop in their HQLA holdings that appear to stabilize in 2013 when Basel III was published, Graph 1 shows an intent to reach the 100% level required by the liquidity framework proposed. However, after 2015 when the LCR became a requirement, HQLA holdings against liquidity needs keep dropping to a 60% in 2017. On the other hand, Low liquidity banks first increase their liquidity holdings until reach 200% after 2008 (that can also be caused by a decline in the liability side). After 2010 cash holdings follow a constant drop that stabilized in 2013 when Basel III was published, a small increase can be observed between 2013-2015 and a slight decline after 2015, when the LCR became binding for the Mexican Banking system. Either case, low liquidity group always kept above 100% during the period of study.

At first glance, conclusions derived from both graphs seem to be contradictory, between banks classified as High Liquidity or Low Liquidity and their actual behavior regarding cash holdings. However, it has an explanation. First, it is important to keep in mind that Banks were classified based on the LCRreal, while the one showed in the graphs is

just a rough construction of the LCR; unfortunately, disaggregation level is not enough to construct a measure closer to LCRreal. Therefore, LCRapproximation only takes into account

liquidity buffers (Cash & Equivalents) in possession of each group against their liquidity needs, while LCRreal is a ratio that takes into account not only Outflows but Inflows in the

next 30 calendar days (Total Net Cashflows). The later has a very strong implication, because it means that High Liquidity group seems to have more future inflows that the low liquidity groups (e.g., consumption loans have monthly payments than other long-term loans).

Second, all the movements described can be explained due to changes in any of the variables that take part in the ratio, that is, I face the same problem that the actual regulation face, the reaction cannot be measured (or control) this way because it depends on many variables.

Nevertheless, a main conclusion can be drawn from this first exercise. LCR ratio was not binding for the High liquidity Banks regarding cash holdings since they kept decreasing cash after the regulation was passed. Low liquidity Banks, kept stable their cash holdings

liquidity banks. Unfortunately, it is very difficult to say at this point if the LCRapproximation was

stable because low liquidity banks augment their cash holdings or because its needs were reduced (e.g., less dependency on wholesale deposits).

I expect that following the technical approach proposed by Benerjee and Mio (2014) help to work out the problems mention above since allows calculating ATE for each dimension of the balance sheet. Differences-in-Differences not only measure the effect but also indicates which of them are not reacting to the introduction of the LCR.

III.5. Descriptive Statistics

Table 4 presents descriptive statistics organized according to the purposes of the methodology, which requires that each dimension of the balance sheet is read as a share of total assets or total liabilities.

Table 4 shows mean and standard deviation for each group and component of the balance sheet subject to analysis. The most important conclusion is Banks in the control group and treatment groups are broadly similar in its balance sheet compositions at this level of aggregation, which allows to calculate the Average Treatment Effect for the establishment of the LCR, using Differences-in-Differences adjusted methodology.

The main difference observed between control and treatment groups is size, but this was already expected due the selection of the treatment group was not random but exclusively based in this factor. For the rest of variables is good news that components in asset side and liabilities size are broadly similar, for some specific dimensions treatment and control group might slightly differ. However, this situation is addressed through the use of conditioning variables, that control for the fact that Banks between the groups might be marginally different among some dimensions of the balance sheet.

Table 4. Descriptive Statistics: Balance Sheet Dimensions (by group) Asset Side Control Group (1) Treatment Group (2) High Liquidity Banks Treatment Group (3) Low Liquidity Banks Total Assets (Ln) 9.588 (1.549) (0.649) 13.002 (0.468) 13.722 Cash and equivalents 0.106 (0.106) (0.038) 0.104 (0.033) 0.109 Securities 0.269 (0.237) (0.114) 0.200 (0.089) 0.253 Total Portfolio (Loans) 0.428 (0.281) (0.144) 0.529 (0.085) 0.480 Productive Sector (Loans) 0.228 (0.209) (0.137) 0.273 (0.020) 0.205 Financial Institutions (Loans) 0.025 (0.032) (0.033) 0.056 (0.023) 0.036 Government (Loans) 0.059 (0.122) 0.046 (0.022) 0.080 (0.034) Consumption (Loans) 0.184 (0.272) (0.038) 0.068 (0.049) 0.083 Mortgages 0.028 (0.055) (0.084) 0.087 (0.021) 0.091 Liabilities Side Total Liabilities (Ln) 9.277 (2.003) 12.842 (0.679) 13.614 (0.477) Retail Deposits 0.389 (0.300) (0.147) 0.540 (0.081) 0.528 Wholesale Deposits 0.124 (0.171) 0.087 (0.129) 0.027 (0.213) Bank Debt 0.032 (0.098) 0.056 (0.093) 0.038 (0.018) Intrafinancial Borrowing 0.146 (0.192) 0.038 (0.017) 0.035 (0.023) Intrafinancial Short-term 0.066 (0.091) 0.030 (0.018) 0.027 (0.020) Intrafinancial Long-term 0.003 (0.016) 0.001 (0.003) 0.0005 (0.0008) Subordinate Debt 0.014 0.009 0.032

