Master International Finance (Msc)
Master Thesis
The effect of the ECB deposit facility interest rate on European banks’ stock returns
Miles Ashton
August 2015
Under the supervision of Razvan Vlahu
Abstract:
This research investigates the impact of fluctuations in the European Central Bank deposit facility rate on the stock returns of Euro area banks spanning between Q12009 and Q12015. The analysis uses a random-effects panel regression model that encompasses 17 European banks using data on 4 key bank-level financial metrics, in addition to various macro-economic variables and the ECB deposit facility rate. The findings indicate that a 1% increase in the ECB deposit facility rate incurs to a 0.196 (3sf) decrease in the percentage change of banks’ stock returns across the sample pool.
Acknowledgements
With pleasure I would like to acknowledge the support that has been provided by Razvan Vlahu. His efforts and input in both an academic and experienced based context have been invaluable and they are much appreciated. Thanks must also be extended to the University of Amsterdam, Amsterdam Business School for their efforts in facilitating this master thesis.
Table of Contents
1 Introduction ... 5
2 Literature Review and Underlying Theory ... 7
3 Data and Methodology ... 14
3.1 Methodology ... 19
3.2 Regression Model ... 20
4 Empirical Results ... 22
4.1 Macro-Economic Regressions ... 22
4.2 Bank-Level Financial Metrics Regressions ... 23
4.3 Complete Regression Testing ... 25
5 Analysis of Results and Discussion ... 28
6 Conclusion ... 32
1
Introduction
The consequences of the 2007-2008 global financial crisis still persist with many major European economies experiencing stagnation, little growth or even deflation as well as lack-lustre lending activity. Central banks have had to action bold, timely and at times unorthodox response to the ensuing issues. The European Central Bank (herein after ECB) determines the monetary policy rates that govern European Union (herein after EU) banks. The deposit facility interest rate is the return earned by Euro area banks when they deposit their money with the ECB, typically overnight. This also acts as a high-level indicator of economic conditions in Europe. It is not unusual for this interest rate to fluctuate; however it does have significant implications for banks and economies as a whole. The core principal is that the ECB reduces the deposit facility interest rate when economies are suffering hardship and low growth. Many of the large European economies are experiencing deflation, where prices of consumer products and commodities fall. This is typical of low growth economic conditions. Strategies to reduce monetary policy rates are implemented to prevent banks hoarding excessive amounts of cash and to make borrowing cheaper for businesses and households, driving demand for loans (Bloomberg QuickTake, 2015). The idea behind this is to reinvigorate the economy as other options have been exhausted.
In June 2014, the ECB took a step into unchartered territory and introduced a negative deposit rate, which reached -0.2% by September 2014. In effect this punished Euro area banks for holding onto their cash. This was widely considered as an aggressive strategy to make banks lend more. Arguably the motivation behind the move is justified, and necessary to maintain liquid markets and promote financial stability in capital markets. Commentators argue that consistently low monetary rates increase banks’ appetite for credit and liquidity risk (Maddaloni and Peydró 2011), which increases the likelihood of financial distress by accruing bank risk1. Clearly then, the topic is diverse with widespread impacts to the financial services, businesses and consumers. There is a growing concentration of research on the effect of ECB monetary rates on banks. The main focus of previous academic research is the consequence of monetary rates on bank risk-taking, loan margins and commercial banks’ stock returns, which
1 See Jiménez et al. (2011) and Maddaloni and Peydró (2011) for more arguments that link monetary policy with
provides a well-founded framework for my own research. My research takes a particular interest in the effect of fluctuations in ECB deposit facility interest rates on EU banks’ stock returns. This is a very current topic that is deeply intertwined with the EU financial crisis recovery process, monetary policy and the import role that banks play in the road to economic revival. This thesis is yet to acknowledge any methodologies that have tackled the effect of negative deposit facility rates. This is because the move into negative territory is unprecedented and new, with the repercussions not fully understood. As such, the recent negative rates will be a focal event for my research in order to improve understanding about their effect on banks.
As a result of preliminary research and in conjunction with an intuitive approach, I anticipate that Euro area banks’ stock returns will prove sensitive to changes in the ECB deposit facility interest rate. I expect the recent decrease in deposit facility interest rates to negatively impact stock returns because banks will earn less interest with the ECB, and cannot immediately counter this by increasing interest rates on loans due to their asset-liability gap structure. In addition, I expect negative interest rates to dent bank performance more aggressively than interest rates above zero.
The aim of the investigation is to provide a better understanding of the impact of changing deposit facility interest rates and specifically whether negative interest rates affect returns to their shareholders. This will be of particular use to various departments of Euro area banks, such as the bank’s treasury, risk management, policy implementers and regulators to name but a few. The consequences for European banks remain rather speculative because of the short period of time that negative monetary policy has been imposed, and this is where this thesis aims to contribute to bridging the gap.
2
Literature Review and Underlying Theory
The foundation of this research is grounded on the premise that international monetary policy is designed with the objective to facilitate long-term, non-inflationary and stable growth (Borio 2014). The ECB governs monetary policy across the Euro Area by constructing a system of financial and economic rules that must be adhered to by Euro Area banks and companies alike. It also plays a pinnacle role in determining key interest rates, which have a direct effect on the banking and financial system, in addition to other important guidelines such as Basel III for capital requirements for banks. In essence, the financial system is under-pinned by the monetary policy that the ECB determines and regulates. However, Dell’Ariccia et al. (2010) examines how monetary policy has been rigorously scrutinised after the recent global financial crisis and blamed for being too loose which permitted credit booms and allowed for global recession. Spectators argues that at the onset of the global financial crisis if monetary policy regulators had aggressively raised interest rates, the ensuing financial damage would have been lessened. Clearly then, monetary policy is a very current topic that is deeply intertwined with the financial system and the economic landscape. There are various avenues to explore within monetary policy, but my own research is focused on the relationship between monetary policy and bank stock returns.
