Cutting jobs to compete : analysing commercial bank profitability

Student number: 11056908

Programme: BSc Economie en Bedrijfskunde Specialization: Finance and Organization Supervisor: Sander Onderstal

This document is written by Student Jelle de Jager 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.

Cutting Jobs To Compete: Analysing Commercial Bank Profitability

Jelle de Jager

Programme: BSc Economie en Bedrijfskunde / Specialization: Finance and Organization / Student Number: 11056908

1. Introduction

News articles from recent years indicate that large banks have been cutting jobs for almost a decade now and that hundreds of thousands of people have been or will be laid off and that more job cuts may follow (Arnold, 2017; Duran, 2018; Henry & Rothacker, 2012; HuffPost, 2011a, 2011b; Kavoussi, 2012; Nelson, 2016; Noonan, 2018b; Rankin, J., 2013; Wilchins, 2013). The reasons for these layoffs seem to vary. Some news articles suggest that in the years after, but close to the financial crisis, banks laid off employees mainly to reduce costs, downsize, or satisfy shareholders in response to decreased revenues (Henry & Rothacker, 2012; HuffPost, 2011a; Kavoussi, 2012). News articles about more recent job cuts in the banking sector indicate that digitalization, increased competition and technological ad-vances have become important factors as well (Arnold, 2017; Nelson, 2016; Pollari, 2017); “Technology is removing a number of the barriers to entry that once insulated our business. We face competition from many different directions, with relatively new players providing more segmented offers to our customers.” (ING, 2018, p. 9)

Furthermore, some news articles indicate that job cuts can be costly because the firm can be confronted with charges and redundancy payments that can increase to billions of dollars (Henry & Rothacker, 2012; Nelson, 2016). News articles also indicate that more recent layoffs are associated with investments in information technology (IT) (Nelson, 2016; Spezzati & Cehic, 2018). In October 2016, for example, ING made public that it planned to save €900 million a year by cutting 5,800 jobs that would become obsolete by investing €800 million in technology over the course of five years, to realize their digital transformation (Nelson, 2016). Justifying this layoff, an ING spokesman stated that their customers expect the same level of service that they get from other major digital platforms (Nelson, 2016), which is in line with the view of Markovitch and Willmott (2014) that the (in 2014 ongoing) digitalization of business processes is (or was) driven by customer demand.

“If you can be the most successful at bringing your staff numbers down the quickest, that’s going to give you the quickest cost advantage”, said an insider of one of Australia’s largest banks (Duran, 2018, para. 4). This statement sug-gests that the benefits that banks can obtain from cutting jobs are dependent on the actions of other banks. Also, banks that do not adapt to technical change (i.e., digitalization in this context) could lose customers that prefer convenience to

U

NIVERSITEIT VAN

A

MSTERDAM

_{& Business}

A B S T R A C T

This paper examines the relationship between the asset employee ratio (AER) and profitability for large U.S. commercial banks over the period 2000–2017 and investigates if this relationship has been stronger after the financial crisis. It shows that (1) the trend in the AER is positive but does not have a steeper slope after 2008, (2) the AER has a negative impact on profitability, although the impact for the period 2009–2017 is less negative than for the period 2000–2008, and (3) above average percentage changes in the AER have an impact on profitability that is negative before, but positive after the financial crisis. Robustness checks were performed by analysing each model again while controlling for more sources of bias. Taking the robustness checks into account; the main findings of this paper are that there is a positive trend in the AER, that there is a negative relationship between the AER and profitability, and that the relationship between above average percentage changes in the AER and profitability is more positive for the period 2009–2017 than for the period 2000–2006.

Type: Bachelor thesis Submitted: 26 June 2018 Field: Organization Keywords: Bank profitability Employment Financial crisis A R T I C L E I N F O

familiarity (Nelson, 2016), which too could mean that early movers can gain an advantage over late movers. The pre-vious two paragraphs suggested that job cuts can be a way to cut costs and a result of digitalization. In addition, this paragraph suggests that both of these motivations are interrelated with the actions of competitors.

So, it seems that (1) the financial crisis may have incited or forced banks to become more cost-efficient, (2) recent technological advances may have increased competition, which in turn may have forced banks to innovate or cut costs, (3) banks may be adapting to digitalization as response to customer demand, (4) each of these three developments may have led to job cuts, and (5) these developments have been taking place after the financial crisis.