IV. Methodology IV.1. Program Evaluation Since the main goal of this study is to measure the effect of regulation, I decide to follow the key steps for Program Evaluation. The core reasoning behind it is to assess the effect of a certain policy, program or treatment. I will refer the implementation of LCR as the policy, regulation or treatment indistinctly. First, imagine two hypothetical situations: 1) Bank “X” is subject to the LCR and 2) Bank “X” is not subject to it. For each of these situations a potential outcome is expected. For Bank “X” in particular, the causal effect would be the difference between the outcome when is subject to the policy and the outcome when is not. However, there is no way to measure the causal effect for a particular bank since the regulation applies to it or not, also known as the counterfactual problem.

For the purposes of program evaluation, measure the mean causal effect it is considered to be enough to derived conclusions about the policy, Stock and Watson (2015). That is, given a population of Banks it is achievable to measure the average causal effect or the average treatment effect (ATE). In this study, ATE refers to the difference in the average outcomes for those banks who were subject to the LCR and the average outcomes for the waivers. IV.2. Local Projection Method and Empirical Strategy Mexican authorities opted for a heterogeneous implementation of the Basel III liquidity regulation which provides a natural control group (waivers). I followed the technical approach presented by Benerjee and Mio (2014): Local Projections Method.

The local projection method was proposed by Jordà (2005). The fundamental idea is to generate local projections for each period ahead instead of inferring distant periods as in other models (e.g., vector autoregressions VAR). That is, to estimate consecutive regressions of the variable of interest as many periods as we want to observe. Based on this method, the authors took different dimensions of the banks’ balance sheet as variables of

interest and computed Difference-in-Difference (DiD) estimators producing a “pre” and “post” information for each period ahead (h-step).

Following the idea, Local Projections Method require to collapse the time series information into one Pre-LCR period and one Post-LCR period for h-step ahead. Bertrand, Duflo and Mullainathan (2004), question the validity of Differences in Differences estimators because they found that conventional DiD methodology usually generates inconsistent standard errors. By using Monte Carlo simulations, they study different techniques to address this problem, and they found that removing the time series dimension through collapsing the information into one pre- and one post- intervention period is a simple technique that performs well solving this problem, especially when the number of entities is relatively small (e.g., ten entities). Taking into account that the number of banks in the Mexican Banking system is relatively small (45 banks during the study period), this technique fits perfectly to calculate the Average Treatment Effect through Differences-in-Difference. Also, it is the reason why Benerjee and Mio (2014) call it adjusted DiD estimates.

Finally, the empirical strategy is to estimate, one by one, the average treatment effect (ATE) of the LCR in every component of the balance sheet that can be adjusted to comply with the requirement. The main addition to this approach is the creation of two treatment groups of Banks: High liquidity and Low liquidity, in order to determine if the reaction to the regulation between the groups is different, as proposed in the hypothesis. IV.3. Estimation Model

• Let LCRi take value 1 to reflects if a bank is subject to liquidity regulation or 0

otherwise, d0 and d1.

• Let Yi be the variable of interest. It can be any component of the Banks’ balance

sheet that it wants to be observed after the regulation was implemented (e.g., loans to the productive sector, short-term funding).

• For each individual bank, the potential outcome is defined as Yi,t+h (dj) – Yi,t. That is

the observed outcome for all possible djÎD,(j=0,1). Zero if the bank had to comply

and 1 if it has not.

• The variable Yi,t+h – Yi,t represents the cumulative change from t to t+h. In the context

of the study, Yi,t should be treated as a benchmark, and it is observed before the

implementation of the LCR.

• The causal effect of the LCR can be write as [Yi,t+h(d1) – Yi,t] – [Yi,t+h(d0) – Yi,t], yet it

should be understood as a random variable that cannot be observed. Therefore, we estimate the Average Treatment Effect.