In the advantage of this research study, there is an abundance of literature surrounding the topic of interest rate changes, including monetary rate changes on bank stock returns and other bank metrics such as non-performing-loans, asset liquidity and credit supply to name a few. The aim of this literature review is to gain a better understanding of the sensitivity of banks’ stock returns to changes in monetary policy rates through investigating the academic work of others who have focused on this general topic. The literature review will help focus the scope of this research study and foster the techniques that will be used in the empirical study. I acknowledge that some of the reviewed literature is dated, as far as back 1984 and are not representative of today’s financial and monetary landscape like negative deposit facility rates. However their work is still valuable in investigating the effect of monetary rate fluctuations and learning from previous methods applied to this type of research.
Flannery and James (1984) performed a similar study to assess the impact of interest rate changes on commercial banks’ stock returns, and highlighted a number of exogenous variables that measure the sensitivity of commercial banks’ returns to changes in interest rates. The data analysis was carried out on a pool of 67 actively traded commercial banks and stock saving and loan associations (S&Ls). Their findings show that common stock returns of these commercial banks are highly correlated with interest rate changes. In addition, they find that the co-movement of a bank’s stock returns with interest rate changes are positively correlated to the size of the maturity difference between a bank’s nominal assets and liabilities. Asset-liability mismatch and asset maturity were two of the variables that Flannery and James used to determine the sensitivity of bank returns to changes in interest rates. The empirical methodology used in Flannery and James’ paper may be of some use to this study, however less attention will be given to the maturity composition of a bank’s balance sheet rather than being a focal interest.
Kwan (1991) used a random coefficient two-index regression model, which controls for the time-varying interest rate sensitivity of commercial banks’ stock returns caused by banks’ changing maturity profile. This research paper has many parallels to that of Flannery and James (1984), and concurs in principle that commercial banks’ stock returns are significantly interest rate sensitive. Similarly, Kwan’s random coefficient model finds a positive correlation between interest rate changes and the maturity mismatch between a bank’s assets and liabilities, but instead as a result of unanticipated interest rate shocks. In addition to these findings, the paper investigates whether these results are consistent in the long-term (longer than 1 year). There is strong evidence to suggest that commercial banks’ stock returns are more sensitive to interest rate fluctuations in the long-term. However, because of the limited number of regressors in the model the results indicate that there are unexplained variables other than duration mismatch unrelated to balance sheet composition that are creating the interest rate sensitivity. Kwan reasons that this is because in the long-term, the maturity of the applied interest rate index has to ‘match’ the duration mismatch in the same maturity category. As such, the result indicates that analysis based over a longer time period must use a more complex multi-factor regression model to incorporate more bank variables, beyond the scope of trivial balance sheet data.
Continuing the theme of regression analyses, Elyasiani and Mansur (1998) utilise a generalised autoregressive conditionally hetroskedastic in the mean (GARCH-M) method to scrutinise bank stock return distribution to changes in the level and volatility of interest rates. Their method entails grouping a pool of 56 commercial banks that are traded mainly on the NYSE in to the following categories, Money Center Banks (10 banks), Large Bank Portfolio (14 banks), and Regional Bank Portfolio (32 banks), based on balance sheet magnitude and portfolio size. This has the advantage of comparing like-for-like banks in addition to smoothing out noise in the data, a useful technique for my own empirical research. Elyasiani and Mansur comment that banks in an effort to reduce their interest rate exposure, one of their most prominent risks, have tried to reduce their duration gaps, taken positions in derivative products and switched to off-balance sheet activities where possible. Despite the measures taken by commercial banks, the research illustrates that long-term interest rates have a significant negative impact on stock returns. In the case of Money Center Banks and Large Bank Portfolio banks, interest rate volatility is found to be an important determinant of stock volatility and bank stock risk premia. And as such, when interest rates are more volatile the risk premium for Money Center Banks and Large Bank Portfolio banks will increase with them in tandem.
Elyasiani and Mansur (2004) furthered their previous work using a multivariate GARCH model to investigate the relative sensitivities of bank stock return distribution to the short-term and long-short-term interest rates and their respective volatilities. They hypothesised that bank stock returns are equally sensitive to short-term and long-term interest rate changes, which is a potential sub-question for my own research. Their findings suggest that bank stock returns are also interest rate sensitive, the direction of the effect is negative, and the magnitude of interest rate sensitivity is portfolio specific and regression model-dependant. In response to the hypothesis, they find that bank stock returns are more sensitive to interest rate fluctuations in the long-term. In addition, interest rate volatility in the short-term and long-term is an important factor in bank equity returns and their volatility. From this research, I have learnt that it is important from an analytical perspective to use an appropriate interest rate variable in the bank asset-pricing model because interest rate risk, market risk and stock return volatility are sensitive to this choice.
Now I will touch upon the first inclusion of an event(s) study in conjunction with a multiple regression, performed by Ghazanfari et al. (2006). The short analysis period spanning 1998 to 2002 encompassed two distinct Federal Reserve interest rate events, 30th June 1999 and 3rd January 2001, an interest rate hike and an interest rate cut respectively. Their work included two stages of analysis, an event(s) study and a multiple regression using specific financial ratios that serve as predictors of interest rate sensitivity: equity-to-total assets, interest income-to-net revenue, deposits-to-total assets; and portfolio securities-to-total assets. The underlying theory is that a bank’s exposure to interest rate risk is partly due to a difference in a bank’s interest income and interest expense resulting from regular re-pricing of short-term assets. The change in interest margin has a knock-on effect on share price. Ghazanfari finds that the interest rate hike during the first event had more statistical significance than that of the interest rate cut in the second event. In conjunction with this, when Federal rates were increased, it was found that portfolio securities-to-total assets had a significant positive relationship to interest rate sensitivity. Finally, when Federal rates were discounted, equity-to-total assets (an indicator of a bank’s financial leverage) had a significant relationship with bank’s stock return sensitivity to interest rates. The work of Ghazanfari et al. (2006) will be useful for my own study because I anticipate using similar financial ratios as control variables in my own regression model. In addition, I will use a defined study period to focus the data analysis, which will include the particularly controversial ECB decision to introduce negative deposit facility rates.