Next, several statistics are discussed to provide more insights into developments in the banking sector. Lázaro-Aguilera and Palomo-Zurdo (2016) show (accounting for enlargement of the EU), that there has been a decrease of about 50% in the number of monetary financial institutions in the Euro-area between 1999 and 2014 and that there has been a decrease in the number of employees and branches in most countries, comparing 2009 and 2013. Statistics from

FRED (n.d.-a) show a 70% decrease in the number of U.S. commercial banks with average assets under $1 billion be-tween 1984 and 2017. The number of commercial banks with average assets greater than $1 billion, on the other hand, increased. Other statistics show a negative peak in net income for U.S. commercial banks between 2006 and 2012, with the minimum in 2009 (FRED, n.d.-b). Statistics on the net interest margin show a difference between banks with aver-age assets below $15 billion and banks with averaver-age assets above $15 billion (FRED, n.d.-c; FRED, n.d.-d; FRED, n.d.-e); the steep decrease of the net interest margin for commercial banks with average assets above $15 billion after 2010 and the dip in net income between 2006 and 2012 could be an underlying factor of the job cuts in large banks that were carried out to cut costs.

Hence, all of the above could make one wonder if banks are indeed using a decreasing amount of labor, how doing so impacts their profitability, and if this impact has become greater after the financial crisis. Therefore, the research question of this paper is: “Regarding large commercial banks, and comparing a period before to a period after the fi-nancial crisis; to what extent is there a decreasing trend in the amount of labor that is used by banks (compared to size), and to what extent can (1) this amount of labor and (2) large job cuts serve as predictors of profitability?”.1

To answer this question, a dataset on large U.S. commercial banks over the period 2000–2017 is used: first to esti-mate the trend in the asset employee ratio (AER) (i.e., the quotient of a bank’s total assets divided by its number of employees); second, to estimate the relationship between the AER and profitability; and third, to estimate the relation2

-ship between above average percentage changes in the AER and profitability.3

Although related literature provides useful insights into the interactions of technology, employment, competition and performance in the context of the banking sector, more recent research on these topics is missing (Beccalli 2007; Bos et al., 2013; Fung, 2006; Fungáčová et al., 2017; Hauswald and Marquez, 2003;Hung et al., 2018; Mainardes et al. 2017; Mithas et al. 2012; Vicente-Lorente & Zúñiga-Vicente, 2012). This paper contributes to the literature on these topics by providing insights into the employment developments in the banking sector and their relationship with prof-itability. Also, articles on bank profitability show that taking different perspectives (to bank profitability) may yield new insights (Albertazzi & Gambacorta, 2010; Beccalli, 2007; Bolt et al., 2012; Bouzgarrou et al., 2018; Dietrich & Wanzenried, 2014; Feng & Wang, 2018), to which this paper contributes by taking the perspective of the AER.

Moreover, this research may be valuable to banks, other financial services companies and investors because it shows the relationship between the AER and profitability. Finally, this research may be valuable to governments be-cause it discloses insights into labor market developments and may indicate if banks are likely to cut more jobs in the future (which is the case if using less labor leads to higher profitability).

The structure of this paper is as follows. Section 2 reviews the related literature. Section 3 defines the hypotheses. Section 4 describes the methodology. Section 5 discusses the data. Section 6 presents the results. Section 7 concludes, discusses limitations and makes suggestions for future research.

2. Literature review

Four topics are considered relevant to this research: technology, labor, competition and performance. Articles that link at least two of these four topics are discussed in Section 2.1. The estimation of bank profitability is considered to be an important aspect of this research as well; literature on bank profitability is discussed in Section 2.2.

“Large” will be defined as relatively large.

1

Note that a decrease in the amount of labor (compared to size) is equivalent to an increase in the AER.

2

The effect of job cuts (or increases in the AER) is investigated separately because doing so may provide more insights than would result from only investigating the impact

3

of the AER as static measure: “above average percentage changes in the AER” takes (to some extent) both the dynamics in the AER and its (potential) relationship to the actions of competitors into account.