• Let 𝑌 UVW,YZ[\]Z be the sample average of the variable of interest for those banks subject to the liquidity regulation before it was implemented.

• Let 𝑌 UVW,^[_Z] (`ab_Zc) be the sample average of the same variable for those banks subject to LCR, for each period ahead after it was implemented.

• Let 𝑌 d\e_]\f,YZ[\]Z and 𝑌 d\e_]\f,^[_Z] (`ab_Zc), be the sample average of the variable for the control group before and after the LCR introduction.

• The average change in Yi over the implementation for those banks subject to LCR is

𝑌 UVW,^[_Z]− 𝑌 UVW,YZ[\]Z .

• The average change in Yi over the same period for those in the control group is

𝑌 d\e_]\f,^[_Z]− 𝑌d\e_]\f,YZ[\]Z .

• Finally, the DiD estimator is the average change in the variable of interest for those in the treatment group minus the average change for those in the control group.

𝛽_jklk _{= 𝑌} UVW,^[_Z]_{− 𝑌}UVW,YZ[\]Z _{− 𝑌} d\e_]\f,^[_Z] _{− 𝑌}d\e_]\f,YZ[\]Z

= ∆𝑌UVW _{− ∆𝑌}d\e_]\f

• The differences-in-differences estimator can be written in regression notation. 𝑌<sub>l,_n`</sub>− 𝑌<sub>l,_</sub> = 𝛽<sub>o</sub>`_{+ 𝛽}

Xi is defined as control variables that could be relevant in explaining the LCRi variable. This

set of variables should include those to control for the fact that waivers were not chosen randomly (e.g., size). Reason to take into account the conditional independence assumption: 𝑦_{l,_n`</sub> 𝑑<sub>q</sub> − 𝑦<sub>l,_</sub> ⊥ 𝐿𝐶𝑅<sub>l</sub>|𝑋<sub>l,_</sub> 𝑓𝑜𝑟 𝑎𝑙𝑙 ℎ³ 0 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑑<sub>q</sub> Assuming the conditional mean can be linearly approximated. The DiD estimator, can be estimated using OLS regression as follows: 𝑌<sub>l,_n`}− 𝑌_{l,_} = 𝛽_o`<sub>+ 𝛽</sub> j`𝐿𝐶𝑅l + 𝛾`´𝑋l,_+ 𝑢l 𝑓𝑜𝑟 ℎ = 0,1, … , 𝐻 IV.4. Selection of Control variables

As mentioned before, selecting control variables is a necessary step to generate an appropriate model because LCR compliers were not randomly determined, rather focused in bank type. According to Mexican regulation, just one characteristic defined if a certain bank had to comply or was waiver: Loans portfolio size. Authorities decide to start with bigger banks; therefore, portfolio size is the natural control variable that captures better preexisting differences between the control group and the treatment groups.

The division above might seem simple, yet can imply many other differences between banks characteristics. This is a crucial step because control variables can be obtained from bank-specific characteristics, country-specific variables or differences between one period or another. Therefore, a wide knowledge of the Mexican banking system and its role in the national economy play a very important part.

Following Benerjee and Mio (2014) I limited the control variables to other banks’ characteristics and constructed different models using diverse combinations of them. To decide which model contained the set of control variables that better captures preexisting conditions between control and treatment groups I run probit regressions and use the

A probit model is a nonlinear regression especially designed for binary variables that allows determining the probability of the binary variable taking the value of 1 given the independent variables, Stock and Watson (2015). The constructed models represent which variables are better associated with the fact that a specific bank was complier (LCR = 1). Finally, the model that contains the best set of control variables is the one with the smallest AIC or BIC information criterion. Tables 5 and 6 present the results for the best five models constructed for High liquidity and Low liquidity treatment group respectively. Table 5. Probit regressions of DLCR for High Liquidity Treatment group Control Variables (1) DLCR DLCR (2) DLCR (3) DLCR (4) DLCR (5) Total Portfolio/TA 0.428*** 1.078*** 0.499*** 3.165*** 3.532*** (0.0507) (0.0766) (0.0840) (0.218) (0.253) Capital/TL -4.193*** -5.011*** -5.431*** -5.971*** (0.249) (0.333) (0.483) (0.514) Retail Deposits/TL 1.257*** -0.178* -0.299** (0.0702) (0.107) (0.126) Intrafinancial Borrowing/TL -12.30*** -15.32*** (0.719) (0.877) Securities/TA -1.088*** (0.130) Constant -1.537*** -1.216*** -1.430*** -1.134*** -0.677*** (0.0333) (0.0396) (0.0458) (0.0594) (0.0873) AIC 2,653.982 2,431.272 2,327.015 1,933.116 1,800.023 BIC 2,666.769 2,450.453 2,352.435 1,964.264 1,836.912 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Based on the descriptive statistics I first focus on the variables that prove to have some differences between the High liquidity group and the control group. From all the models generated I present the five more representative since only contained statistically significant variables, an indicator that they have an impact in the LCR classification. In simple words, the probability of a bank being a complier (LCR=1) is more related to these variables and therefore can be used as control variables. Using AIC criterion as guidance, I kept model DLCR (5) as the one that contains the best group of control variables. Probit models can tell