At this time it appears valuable to strengthen understanding of a bank’s more apparent stock return drivers. Castrén et al. (2006) first outline the importance of a bank’s equity price and the responsibility of the dividend-discount model to reflect as much market information as possible in a single number. The paper concentrates on bank-level financial metrics to explore the volatility in banks’ equity prices, log excess returns, return on equity (RoE), log average and log book-to-market ratio. In conjunction with these metrics, the dividend-discount model is used with an auto-regression framework that illustrates; in the short-term, momentum of past returns and historic levels of leverage predominantly drive equity returns. In the long-term, equity returns are impacted by market shocks that effect cash flow. Castrén et al. (2006) find that the positive covariance between return news components indicated that the market tends to underreact to positive news on bank-specific fundamentals and instead only gradually incorporates market news into equity prices. This result is most apparent for
bigger banks, which appear more prone to market-wide shocks. In essence, Castrén et al. (2006) study reminds us of the vast scope of this topic and demonstrates to me the importance of controlling for bank-level metrics due to their hand-in-hand relationship with bank performance.
The ECB monthly bulletin (2010) explains the importance of the ECB’s interest rate strategy specifically in response to the global financial crisis, which again aids our understanding of the relationship between monetary rates and bank performance. Immediately after the financial market’s collapse in 2007-2008, central banks and governments had to respond decisively to reduce systematic risk, maintain liquid markets and price stability. When there are changes in key monetary rates, the resulting effects are widespread and relevant to banks, businesses and households among others. Once the ECB determines interest rates, they are implemented by allotting the required amount of liquidity needed by the banking sector to meet the demand resulting from so-called autonomous2 factors and to fulfil the reserve requirements. By allowing banks to conform to the reserve requirements3 over a monthly period, the minimum reserve4 system ensures that the overnight money market rate mirrors the official interest rate. Consequently the effects of the ECB’s interest rate strategies are transmitted to the financial markets and, with lags, to the real economy.
Armed with a deeper understanding of the implications of monetary rates and their wider consequences, I turn to Tahmoorespour and Ardekani (2012) and their undertaking of a study to analyse the effect of interest rates on banks in 14 international markets. Their research aims to examine the impacts of interest rates with respect to 7 bank-level financial ratios, while controlling for bank size. The technique of controlling for an independent variable provides valuable lessons for my own regression approach. Their ordinary least squares (OLS) regression across international markets illustrates that the behaviour of banks towards interest rate fluctuations is highly dependent on the market that the bank operates in. In only 5 out of 14 markets banks’ size was a significant variable impacting bank returns. The financial ratios incorporated into the regression can be considered collectively as a comprehensive measure of
2 Autonomous factors = sum of banknotes in circulation + government deposits – foreign asset + other factors 3
See Basel III capital requirements for banking sector, particularly Pillar 1 requirements regarding Quality and Level of Capital, and Capital Conservation Buffer
bank performance and are ratios that I should use as well. The ratios used: net interest/ revenues, common shareholders’ equity/ total assets, net loans/ total deposits, total debt/ common shareholders’ equity, net loans/ total assets, provision for loan losses/ net loans, and RoE. From my own perspective, it is interesting to see that provision for loan losses/ net loans has been embedded in the regression, which prompts me to include size of non-performing loans (NPLs) in to my own regression as a measure of bank risk. Dia (2013) analyses the banking model and how it responds to monetary policy shocks with a particular focus on the size of a bank’s loan book. The results indicate that banks increase the size of their portfolios by lending more when there are positive interest rate shocks; they are more profitable and generate new liquidity. Negative interest rate shocks cause a scaling back of the loan book because the expected risk of default increases. During these times the bank reacts by increasing interest on loans, however they do not pass on all of the increased cost of the loan. In this way a bank can smooth-out interest rate fluctuations by providing insurance against monetary policy shocks. This is interesting because there is typically a connection between monetary rate changes, bank loan books and risk-taking, which provides another reason to incorporate a NPL metric into the regression if consistent data permits me to do so.
To summarise, the argument for decreasing the deposit facility rate is to reduce borrowing costs for companies and households in order to encourage lending and drive business. However, if rates remain low over a prolonged period of time there is a high risk of disrupting the money markets that financial institutions use to fund themselves. Banks may offset their loss in earnings from low deposit rates by increasing interest rates on their loans, however this involves a complex change in a bank’s asset-liability gap, which is a difficult and lengthy process.
The literature review has been useful for a number of reasons. The exercise has provided a comprehensive overview of the topic, increased focus and concentration on particular areas of the research, as well as providing helpful guidance for the empirical techniques.
Despite some conflict between academic papers, it would appear that the majority of research in this area finds that banks’ stock returns are sensitive to the interest rate environment. These results have been produced by various regression techniques, with the differences between
them found in the variety of independent variables used in the regression expression. This indicates that in spite of some common ground between the different techniques, conclusions drawn from the regressions are model dependent. As a result, my own findings will be specifically geared towards the independent variables that construct the regression expression, which will be explained in the following sections. The literature review has also shown that in general, interest rate fluctuations have a greater impact on stock returns in the long-term, with short-term rates being less statistically significant. The financial metrics that have been used in the research covered in the literature review have responded differently to interest rate hikes and interest rate cuts, with metrics such as portfolio securities-to-total assets showing a significantly positive correlation to interest rate sensitivity during a rate hike, whereas during an interest rate cut equity-to-total assets was more significant. This suggests that the bulk of stock return variability is due to cash flow shocks. The literature review has shown that typically larger banks are more vulnerable because it is more complex for these organisations to restructure their assets and liabilities. The findings indicate that banks with shorter asset-liability duration gaps significantly reduce their interest rate exposure, hedging against negative monetary changes.
The main findings of the literature review support my own expectations in principal, but clearly show the variability in results created by the model used. The summary also reinforces that bank-level data must be appropriately controlled for, otherwise the relationship between banks’ stock returns and ECB monetary rates will be aggressively skewed.
3
Data and Methodology
The nature of the relationship under investigating requires data from a number of different sources covering various aspects of the economic and banking world. The regression will encompass bank balance sheet data, macro-economic data and the explanatory variable ECB deposit facility rates.
Table 1 – Banks in the sample pool
Bank Name
HSBC Holdings Plc. ING Groep NV
BNP Paribas SA UniCredit SpA
Crédit Agricole SA Credit Suisse Group AG
Barclays Plc. Nordea Bank AB (publ)
Deutsche Bank AG Intesa Sanpaolo
Royal Bank of Scotland Group Plc. (The) Banco Bilbao Vizcaya Argentaria SA
Société Généralé SA Natixis SA
Banco Santander SA Commerzbank AG
Lloyds Banking Group Plc.