2.1. Technology, labor, competition and performance

This paragraph provides an overview of the main findings of the articles on technology, labor, competition and performance. Each article is discussed in more detail in the subsequent paragraphs of this section. In brief, labor-saving technologies could decrease or substitute, but also increase or complement employment (Fung, 2006; Mainardes et al. 2017; Vicente-Lorente & Zúñiga-Vicente, 2012). IT investments seem to have a positive impact on profitability for firms in general, but little or no impact on profitability for banks specifically (Beccalli, 2007; Mithas et al., 2012). Low and high (compared to moderate) competition seem to stimulate innovation for U.S. banks (Bos et al., 2013). Advances in IT could increase but also decrease competition in the financial services sector (Hauswald & Marquez, 2003) and competition seems to have a negative impact on profitability for U.S. banks (Fungáčová et al., 2017; Hung et al., 2018).

Based on the vast 30% decrease in the average labor hours per dollar of banking output between 1992 and 2002,

Fung (2006) examined the relation between labor-saving technologies and employment. He found that labor-saving technologies can, depending on the effect of increased output that results from these technologies, both increase and decrease employment: despite a lower average of labor hours per dollar of output, the number of employees per bank could increase because of (on) innovation (based growth opportunities). Accordingly, Mainardes et al. (2017) found that the use of information technology rather complements than substitutes human capital in the service sector as com-pared to the non-service sector, using a Brazilian database and comparing data from 1985 and 2002. Vicente-Lorente and Zúñiga-Vicente (2012) found evidence that supports the statements that (1) more product innovation and (2) process innovations related to the acquisition of complementary production assets reduce the probability of downsiz-ing. They used data on Spanish manufacturing firms over the period 1994–2006.

Based on a dataset of 400 global firms over the period 1998–2003, Mithas et al. (2012) found that IT investments have a positive impact on profitability and that, while IT investments enable both revenue growth and reductions in operating expenses, the IT-enabled revenue growth has the greatest impact on profitability. Beccalli (2007), however, found little relationship between IT investments and improved profitability for European banks between 1995 and 2000. To explain her findings, she suggested that IT investment affects the competitive environment of the banking sector, which in turn could reduce average profits.

Bos et al. (2013) examined the effect of competition on innovation for U.S. banks between 1984 and 2004 and found evidence for an inverted-U relationship. Hauswald and Marquez (2003) investigated how advances in IT affect competition in the financial services sector and, based on the application of their model on banks, state that technologi-cal progress could decrease competition and increase interest rates, but could also increase competition and decrease interest rates. They suggest that technological advances in information processing could decrease competition because firms that engage in gathering information would gain an advantage over those who do not. However, when technolog-ical progress makes obtaining information less costly, IT advances could increase competition (Hauswald & Marquez, 2003).

Fungáčová et al. (2017) found a positive effect of competition on the cost of credit for European banks in the peri-od 2001–2011. In line with economic competition theory, Hung et al. (2018) found that competition has a negative im-pact on profitability for U.S. banks between 1986 and 2013. Therefore, the findings of Fungáčová et al. (2017), Hung et al. (2018) and Hauswald and Marquez (2003) could explain the finding of Beccalli (2007) that IT investments do not lead to better performance if IT investments would indeed lead to an increase in competition. The findings of

Hauswald and Marquez (2003) could explain the finding of Mithas et al. (2012) that IT investments have a positive impact on profitability if IT investments decrease competition. However, given that the findings of Beccalli (2007)

concern European banks and that the findings of Mithas et al. (2012) concern global companies in general, the findings of Beccalli (2007) are more likely to apply to U.S. commercial banks.

2.2. Bank profitability

This section discusses literature on bank profitability to (1) create a theoretical framework on which may be relied in the remainder of this paper and (2) specifically select one article that will serve as the basis for the estimation of bank profitability (which is covered in the methodology).

With literature dating back to at least 1979, bank profitability has been an extensively studied topic (Dietrich and Wanzenried, 2014). Six articles are considered relevant because they investigate bank profitability for the same type of bank (commercial banks), country (U.S.) or for a period similar to 2000–2017. This selection includes the articles of

Beccalli (2007), Albertazzi and Gambacorta (2010), Bolt et al. (2012), Dietrich and Wanzenried (2014), Feng and Wang (2018) and Bouzgarrou et al. (2018) (see Table 1).