much more, however, its application here is limited to find the best combination of control variables.

The same steps were applied to the Low liquidity group. I constructed diverse models based on the differences between treatment (2) and control group. Surprisingly the probit models prove that control variables differ from the High Liquidity group. Table 6 contains the five more representative models and shows that for the second treatment group adding more control variables is not the best strategy. Actually, while keep adding control variables these became not statistically significant. Based on these results and the AIC criterion I decide to keep model number (4), omitting retail deposits and intra-financial from the optimal model since they are not statistically significant, that is, they are not relevant explaining liquidity regulation for low liquidity treatment group, only portfolio size and capital. Table 6. Probit regressions of DLCR for Low Liquidity Treatment group Control Variables (1) DLCR DLCR (2) DLCR (3) DLCR (4) DLCR (5) Total Portfolio/TA 0.847*** 1.059*** 0.729*** 2.479*** 1.344*** (0.0664) (0.0716) (0.0847) (0.153) (0.0878) Capital/TL -1.056*** -0.899*** -0.802*** -0.485*** (0.0766) (0.0715) (0.118) (0.0801) Retail Deposits/TL 0.974*** 0.139 (0.0710) (0.0989) Intrafinancial Borrowing/TL -10.43 (0.480) Wholesale Deposits/TL -1.586*** (0.204) Constant -1.599*** -1.473*** -1.783*** -1.482*** -1.391*** (0.0385) (0.0419) (0.0462) (0.0510) (0.0482) AIC 3,195.86 3,103.05 2,989.44 2,385.32 2,769.85 BIC 3,208.70 3,122.32 3,014.98 2,416.64 2,794.36 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

From the chosen models, it can be derived that results reinforce the idea that liquidity condition, capture by different components of the balance sheet, lead banks to react differently to liquidity regulation. For the high liquidity group, the optimal set of

Deposits, intra-financial borrowing and securities access, allowing them to follow different paths to comply with the regulation. I conclude this section mention that Mexican authorities could have considered that targeting bigger banks not necessary means that they are homogeneous, thus require them the same general rule could cause different responses. Taking this idea one step forward is expected that banks will approach the benchmark based on its liquidity condition, either by reducing its liquidity levels or by increasing them. The remain question is through which channels. To address this point, I next present for each dimension of the balance sheet chosen, the Average Treatment Effect of the introduction of liquidity regulation, using adjusted Differences-in-Differences. Separately for High liquidity treatment group and Low liquidity treatment group. V. Baseline Results The Mexican design for the introduction of LCR allows differentiating several periods, each of them implies different treatment and control groups and/or changed required liquidity level, as can be observed in Table 2. I focused on the first period that covers from the introduction of the LCR on December 2014 until June 2015. During this time circumstances remain unchanged for the treatment and control groups based on the regulation design. Therefore, the benchmark period or pre-policy correspond to t = 2014M12, from which the cumulative change is measured, until the post-policy period H = 2015M06.

For robustness check section the benchmark period (t) will remain the same, but the h-step ahead period will be extended to H = 2015M12, since the required level for the treatment group changed from 60% to 70% between July 2015 until December 2015, but the treatment and control groups remained the same.

To interpret the results is essential to keep in mind that the variables examined represent the most important aggregate components of the balance sheet and each of it is measured as a share of total assets or total liabilities. This technique admits one extra check for the results obtained because an adjustment (up or down) in the asset side of the balance sheet must be offset by a movement in the liabilities side. From basic accounting rules is

expected that all the movements together (assets/liabilities) must sum zero, which can be read as an indicator of consistency throughout the results.