The pool of banks used in the analysis initially focused on the top 50 EU commercial banks. The dataset was obtained through the Bankscope5 data platform, which holds reliable and comprehensive datasets for banks around the world. There were two criteria that determined whether a European bank was included in the study. Firstly, the bank’s stock had to be traded continuously on an equity index over the period between Q12009 and end Q12015. The second criterion was that there is complete balance sheet data to determine whether a bank had unique circumstances that may impact the sensitivity of the stock returns. After filtering the original pool of banks by the aforementioned criteria, there was credible and continuous data for 17 European Bank of various sizes as shown in Table 1. The banks in this pool are ranked in terms of the magnitude of their assets. For example, the largest bank in the pool in 2014 HSBC Holdings Plc. owned assets worth €2.17 trillion. Bankscope has provided detailed balance sheet data that the regression model will have to control for in order to
5
Bank scope is the most comprehensive, global database of bank’s financial statements, ratings and intelligence. Each bank report contains a detailed consolidated and/or unconsolidated balance sheet, an income statement plus interim reports. Financial data is available for up to 16 years.
highlight the specific relationship that I am looking for. Without controlling for balance sheet variables, any perceived relationship found between the sensitivity of stock returns and deposit facility rates may be owed to bank specific variables rather than the ECB rates themselves. There are 4 key bank-level metrics that the regression model will control for: Total Assets, Net Income, Equity/ Total Assets, and Net Loans/ Total Assets. These bank-level metrics provide a well-rounded indication of a bank’s performance and illustrate periods where stock return fluctuations may be owed to the bank’s environment beyond the effect of ECB deposit rate changes.
Table 2 – Descriptive statistics of bank-level financial metrics
Obs* Mean Standard Deviation Minimum Maximum
Total Assets (Th EUR) 1,184 1.23E09** 4.99E08 5.08E08 2.48E09
Net Income (Th EUR) 1,184 2,422,083 4,708,666 -1.36E07 2.12E07
Equity/ Total Assets (%) 1,184 5.14 1.44 2.53 8.61
Net Loans/ Total Assets (%) 1,184 40.85 13.48 17.2 61.03
*Number of observations for each data variable
**E(x) refers to exponential (x), example 1.23E09 = €1,230,000,000.00
More detailed analysis of the bank-level data in Table 2, on both an individual bank and collective basis, shows that the large standard deviation in both Total Assets and Net Income metrics is owed to the variation in size of the banks in the pool. On an individual basis, most banks in the pool had relatively stable total asset holdings. Deutsche Bank lacked performance exiting the financial crisis and regained strength rapidly near the end of the analysis period. In terms of net income, UniCredit suffered hardship throughout 2013 with negative income. Comparing Equity/ Total Assets and Net Loans/ Total Assets across the banks is rather unfruitful because these metrics are relative to the individual bank’s capital structure, however we notice that the standard deviation of both metrics is quite low which suggests fairly stables performance across the sample pool.
Table 3 - Correlations bank-level financial metrics
Total Assets Net Income Equity/ Total Assets Net Loans/ Total Assets
Total Assets 1.0000 - - -
Net Income 03043 1.0000 - -
Equity/ Total Assets -0.2993 0.1505 1.0000 -
Net Loans/ Total Assets -0.5489 -0.0627 0.6605 1.0000
As might be expected there are distinctive correlations between each of the bank-level metrics. What is most notable from Table 3 is the relatively strong inverse correlation (-0.5489) between Net Loans/ Total Assets and Total Assets, as well as the strong positive correlation (0.6605) between Equity/ Total Assets and Net Loans/ Total Assets. Given the prominent correlations between these variables, it would be prudent to monitor their values over the investigation period to identify if there are any major swings that could effects the regression results.
In addition, macro-economic variables are predicted to be responsible in part for bank stock returns and as such, they must also be entered into the regression model and appropriately controlled for. Macro-economic data has been sourced from IHS Connect, which has provided access to a number of macro-economic factors that I believe will affect the sensitivity of banks’ stock returns. The following macro-economic factors: nominal GDP (€), unemployment rate, GDP deflator, Consumer Price Index (CPI) and the ECB’s HICP index are likely to influence bank performance and lending patterns and consequently need to be controlled in the regression. The dependent variable in the regression, bank stock returns, have been sourced from Yahoo Finance. This is considered as a reliable platform to obtain stock returns of publically listed companies and/ or banks, with consistent and up-to-date data. The log of returns will be used in the data analysis. The final dataset required is the explanatory variable, the ECB deposit facility rate (sometimes referred to as the overnight rate, or interest rate in this paper). The deposit facility rate is sourced from the ECB database, which holds a reliable record of the rates that the ECB set.
Table 4 - Descriptive statistics of European Macro-economic data Q12009 to Q12015
Obs* Mean Standard Deviation Minimum Maximum
Unemployment Rate (%) 1,178 10.29 0.70 8.52 11.35
GDP Deflator (y-on-y % change)* 1,184 1,17 0.20 0.75 1.58
CPI 1,184 103.71 3.18 97.93 107.29
Nominal GDP (Th EUR) 1,184 13,132.4 612.94 11,901.42 14388.7
HICP (%) 1,184 1.40 1.05 -0.6 3
*y-on-y refers to year-on-year percentage change in the GDP deflator metric
The descriptive statistics in Table 4 illustrate the macro-economic environment across the Euro Area. A deliberate decision was made to apply Europe wide macro-economic data rather than country specific macro-economic data to each bank. This is likely to encounter scrutiny because the banks in the sample pool are of various sizes and in some cases compete in quite different segments of the European market. For example, the mainstream banks such as Barclays Plc., HSBC Holdings Plc., and Deutsche Bank AG a.o operate in the same markets and are equally affected by macro-economic changes. This is not necessarily the case with smaller banks that are the dominant players in the Nordics or Southern Europe for example. Banks such as Nordea Bank AB or Intesa Sanpaolo may not be affected by Europe wide macro-economic variables to the same extent as the larger banks in the pool. The rationale behind this decision was to create a model that is specifically founded on the economic environment across Europe. If banks were tested against country specific macro-economic variables there would be little means of comparison between the results. Thus in order to effectively control for macro-economic variables, the banks in the pool have to be benchmarked against the same macro-economic environment. Pursuing the same tests using country specific macro-data would inevitably create an array of different relationships between bank stock returns and the ECB interest rate, making a conclusion very difficult to pinpoint.