Beccalli (2007) investigated the impact of IT investments on profitability for European banks for the period 1995– 2000 and found that IT investments do not necessarily lead to better performance (as discussed in Section 2.1). Alber-tazzi and Gambacorta (2010) examined the impact of the corporate income tax on profitability for European and U.S. banks and found that this taxation (substantially) affects the composition of revenues. Bolt et al. (2012) investigated the profitability of banks during recessions. Their main findings are that output growth had a greater impact on bank prof-itability than was commonly found in literature and that long-term interest rates of “previous” years are important pre-dictors of profitability in periods of high economic growth.

Dietrich and Wanzenried (2014) investigated determinants of commercial bank profitability, distinguishing low-, middle-, and high-income countries, and show that the used determinants can explain differences in the profitability of commercial banks. Feng and Wang (2018) examined why European banks were less profitable than U.S. banks in the period 2004–2014. They show that the profitability of European banks deteriorated over time and that this could be explained by declines in, inter alia, the relative lending rate, relative return on securities and technical efficiency, and by an increase in funding costs. Bouzgarrou et al. (2018) investigated the profitability of in the French market operat-ing domestic and foreign banks for before and duroperat-ing the financial crisis. Their results show that the foreign banks were more profitable, especially during the financial crisis.

Next, this selection of six articles is, by comparing important aspects of each paper, narrowed down to one that will be used to estimate bank profitability. Only the article of Bouzgarrou et al. (2018) is published in a journal that is not from the Tinbergen Journal List. The article published in the journal with the highest Tinbergen Journal List classifica-tion is the article of Dietrich and Wanzenried (2014) (with journal classification AA). Table 1 provides an overview of characteristics of each article. Looking at “Type of banks”, Beccalli (2007), Dietrich and Wanzenried (2014) and

Bouzgarrou et al. (2018) are the best options because they focus on commercial banks only. Looking at “Countries”,

Beccalli (2007) and Bouzgarrou et al. (2018) could be less good options because the U.S. is not among the countries they investigate. Looking at “Period”, Dietrich and Wanzenried (2014), Feng and Wang (2018) and Bouzgarrou et al. (2018) are the best options because their time frames have the largest overlap with the time frame of this paper (which is 2000–2017).

So, given this evaluation, it could be concluded that the article of Dietrich and Wanzenried (2014) would provide the best basis for the estimation of bank profitability because only their paper is for every discussed aspect among the best options. The fact that they succeeded in identifying determinants of bank profitability substantiates this choice.

Table 1

Overview and features of the selected articles on bank profitability

Authors Type of banks Number of banks Countries Period Perspective

Beccalli (2007) Commercial 737 France, Germany, Italy,

Spain, U.K. 1995–2000 IT investments

Albertazzi and Gambacorta (2010)

Commercial, Cooperative, Mutual,

Saving and more

195 U.S., 8 European

countries 1981–2003 Corporate income tax

Bolt et al. (2012)

Commercial, Cooperative, Investment, Real estate,

Saving

On average 166 U.S. banks and 733 banks from other countries.

U.S., 18 other countries 1990–2007 Profitability during recessions

Dietrich and

Wanzenried (2014) Commercial 10,165 118 countries 1998–2012

Determinants of profitability in low-,

middle-, and high-income countries

Feng and Wang (2018) Large 220 Europan and 301 U.S. banks

U.S., 12 European

countries 2004–2014

Profitability of European versus U.S.

banks

Bouzgarrou et al.

(2018) Commercial 170 France 2000–2012

Profitability during and before the financial crisis: domestic versus

3. Hypotheses

I base my hypotheses on the related literature and on the developments in the banking sector that were discussed in the introduction. The hypotheses can be divided into three sets: one about the trend in the AER, one about the relation-ship between the AER and profitability, and one about the relationrelation-ship between job cuts (i.e., increases in the AER) and profitability. Each hypothesis applies to one or more of the following periods: pre-financial crisis (2000–2006), the financial crisis (2007–2008) and post-financial crisis (2009–2017).4

My first hypothesis is that the AER has increased after the financial crisis (H1a). I also expect that the slope of the trend in the AER has been steeper after 2008, as compared to the period 2000–2008 (H1b). I reason H1a and H1b as follows. Although the related literature is unclear about the impact of technological progress on employment and the consulted statistics do not indicate a direction in the trend in the AER for the period 2000–2008 (FRED, n.d.-a; Fung, 2006; Lázaro-Aguilera & Palomo-Zurdo, 2016; Mainardes et al. 2017; Vicente-Lorente & Zúñiga-Vicente, 2012), the news articles discussed in the introduction show that there have been many large job cuts in the banking sector after the financial crisis. So, while banks may not have been triggered to cut costs before the financial crisis, it certainly seems that they have been forced or stimulated to do so thereafter (e.g., Henry & Rothacker, 2012; HuffPost, 2011a; Kavous-si, 2012; Nelson, 2016; Spezzati & Cehic, 2018). The following hypotheses will be tested:5

(H1a) There is a positive trend in the AER in the period 2009–2017.