Additionally, the average treatment effect is calculated separately for each component of interest. That is, for each treatment group I run 14 separately regressions. It can be already assumed that not all components of the balance sheet change as a response to the liquidity regulation, indeed for those components the regression will show a non-statistically significant coefficient for the interaction dummy LCR. For simplicity, the following tables present only the dimensions that have been impacted by the introduction of the LCR, making straightforward the visual and numerical interpretation of the results. However, Appendix A and B show the fourteen regressions for each group. The process was conducted in the same way for High liquidity and Low liquidity treatment groups. V.1. High Liquidity treatment group Table 7. Impact of Liquidity Coverage Ratio on Balance Sheet composition: High Liquidity Group Assets Side Liabilities Side (1) Cash & Equivalents (2) Financial Institutions (3) Consumption (4) Mortgages (5) Wholesale Deposits LCRInteraction -0.036** 0.030** -0.141** 0.054* -0.130*** (0.016) (0.014) (0.059) (0.031) (0.045) TotalPortfolio/TA -0.073 0.032 0.458* -0.019 0.330*** (0.051) (0.032) (0.262) (0.051) (0.101) Capital/TL -0.000* -0.000 -0.000 -0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) Retail/TA 0.044 0.019 0.024 0.093 -0.165* (0.048) (0.026) (0.224) (0.062) (0.095) IntrafinBorrw/TA -0.019 -0.019 -0.151 0.116 -0.207 (0.042) (0.054) (0.286) (0.093) (0.150) Securities/TA -0.042 -0.015 0.069 -0.020 -0.027 (0.067) (0.024) (0.229) (0.064) (0.101) Constant 0.138*** 0.011 -0.055 -0.005 0.097** (0.040) (0.014) (0.107) (0.026) (0.046) Observations 42 37 39 31 40 R-squared 0.139 0.290 0.232 0.221 0.270

The High Liquidity group reacted to five of the fourteen dimensions chosen, and it can be observed in Table 7 that adjustments occurred mainly in the asset side of the balance sheet, contrary to the results presented by Duijm and Wierts (2014) and Benerjee and Mio (2014). The first row presents estimates of the Average Treatment Effect of the LCR for each component of the balance sheet.

The first column (1) indicates that High Liquidity Banks decrease its Cash & Equivalents by 3.6 percentage points after the introduction of the LCR. This movement is almost offset by an increase in loans to Financial Institutions by 3 percentage points, as it shows the second column. The 0.6 percentage points difference can be accounted for the increase in Mortgages by 5 percentage points in column (4).

The mentioned adjustments, clearly indicate that banks with a High Liquidity condition substitute their own assessment of liquidity risk with the benchmark stated in the regulation, which allows them to transfer the “extra” liquidity they hold (compared to the requirement) to more profitable assets. In the Mexican banking case, they are moving cash and equivalents to loans to financial institutions and Mortgages.

Surprisingly, they also react reducing its dependence to Wholesale Deposits by 13 percentage points in the liabilities side, adjustment that is offset in the asset side by a reduction to Consumption loans (credit cards, automobile and home appliances credits, etc.) by 14.1 percentage points according to column (3). The difference between these movements can also be accounted for the increase in Mortgages. This behavior was not expected since the group belongs to the High Liquidity group. My interpretation about this movement is that High Liquidity Banks are taking advantage of the partial introduction of liquidity regulation to make a smooth transition to fully comply with the whole framework, that is not only about LCR but the Net Stable Funding Ratio (NSFR), which is focused in maturity mismatch. In my opinion, they are slowly adjusting maturity mismatch at the same time they compensate the reduction in wholesale deposits and consumption loans by transferring the cash and equivalents available to more profitable assets.

In some good news, the introduction of the LCR did not have a direct impact on the productive sector as did not have an impact on any other funding source for this group. As a final remark, the 5 percentage points increase in Mortgages it is only statistically significant at the 10% level, and only 2 percentage points can be explained consistently with the results overall. It can be an indication that not all the increment is associated with the introduction of LCR. V.2. Low Liquidity treatment group Table 8. Impact of Liquidity Coverage Ratio on Balance Sheet composition: Low Liquidity Group Assets Side Liabilities Side (1) Consump tion (2) Mortgages Retail (3) Deposits (4) Wholesale Deposit (5) Bank Debt Intrafinancial (6) Funding LCRInteraction -0.120** 0.048*** 0.123** -0.121*** 0.041*** -0.154*** (0.047) (0.012) (0.053) (0.028) (0.006) (0.041) Total Portfolio/TA 0.484*** 0.061* 0.356** 0.198*** 0.013 0.438*** (0.169) (0.035) (0.169) (0.071) (0.021) (0.134) Capital/TL -0.000 -0.000 -0.001*** 0.000*** -0.000 -0.000* (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Constant -0.044 0.009 0.226*** 0.051* 0.006 -0.023 (0.046) (0.013) (0.068) (0.027) (0.009) (0.040) Observations 38 30 42 38 15 42 R-squared 0.257 0.189 0.177 0.182 0.344 0.318