Table 5 – Correlations macro-economic variables Unemployment Rate GDP Deflator CPI Nominal GDP HICP ECB Deposit Rate Unemployment Rate 1.0000 - - - - - GDP Deflator 0.2932 1.0000 - - - - CPI 0.8836 0.0659 1.0000 - - - Nominal GDP 0.7136 -0.0559 0.9435 1.0000 - - HICP* 0.0818 0.7645 -0.0085 -0.0811 1.0000 -
ECB Deposit Rate -0.5634 0.3064 -0.5927 -0.6037 0.5563 1.0000
*HICP is the harmonised index of consumer prices, a metric commonly used to assess whether a country is ready to join the Euro Area
Table 5 illustrates interesting correlations between the ECB deposit rate and the other independent macro-economic variables. The deposit rate appears to have quite a significant correlation with all of the other factors, but most notably inverse relationships with the European unemployment rate, CPI and nominal GDP. This is useful information to consider when analysing the output of the complete regression.
The time period of analysis is important to define, which is being termed the ‘post-crisis’ period between Q12009 and Q12015, although there is still much debate over whether the European Union has actually departed the financial crisis. The rationale behind this period of analysis is to alleviate the macro-economic and bank-level data from any abnormal observation resulting from the 2007-2008 global financial crisis and to make the research current enough to include the ECB’s introduction of negative deposit facility rates in June 2014. This is one of my focal objectives to understand whether negative deposit facility rates have a greater effect on stock returns.
3.1 Methodology
I have opted to use STATA a data analysis and statistical software to perform the data manipulations. This software package is reliable and allows for complex data manipulations, including regression models that are designed to describe changes in a variable with respect to changes in other variables.
The aforementioned datasets were compiled into a single file so that STATA could perform regression analyses internally across all banks and avoid the complexity of regressing across multiple individual datasets. The data has both time-series and cross-sectional components, which forms a panel dataset. Consequently it is most fitting to employ a panel regression model. There are a number of advantages of using a panel regression model, such as testing multiple dynamic hypotheses over a continuous time frame while mitigating the effects of omitted variables that are correlated with the explanatory variable, the deposit facility rates. It is noted that the panel dataset was found to be unbalanced, which means that the time component of each independent variable is different. This is a consequence of some data having a monthly period, and other data recorded on a quarterly basis. Although this is not that problematic for the regression, the data should nevertheless be treated with care. Using more data observations increases the accuracy of the panel regression, which poses the question whether the dataset contains a large enough pool of banks. However, I feel strongly that after extensive data mining and assessment of available data the pool of banks in the sample are representative of the EU’s most notable banks. The next phase was to determine whether a fixed-effects or random-effects model was most appropriate given the properties of the dataset. This was investigated using a Hausman Test to distinguish if the dataset was a random panel or a fixed panel. The P-value6 of Chi-Sq tested at 5% confidence interval was greater than 0.05, suggesting that the data was a random panel and it is most fitting to apply a random-effects model panel regression.
Once the regression technique was settled, I could start running preliminary random-effects regressions between the dependent variable (stock returns) and single independent variables.
6
The rationale behind this technique was to gain an understanding of the sensitivity of stock returns with respect to each independent variable in the regression expression. This exercise indicates whether a particular independent variable is likely to be bias and deform the results. On completion of this task, the full regression expression can be applied to the dataset effectively modelling the relationship between bank stock returns and the ECB deposit facility rate. This will allow me to test my hypothesis that EU bank stock returns are not only sensitive to deposit facility rate fluctuations, but I anticipate that returns will be negatively impacted by decreasing deposit interest rates, particularly those below zero. Following the regression analysis, section 4 of this report will display the empirical results and section 5 will provide a comprehensive analysis of the output and comparison to my hypothesis that the recent decrease in deposit facility interest rates to negatively impact stock returns.
3.2 Regression Model
In addition to the results of a Hausman test, another reason to use a random-effects panel model is because of the perceived influence of the independent variables on stock returns. The random-effects model takes the following form:
Yit =α + β1,i[Xt] + Uit + εit (1)
Where Uit represents the error between the entities and εit represents the error with-in the entities. The random-effects model assumes that the error between and with-in the entities is uncorrelated to the predictors, which in effect allows for the time in-variant variables to play a role as explanatory variables. This is particularly useful because the model caters well for the time component of my dataset and increases the accuracy of coefficient estimates. The regression model constructed to test my hypothesis will follow the same format as that of expression (1) and will include all of the previously mentioned independent variables in section 4.0.
The relationship that I wish to assess by means of a random-effects panel regression can be assimilated by the following regression equation:
STKRi = α + β1,i[ECBRt] + β2,i[UERt] + β3,i[GDPDt] + β4,i[CPIt] (2)
Where
εit
is the error term, i = bank i, t = monthly observations from 2009… 2015 Denote:• STKRi,t is the percentage change in stock return of banki at timet;
• α regression constant;
• βi constant with relation to independent variblei;
• ECBRt is the ECB set deposit facility rate over period t;
• UERt European unemployment rate over period t;
• GDPDt European gross domestic product deflator (year-on-year change) over period t;
• CPIt European consumer price index over period t;
• NGDPt European nominal gross domestic product in Euros over period t;
• HICPt harmonised consumer price index over period t, in accordance with ECB
harmonised statistical methods; • TAi total assets (€) of banki;
• NIi net income of banki;
• EQTAi equity/ total assets of banki; and
4
Empirical Results
In accordance with the methodology the following section will show the empirical results for preliminary regressions between stock returns and various independent variables, finalising with the complete random-effects panel regression in expression (2).
4.1 Macro-Economic Regressions
Table 6 illustrates the first regression between the percentage change in stock returns across the pool of banks and the collection of European macro-economic factors.