(H1b) The trend in the AER has a more positive slope after 2008, compared to the period 2000–2008.

My second hypothesis is that the AER has had a positive impact on profitability after the financial crisis (H2a). I also expect that the impact of the AER on profitability has been more positive after 2008, as compared to the period 2000–2008 (H2b). I reason H2a and H2b as follows. The related literature discussed in Section 2.1 seems to imply that

employment, IT advances, competition and performance are, to a certain extent, interdependent factors with an unclear

polarity of impact, which does not facilitate proposing a substantiated hypothesis about the effect of the AER on prof-itability (Beccalli 2007; Bos et al., 2013; Fung, 2006; Fungáčová et al., 2017; Hauswald & Marquez, 2003;Hung et al., 2018; Mainardes et al. 2017; Mithas et al. 2012; Vicente-Lorente & Zúñiga-Vicente, 2012). The post-financial crisis developments and events related to employee layoffs discussed in Section 1, however, do.

If using less labor results in lower personnel expenses, doing so will have a positive impact on profitability, ceteris

paribus. However, when labor has to be substituted by some sort of capital, a negative impact on profitability is also

possible (e.g., when the costs of capital exceed the reduction in costs that results from using less labor). Indeed, from the introduction it seems clear that increasing the AER (by cutting jobs) can go along with large IT investments or re-dundancy payments (Henry & Rothacker, 2012; Nelson, 2016). Yet, given the fact that banks have been laying off em-ployees and have been substituting labor with capital could mean that doing so (eventually) leads to more benefits than costs (and therefore positively affects profitability), (which is also what the news articles suggest (e.g., Duran, 2018; Henry & Rothacker, 2012; Nelson, 2016)). Accordingly, and because I expect that the AER has increased after the fi-nancial crisis (H1a) and that the slope of the trend in the AER has been steeper after 2008, as compared to the period 2000–2008 (H1b), I also expect that the AER has had a positive impact on profitability after the financial crisis, and that the impact of the AER on profitability has been more positive after 2008, as compared to the period 2000–2008. Therefore, the following hypotheses will be tested:

(H2a) The AER has a positive impact on profitability in the period 2009–2017.

(H2b) The AER has a more positive impact on profitability after 2008, compared to the period 2000–2008.

My third hypothesis is that above average percentage changes in the AER have had a positive impact on profitabil-ity after, but not before the financial crisis (H3a; H3b). I also expect that the impact of above average percentage changes in the AER on profitability is (regardless of H3a and H3b) more positive for the post-financial crisis period than for the pre-financial crisis period (H3c). H3 takes a more sophisticated approach to the AER than H2 and may 6

substantiate the findings of H2. I reason H3a, H3b and H3c as follows. First of all, large job cuts are likely to be care7

-fully considered operations to which many actions precede and are therefore considered to be requiring a vast amount

According to Degl’Innocenti et al. (2016), the financial crisis can be divided into three phases: the U.S. subprime crisis (2007–2008), the global financial crisis (2009–2010)

4

and the sovereign debt crisis (2010–2012). The U.S. subprime crisis, given the many events that occurred in the banking sector during this period (e.g. the bankruptcy of Lehman Brothers), is considered to have been the most prominent financial crisis phase for U.S. commercial banks. Therefore, the period 2007–2008 will be referred to as the “financial crisis”.

Almost all identified news articles about job cuts in the banking sector reported job cuts that did (or were about to) take place after the financial crisis.

5

Note that H3c is supported when H3a and H3b are supported, but that H3c can also be supported while H3a and H3b are not; “more positive” can also mean less negative.