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

The Low Liquidity group reacted to six of the fourteen dimensions chosen. The first-row present estimates of the Average Treatment Effect of the LCR for each component of the balance sheet.

It is remarkable that the introduction of the regulation did not impact Cash and equivalents dimension, considering that it is its main purpose. The interpretation of the situation is that once they disclose its liquidity levels, has been more difficult or expensive

are primarily made through the liabilities side, aligned to the results of Duijm and Wierts (2014) and Benerjee and Mio (2014) as can be observed in Table 8.

Findings prove that, under a Low Liquidity condition, Banks will react as to the introduction of LCR as following: Column (3) and (4) show that wholesale deposits have been replaced by retail deposits in 12 percentage points, this implies a transition to more stable funding source since retail deposits are insured. Column (6) and (1) shows that Intrafinancial borrowing decrease in 15 percentage points which ultimately led to a 12 percentage points reduction in loans to Consumption in the asset side. Finally, Bank Debt in column (5) indicates an increment of only 4 percentage points that can be translated to an increase in Mortgages (2) by the same amount in the asset side.

The behavior of low liquidity banks group can be explained from two perspectives. First, they are willing to make all the adjustments since they have to comply with the requirement one way or another, they know that the legally required amount will be increased until reach its original 100% binding level. Second, as a consequence of the regulation, this group was subject to market response to disclosure and concentration. Low Liquidity group represents the biggest banks in Mexico that had to disclose its low liquidity levels, causing the market to decide to redirect the resources to other destinations, forcing them to replace less stable funding with more stable one and even reduce their lending to consumption. Either case, the results show a reaction of low liquidity banks to the introduction of Liquidity Coverage Ratio as expected, which is consistent with previous studies. Another notable reaction is that the reduction in lending was not targeted to the productive sector. However, considering that Mexican economy relies importantly on credit to consumption, this situation could have an indirect effect on the productive sector and could constitute a new topic to further study.

In summary, for banks with a high liquidity condition, liquidity regulation allows them to adapt the new circumstances by stop assessing its own liquidity risk and trade the “extra” liquidity holdings to more profitable assets. For banks with low liquidity condition, regulation effectively provides banks with more resilience to survive a stress scenario of 30

days but not by increasing its liquidity holdings but by reducing its outflows (adjusts through liabilities side) which ultimately could represent a negative externality to the real economy. In the general context, an accounting ratio like LCR that does not take into account entity level characteristics (e.g., liquidity conditions), can solve some problems for a sub-group of the banking system while creating others. VI. Robustness Check Baseline results introduced the first reaction of both treatment groups to the establishment of LCR for the first period, according to the General Provisions on Liquidity Requirements for Commercial Banks. During the first semester of 2015 conditions remained the same in terms of groups and liquidity level required. However, for the second semester, the required LCR increase from 60% to 70%, while the treatment and control group were not altered. This change in the regulatory level allows to make a robustness check through the extension of the h-step period ahead for the second semester of 2015 and observe if the behavior follows the same trend found before. For the robustness check the benchmark period do not move t = 2014M12, while the targeted period is extended to H = 2015M12. Results for High and Low liquidity are presented in Table 9 and Table 10 respectively.

After the required level increased from 60% to 70%, it can be observed in Table 9 that Mortgages is no longer a dimension that responds to LCR, confirming the suspicions rise during the first period about its significance in the future. For the rest of variables presented in Table 7 the effect remains roughly the same, with consistent direction and a similar degree of reaction. Cash and equivalents decreased by 3.6 percentage points almost offset by the increase in 2.8 percentage points to loans for Financial institutions (2). Wholesale deposits in the liabilities side decrease by 12.7 percentage points which are translated completely to 14.4 percentage points decline in loans to Consumption. In general, High liquidity group follows the same tendency over 2015, even considering the fact that the required level increased by 10% in July 2015.