Table 6 – Regression stock returns against group of macro-economic variables
Percentage Change in Bank
Stock Returns Coef. Std. Err. z P>[z] [95% Conf. Interval)
Unemployment Rate -0.1395456 0.0364319 -3.83 0.000 -0.2109508 -0.0681403 GDP Deflator 0.200665 0.0585752 3.43 0.001 0.0858598 0.3154702 CPI 0.0639325 0.0155149 4.12 0.000 0.0335238 0.0943411 Nominal GDP (EUR) -0.0002484 0.0000504 -4.92 0.000 -0.0003473 -0.0001495 HICP -0.0327346 0.0102132 -3.21 0.001 -0.0527521 -0.012717 _cons -2.082147 0.6788807 -3.07 0.002 -3.412729 -0.7515655 R-Squared 0.1117
The coefficients displayed in the regression output in Table 6 describe the estimated change in the dependent variable as a result of changes in the independent variables. The regression coefficient of the European unemployment rate indicates that a 1% increase in unemployment translates to a 0.139 (3sf) decrease in the percentage change of all banks’ stock returns, holding all other independent variables constant. In contrast, a 1% increase in GDP deflation across the Eurozone accounts for a 0.201 (3sf) increase in the percentage change of all banks’ stock returns, holding all other independent variables constant. A 1% increase in the CPI metric translates to a 0.064 (3sf) increase in the percentage change of all banks’ stock returns and an equivalent increase in nominal GDP of the Eurozone reflects a 0.0002 (3sf)
(negligible) decrease in stock returns. Finally, a 1% increase in ECB calculated HICP metric leads to a 0.033 (3sf) decrease in the percentage change of all banks’ stock returns. Turning to an alternative statistic, the R-squared value indicates how closely the data is fitted to the regression line (described by _cons and/or line of best fit if the results were graphed). For this random-effects panel regression we are most interested in the overall squared value. R-squared = 0.1117, shows that 11.7% of the fluctuations in the percentage change in stock returns across all the banks in the sample is explained by the combined effects of the group of macro-economic factors tested in the regression. This result clearly indicates that the association between the group of macro-economic factors and dependent variable is weak. The z-value and 2-tail p-values are used to test the hypothesis that the coefficients of the independent variables are equal to 0, i.e. whether they are statistically significant or not. Because the p-values of all the independent variables are less than α set at 0.05, the coefficients of each macro-economic variables is statistically significant at 5% significance level.
4.2 Bank-Level Financial Metrics Regressions
Now I will present the regression results between the banks’ stock returns and each of the respective bank-level financial ratios. This will provide an indication of whether a single financial metric alone is taking a significant effect on the banks’ returns. I chose to run
Table 7 – Regression stock returns against Total Assets (Th EUR)
Percentage Change in Bank
Stock Returns Coef. Std. Err. z P>[z] [95% Conf. Interval)
Total Assets (Th EUR) -2.34E-12 7.49E-12 -0.31 0.755 -1.70E-11 1.23E-11
_cons 0.0108609 0.0099265 1.09 0.274 -0.0085948 0.0303165
R-Squared 0.0001
Performing a random-effect panel regression between the banks’ stock returns and the total assets of each bank as in Table 7, shows that across the pool a 1% increase in total assets leads to a negligible change (2.34E-12 decrease) in the percentage change in stock returns of
all banks. The P>[z] value 0.755 (3sf) far exceeds α = 0.05, which reinforces that total assets alone is insignificant.
Table 8 – Regression stock returns against Net Income (Th EUR)
Percentage Change in Bank
Stock Returns Coef. Std. Err. z P>[z] [95% Conf. Interval)
Net Income (Th EUR) 4.12E-10 7.93E-10 0.52 0.603 -1.14E-09 1.97E-09
_cons 0.0069939 0.004209 1.66 0.097 -0.0012556 0.0152433
R-Squared 0.0002
The next bank-level metric in Table 8 is relatively uneventful as well, with a 1% increase in banks’ net income reflecting a 4.12E-10 increase in the percentage change of all banks’ stock returns. In addition, the P>[z] value shows further statistical insignificance of net income.
Table 9 – Regression stock returns against Equity/ Total Assets
Percentage Change in Bank
Stock Returns Coef. Std. Err. z P>[z] [95% Conf. Interval)
Equity/ Total Assets -7.17E-06 0.0026031 -0.00 0.998 -0.0051091 0.0050948
_cons 0.0080369 0.0139025 0.58 0.563 -0.0192115 0.0352853
R-Squared 0.0000
Table 10 – Regression stock returns against Net Loans/ Total Assets
Percentage Change in Bank
Stock Returns Coef. Std. Err. z P>[z] [95% Conf. Interval)
Net Loans/ Total Assets 0.0000394 0.0002756 0.14 0.886 -0.0005007 0.0005794
_cons 0.0064572 0.0119375 0.54 0.589 -0.0169398 0.0298542
Table 9 and Table 10 illustrate the final bank-level metrics, which are both statistically insignificant. A 1% increase in the financial metric equity/ total assets results in a 7.17E-06 decrease in the percentage change of all banks’ stock returns. The same unit change in net loans/ total assets produced a 0.0000394 increase in stock returns across the pool of banks. The P>[z] value for both bank metrics proved to be insignificant in this model.
4.3 Complete Regression Testing
After setting the scene with the preceding regressions, I now perform the full regression from equation (2). Prior to this we have seen the effect of the group of macro-economic variables on the banks’ stock returns, as well as 4 bank-level financial metrics. However the previous results will differ from the full regression due to the collinearity between variables once they are regressed collectively. This regression provides the full and non-manipulated view on how banks’ stock returns are influenced by ECB deposit facility rates.