6

H3 uses the same measure (the AER), but incorporates its dynamics, compares these dynamics to competitors and excludes the financial crisis; which may result in different

7

of anticipation. Second, it seems that after the financial crisis, banks have been forced to cut jobs either as a conse8

-quence of decreased revenues or adapting to technological change (Arnold, 2017; Henry & Rothacker, 2012; HuffPost, 2011a; Kavoussi, 2012; Nelson, 2016; Pollari, 2017). Therefore, “having above average percentage changes in the AER” may reflect to what extent banks are able to outperform their competitors in adapting to those market changes that require or result in job cuts; which is expected to positively affect profitability (Duran, 2018; Nelson, 2016). While many sources indicate that it has been common that job cuts are a part of a goal or strategy after the financial crisis (2009–2017) (e.g., HuffPost, 2011a, 2011b; Nelson, 2016; Wilchins, 2013), I did not manage to find any source that indicates that the same holds for the pre-financial crisis period (2000–2006). So, to investigate if large job cuts can in-deed serve as a predictor of profitability after the financial crisis, while not before, and (otherwise) to determine if there is (at least) a more positive relationship between large job cuts and profitability after the financial crisis (as compared to before), the following hypotheses will be tested:

(H3a) Above average percentage changes in the AER do not have a positive impact on profitability in the period 2000–2006.

(H3b) Above average percentage changes in the AER have a positive impact on profitability in the period 2009–2017.

(H3c) The impact of above average percentage changes in the AER on profitability is more positive in the period 2009–2017 than in the period 2000–2006.

4. Methodology

This section outlines the methodological approach. Each set of hypotheses (H1, H2 and H3) is tested by a different model. The first model only uses the AER and a time variable. The second and third model use the AER, profitability measures and a set of control variables. Section 4.1 describes all variables that are used. Section 4.2 presents the mod-els and discusses how the modmod-els and variables are used to test the hypotheses.

4.1. Variables

Section 4.1.1 discusses the dependent variables and argues why the return on assets (ROA), the return on equity (ROE) and the net interest margin (NIM) (and not the return on average assets (ROAA), return on average equity (ROAE) and NIM) are used to measure profitability. Section 4.1.2 describes the independent variables.

4.1.1. Dependent variables

The most common measures of bank profitability are ROAA, ROAE and NIM (Dietrich & Wanzenried, 2014). ROAA measures how profitable a firm’s assets are, or, put differently, how efficiently a firm uses its assets to generate profits. ROAA is calculated by dividing total profits by the yearly average of total assets. ROAE measures how much profit a firm generates with its shareholders’ investments and is calculated by dividing total profits by the yearly aver-age of total equity. NIM is calculated by dividing the net interest income by the averaver-age interest-earning assets and measures how profitable a bank’s interest activities are. 9

However, the dataset that is used lists merged banks under the same name as one of these previously unmerged banks, which means that an observation for bank A and year t, could actually be the observation for the merger of bank A and bank B that continued under the name “A”, while the observation for year t-1 is the observation for (the un-merged) bank A. The same principle holds for acquisitions. For the data, it means that total assets and total equity (or any quantitative variable) could double or even triple from one fiscal year to the next due to mergers and acquisitions, resulting in meaningless values of ROAA and ROAE because they are based on yearly averages (while profits are not). (NIM is reported by banks and present in the database and therefore assumed to be accurate.)

Identifying and filtering out all mergers and acquisitions could cause problems as well because there have been so many in the banking sector. Also, many banks have been involved in more than one merger or acquisition. Adjusting all observations for banks that have been involved in mergers and acquisitions is beyond the scope of this paper. Dropping observations from before or after a merger or acquisition would still be a complicated process and would result in the loss of too many data. Hence, ROAA and ROAE will not be used to measure profitability. ROA and ROE (also used by

Beccalli (2007) and Bouzgarrou et al. (2018)) will be used instead. ROA and ROE are less accurate when a firm grew

Actions such as setting goals, changing strategies, legal actions and having any form of replacement ready for use (e.g., digital platforms and IT).

8

For each measure (ROAA, ROAE and NIM), the “average” is the average value for the corresponding fiscal year.

or shrunk a lot within a fiscal year, but this is expected to cause less bias than changes in size between fiscal years (caused by mergers or acquisitions) would. Because ROE disregards financial leverage (Dietrich & Wanzenried, 2014) and NIM is also a less comprehensive measure than ROA (e.g., because it disregards all expenses other than interest expenses), the ROA is considered to be the most important measure of profitability.

4.1.2. Independent variables

The main independent variable is the AER. When not used to measure percentage changes, the natural logarithmic equivalent of the AER is used to control for its highly positively skewed distribution. 10

As argued in Section 2.2, the work of Dietrich and Wanzenried (2014) will provide the basis for the estimation of profitability. Accordingly, the following independent (control) variables will be used: capital ratio, cost-to-income

ra-tio, loan loss provisions, deposits to assets, bank size, interest income share, funding costs, and non-performing assets. Inflation, measured by the consumer price index (CPI), will be used as well, but only to adjust total assets for inflation

when estimating the trend in the AER. Next, each control variable is discussed. It is also argued why some variables will not be used.

The capital ratio (also used by Bouzgarrou et al. (2018), Feng and Wang (2018) and Iannotta et al. (2007)) is the ratio of total equity to total assets and represents capital strength (Dietrich & Wanzenried, 2014). The cost-to-income

ratio is the ratio of operating costs to total revenues and measures how cost-efficient a firm is in generating revenue. Loan loss provisions is measured by the ratio of loan loss provisions to total loans and proxies credit risk (Dietrich & Wanzenried, 2014).

Growth deposits measures the annual growth rate of total deposits. Using this variable, however, could be

prob-lematic because of the merger-acquisition issue (described in the previous section). An option would be to measure the growth in the quotient of deposits over total assets (which controls for changes in firms size). Yet, this measure could still be inaccurate when the deposit-asset structure of a bank changes after a merger or acquisition. Therefore, instead of using a variable that measures the growth in deposits, another variable that incorporates deposits will be used:

de-posits to assets. Bolt et al. (2012) found that deposits to assets (although using it as lagged variable), has a significant impact on profitability.

Bank size is most commonly measured by total assets (Dietrich & Wanzenried, 2014). Accordingly, and based on the articles of Beccalli (2007), Bouzgarrou et al. (2018) and Iannotta et al. (2007), the natural logarithm of total assets is used to measure bank size. The interest income share is the ratio of total interest income to total income and accounts for the potential differences between returns from asset management and returns from (traditional) interest operations (Dietrich & Wanzenried, 2014). Funding costs represents the deposit interest payments and is measured by interest ex-penses over total deposits.

The effective tax rate, defined as taxes paid over before-tax profits, does not seem to have a significant impact on profitability for high-income countries (Dietrich & Wanzenried, 2014). Also, in the profitability model of Albertazzi and Gambacorta (2010), the corporate income tax rate does not have a significant impact on after-tax profits. Because of these findings, the effective tax rate will not be used.

Finally, non-performing loans are loans on which no interest payments have been made for at least 90 days and that are therefore considered to be in default. Non-performing loans had a negative impact on bank profitability in the model of Feng and Wang (2018). Because of a lack of available data on non-performing loans specifically,

non-per-forming assets is used instead. Like total deposits, non-pernon-per-forming assets is scaled by total assets.11 4.2. Models

This section explains how the above-described variables are used. Sections 4.2.1, 4.2.2 and 4.2.3 describe the models, and Section 4.2.4 shows that the use of these models is appropriate. To test the hypotheses, six ordinary least squares (OLS) regression analyses are performed: the first to test H1a and H1b (Model 1), the second, third and fourth to test H2a and H2b (Model 2), and the fifth and sixth to test H3a, H3b and H3c (Model 3).

The type of data that is available allows for the elimination of the omitted variable bias (that the OLS estimators of the regression coefficients could have) that is caused by omitted variables that differ across entities but are consistent over time or change over time but are constant across entities (Stock & Watson, p. 347). The fixed effects regression controls for these so-called entity and time fixed effects (Stock & Watson, p. 347) and will be applied to Model 2 and

The natural logarithmic equivalent of the AER is the natural logarithm of the quotient of total assets over employees.

10

In all year reports that were examined, the value for non-performing loans nearly equaled the value for non-performing assets (that was obtained from the database).

Model 3. A multiple regression analysis will be applied to Model 1. Next, each model is discussed and linked to the corresponding hypotheses.