Table 11 – Regression in complete form from expression (1)
Percentage Change in Bank
Stock Returns Coef. Std. Err. z P>[z] [95% Conf. Interval)
ECB Deposit Rate -0.1960726 0.050994 -3.85 0.000 -2.96019 -0.0961262
Unemployment Rate -0.2217594 0.041961 -5.28 0.000 -0.3040015 -0.1395173
GDP Deflator 0.2239515 0.0581576 3.85 0.000 0.1099647 0.3379383
CPI 0.0892227 0.016844 5.30 0.000 0.056209 0.1222363
Nominal GDP (EUR) -0.000346 0.0000567 -6.10 0.000 -0.0004571 -0.0002348
HICP -0.0113417 0.0116339 -0.97 0.330 -0.0341436 0.0114603
Total Assets (EUR) 7.96E-12 1.61E-11 0.49 0.621 -2.26E-11 3.95E-11
Net Income (EUR) -8.15E-10 1.52E-09 -0.54 0.591 -3.79E09 2.16E-09
Equity/ Total Assets -0.0032792 0.0064505 -0.51 0.611 -0.0159219 0.0093635
_cons -2.615389 0.6942046 -3.77 0.000 -3.976005 -1.254773
R-Squared 0.1457
The random-effects panel regression produced the results above, and a summary of the regression coefficients is displayed in Table 12. The overall R-squared value indicates that the ECB deposit facility interest rate accounts for 14.6% of the changes in the banks’ stock returns while controlling for the collections of independent variables.
The regression coefficient of the explanatory variable, the ECB deposit facility rate, which is the focal interest of this research shows that for a 1% increase in ECB interest rate banks’ stock returns are inversely affected by 0.196 (3sf) while controlling for all other variables. This shows that there is an inversely correlated relationship between the ECB deposit facility rate and banks’ stock returns.
Table 12 summarises the effect of a 1% increase in each independent variable on the banks’ stock returns. This illustrates the variables that stock returns are most sensitive to. The results in Table 12 indicate that bank stock returns are more sensitive to macro-economic factors over bank-level financial ratios because the macro-economic variables have more significant regression coefficients. At first glance this might seem counter-intuitive, and so it will be
investigated further in section 5.
Table 12 – Regression coefficients, the effect of each independent variable on banks’ stock returns
Effect on Percentage Change of Bank Stock Returns
Variable Negative change Positive change
ECB Deposit Rate -0.196
GDP Deflator 0.224
CPI 0.089
Nominal GDP 0.000
HICP -0.011
Total Assets 7.96E-12
Net Income -8.15E-10
Equity/Total Assets -0.003
5
Analysis of Results and Discussion
This section of my investigation will delve deeper into the results found in the empirical analysis and attempt to reason with the results and gauge their impact for the European banking sector. Firstly I will analyse the results from the macro-economic regression test, before taking a more detailed look at the results of the bank-level regressions and finally analyse and discuss the complete regression.
The overall R-squared value describing the correlation between the percentage change in bank stock returns across the pool and the collection of macro-economic factors is 0.1117, indicating that only 11.17% of the changes in banks’ stock returns results from the macro-economic influences. This is significantly lower than anticipated given the expected relationship between the banking sector and the general economic landscape. From the collection of macro-economic variables, the European unemployment rate has the most significant negative relationship with banks’ stock returns, which impacts returns by -0.139% (3sf) per 1% increase in unemployment. On the contrary, for a 1% increase in GDP deflation, bank stock returns are boosted by 0.201% (3sf). Correlation between these variables is 0.2932, which is considered low. The other macro-economic independent variables have smaller effects of bank stock returns, at least while the model takes this form. This evidence is rather weak and coincides with the work of Çiftçi (2014), suggesting that macro-economic factors alone do not cause substantial changes in bank stock returns. The z-values of each of the economic variable in the test appear to contradict this, showing that each of the macro-variables are significant when tested at 5% significance level.
I now turn to each bank-level financial metric that was regressed against the percentage change of banks’ stock returns. It is beneficial to analyse each bank metric individually first, before combining their impacts on stock returns, to assess whether there is a single financial metric that has a profound impact. The regression between total assets and stock returns shows little significance. The coefficient of total assets is negligible (-2.34E-12) and negatively correlated to stock returns for which the p-value supports this finding, far exceeding the value of α set at 5% significance level. The net income of the banks’ in the sample pool also proves insignificant, with a minor positive coefficient (4.12E-10). The
p-value is 0.603 (3sf), which again shows that net income is insignificant with regard to changes in banks’ stock returns. It is important to note however that these results are clearly model dependent, which was also found by Elyasiani and Mansur (2004), because typically a bank’s net income is expected to be fundamental in driving stock returns and providing dividends to shareholders. It is possible that net income is insignificant in this instance because the model focuses on the percentage changes (fluctuation) in stock returns, distinct from the magnitude of the returns. The final bank-level metrics, Equity/ Total Assets and Net Loans/ Total Assets, also proved to be insignificant when regressed alone against the percentage change in bank stock returns. The regression output from both metrics exhibit high p-values in addition, and together have R-squared values equal to zero. This exercise has shown that none of the bank-level financial metrics alone are impacting a bank’s stock returns using this model, and now I can progress to analysing the complete regression from expression (2).
The ECB deposit facility rate is pivotal for this research and was originally anticipated to catalyse changes in banks’ stock returns. The results observed however do not suggest such a straightforward relationship between the deposit facility interest rate and bank stock returns. The random-effects panel regression model is designed such that the various independent variables are controlled for over time, which aims to single-out the relationship between the ECB deposit rates and the stock returns of all the banks in the sample pool. With this model in place, the regression coefficient of the ECB deposit rate is -0.196 (3sf), which indicates that for a 1% increase in the ECB deposit rate there is a corresponding 0.196 (3sf) decrease in the percentage change of bank stock returns across the sample pool, holding all other variables constant. This is confirmed statistically significant with respect to changes in bank returns through the p-value of 0.000, tested at 5% significance level. In spite of this, the statistical impact of the ECB deposit rate is small on average across all the banks; this is perhaps because each bank in the pool holds varying amounts of capital with the ECB overnight. Large banks, ranked according to total assets, such as HSBC and BNP Paribas will hold greater sums of capital in the ECB’s overnight deposit facility and therefore have a larger interest rate risk exposure. This would mean that their stock returns could be more heavily impacted by deposit facility interest rates. The random-effects model does not capture this because the regression coefficient of the ECB’s deposit rate is averaged across the banks.
Now that the all of the independent variables are pooled into a single regression, the observations previously explained have changed marginally. The macro-economic factors are significant as shown by their respective p-values, with varying impacts on bank stock returns. European unemployment rate and GDP deflation still remain the most impactful on stock returns with the coefficient of unemployment rate decreasing to -0.222 (3sf), and the coefficient of GDP boosting bank stock returns with a regression coefficient of 0.224 (3sf). The remaining macro-economic variables have little effect on bank stock returns, which is likely explained by the weak relationship between the banking sector and CPI and HICP. The bank-level metrics also experience a few changes once they are compiled in the complete regression, yet still remain statistically insignificant judging by their p-values in Table 11. Their regression coefficients also suggest that their impact on bank stock returns on a monthly basis is negligible. This is unexpected because when banks suffer financial shocks their share prices tend to be hit rapidly. Such unfruitful results possibly indicate that the banking sector has experienced a stable and less risk-driven period since the financial crisis, which is confirmed by low volatility in net income for most of the banks in the pool. An additional explanation is that when the ECB reduces deposit rates, banks do not restructure their asset-liability spread. This means that while interest rates on bank loans remain constant, banks’ obligations on their deposits are less costly. As a result, the spread between a bank’s deposit rate and loan rate expands, translating into greater margins which improves performance.
The overall R-squared value for the complete regression is 0.146 (3sf), showing that 14.6% of changes in bank stock returns are owed to the ECB’s deposit facility interest rate, using this model and controlling for the various factors above. However, we cannot place much significance on the Squared value since it is entirely model dependent. The resulting low R-Squared value does not necessarily imply that the results are not meaningful. The relationship found between bank stock returns and the ECB deposit facility rate may indicate that when deposit rates are very low, consumers and businesses alike tend to invest their capital in the stock markets rather than holding it in fruitless deposit accounts. As a result, the entire stock market increases, propping up bank stock returns as well, even if they are suffering poor performance. In addition, we have learnt previously from Dia (2013) that banks typically scale-back their loan books during negative interest rate shocks, however deposit facility rates have a counterintuitive dynamic to other interest rates. The premise behind reducing deposit facility rates is to encourage more lending, and as such over the analysis period banks are
likely to have increased the size of their loan books, which has generated greater returns with higher interest rates on loans and consequently boosted stock returns. An interesting extension to this work is be to investigate the reaction of banks’ asset-liability maturity gaps to interest rate changes.
6
Conclusion
My research investigates how European banks’ stock returns are affected by changes in the ECB deposit facility interest rate between Q12009 and Q12015. In order to successfully evaluate the impact of the ECB’s deposit rate on banks’ stock returns, a collection of both macro-economic variables and bank-level financial metrics were controlled for in a regression. The model took form as a random-effects panel regression, which was implemented using STATA software. The results from the regression modelling contradict the original expectation and illustrate a negatively correlated relationship between the percentage change in bank stock returns and the ECB deposit facility rate. While controlling for the macro-economic and bank-level metrics, the random-effects model indicated that for a 1% increase in the ECB deposit rate there is a corresponding 0.196 (3sf) decrease in the percentage change of bank stock returns across the sample pool. ECB deposit rates are statistically significant with a p-value of 0.000. There are a number of explanations that justify these findings. Firstly, during relaxed monetary policy environments banks reduce their deposit rates while maintaining interest rates on loans, which increases their margins. Secondly, investors move away from banking products and take more risk in the stock market; this increases stock prices in general including those of banks. This means that the ECB deposit rate has less of an impact on banks, which is supported by the findings that bank fundamentals are less significant. This coincides with increasing bank valuations as the market rises across the board.
The random-effects panel model was selected because of the time series and cross-sectional components of the data. However, the prevalent result that bank-level metrics are insignificant suggests that the regression model has struggled to deal with collinearity between the variables. Nevertheless, the results still indicate a clear relationship between bank stock returns and the ECB deposit facility interest rate, which could be made more valuable for banks with some fine-tuning of the regression model.
7
References
1 Bloomberg QuickTake, Randow, J., April 13th 2015, Less than zero
2 Borio, C., 2014, The international monetary and financial system: its Achilles heel and what to do about it, BIS Working Papers, No 456
3 Castrén, O., Fitzpatrick, T., Sydow, M., What drives EU banks’ stock returns? Bank-level evidence using the dynamic dividend-discount model, European Central Bank, Working Paper Series, No. 677, Sep. 2006
4 Çiftçi, S., The influence of macroeconomic variables on stock performance, University of Twente, Dec. 2014
5 Dell’Ariccia, G., Laeven, L., Marquez, R., 2010, Monetary policy, leverage, and bank risk-taking, IMF December 2010
6 Dia, E., 2013, How do banks respond to shocks? A dynamic model of deposit-taking institutions, Journal of Banking & Finance, 37 (2013) 3623-3638
7 European Central Bank (ECB), The ECB’s response to the financial crisis, Article: monthly bulletin, Oct. 2010
8 Elyasiani, E., Mansur, I., 1998, Sensitivity of bank stock returns distribution to changes in the level and volatility of interest rate: A GARCH-M model, Journal of Banking & Finance 22 (1998) 535-563, Temple University and Widener University
9 Elyasiani, E., Mansur, I., 2004, Bank stock return sensitivities to long-term interest rates: a multivariate GARCH approach, Managerial Finance, Vol. 30 Iss 9 pp. 32-55
10 Flannery, M., James, C., 1984, The effect of interest rate changes on common stock returns of financial institutions, The Journal of Finance, Vol. 39, No. 4 (Sep., 1984), pp. 1141-1153
11 Ghazanfari, F., Rogers, H., Sarmas, P., 2006, The effect of Federal Reserve interest rate policies on the returns of commercial banks stocks, Journal of Financial Services Marketing, Vol. 11, 4 349-359
12 Kwan, S. 1991, Re-examination of interest rate sensitivity of commercial bank stock returns using a random coefficient model, Journal of Financial Services Research, 5:61-76, The University of Arizona
13 Maddaloni, A., Peydró, J-L., 2011, The low monetary rates paradox, banking stability and credit: evidence from the Euro Area, 12th Jacques Polak Annual Research Conference, page 2
14 Tahmoorespour, R., Ardekani, A. M., 2012, Examine banks’ share price sensitivity due to interest rate changes: emerging markets and advanced countries, International Conference on Economics, Business Innovation IPEDR vol. 38 (2012)