4.2.1. Trend in the AER

The first sub-hypothesis (H1a) is that there is a positive trend in the AER in the period 2009–2017. The second sub-hypothesis (H1b) is that the trend in the AER has a more positive slope after 2008, compared to the period 2000– 2008. Both hypotheses are tested at once by analysing Model 1. AERit is the asset employee ratio, FYt is the fiscal year

(that corresponds to the observations for total assets and employees) reduced by 2000, ! is a dummy that equals 1 if

FYt indicates a year after 2008, and 0 otherwise, and "it is the error term. The total assets values are in- or deflated to 12

2010 prices to control for inflation. (Model 1)

For both #1 and #2, it is tested if they significantly differ from zero. H1a is not rejected when the sum of #1 and #2

is positive. H1b is not rejected when #2 is significantly positive. If both hypotheses are not rejected, there seems to be

evidence that the banks in the sample have been using a decreasing amount of labor (compared to size), on average, in the period 2009–2017, and that the decreases in employment have become larger after the financial crisis as compared to before.

4.2.2. Impact of the AER on profitability

The first sub-hypothesis (H2a) is that the AER has a positive impact on profitability in the period 2009–2017. The second sub-hypothesis (H2b) is that the AER has a more positive impact on profitability after 2008, compared to the period 2000–2008. Both hypotheses are tested at once by analysing Model 2. Πit is one of the profit measures (ROA,

ROE or NIM), AERit is the asset employee ratio, ! is a dummy that equals 1 if FYt indicates a year after 2008, and 0

otherwise, Xit is the set of control variables discussed in Section 4.1.2 and "it is the error term.

(Model 2)

For both #1 and #2, it is tested if they significantly differ from zero. H2a is not rejected when the sum of #1 and #2

is positive. H2b is not rejected when #2 is significantly positive. If both hypotheses are not rejected, there seems to be

evidence that the AER has a positive impact on profitability for the banks in the sample, on average, in the period 2009–2017, and that this impact has become greater after the financial crisis as compared to before.

4.2.3. Impact of above average percentage changes in the AER on profitability

The first sub-hypothesis (H3a) is that above average percentage changes in the AER do not have a positive impact on profitability in the period 2000–2006. The second sub-hypothesis (H3b) is that above average percentage changes in the AER do have a positive impact on profitability in the period 2009–2017. The third sub-hypothesis (H3c) is that the impact of above average percentage changes in the AER on profitability is more positive in the period 2009–2017 than in the period 2000–2006. These three hypotheses are tested by analysing Model 3 twice, once for the period 2000–2006 and once for the period 2009–2017. The main independent variable, Zit, is a dummy that equals “1” if the percentage

change in the AER for bank i between year t and t-1 is greater than the yearly average percentage change in the AER, and “0” otherwise. The dependent variable is ROA. The other variables are the same ones that are used in Model 2.13 14

(Model 3)

For both periods (2000–2006 and 2009–2017) it is tested if #1 significantly differs from zero. H3a is not rejected

when #1 is not significantly positive for the period 2000–2006. H3b is not rejected when #1 is significantly positive for

the period 2009–2017. H3c is not rejected when #1 for the period 2009–2017 is more positive than, and significantly

The fiscal years are reduced by 2000 for the convenience of the interpretation of the regression results.

12

One might wonder if the merger-acquisition issue applies to Z; it is considered that it does not. The reasons for this assumption are that the AER is a quotient of two firm

13

size measures (that may not even change after a merger or acquisition) and that it could be argued that mergers and acquisitions are a way to (intentionally) increase (or more generally, change) the AER.

Observations for total assets and employees for the year 1999 are used to calculate the values of Zit for the year 2000.

14

AER

=

β

+

β

FY

+

β

δ FY

+

ε

Π

=

β

+

β

AER

+

β

δ AER

+

β

X

∑

+

ε

Π

=

β

+

β

Ζ

+

β

X

∑

+

ε

different from #1 for the period 2000–2006, which will be tested by comparing their confidence intervals. If all three

hypotheses are not rejected, there seems to be evidence that large job cuts can serve as predictor of profitability for the period 2009–2017, while not for the period 2000–2006.

4.2.4. Appropriateness of the entity and time fixed effects models

First, Hausman specification tests were performed to test whether or not a fixed effects model would be more ade-quate than a random effects model (StataCorp, n.d.-a). For each model, (Model 2 (ROA), Model 2 (ROE), Model 2 (NIM), Model 3 (2000–2006) and Model 3 (2009–2017)), the null hypothesis that the individual-level effects are more appropriately modelled by a random-effects model was rejected (p-value<0.001) (see Appendix A); justifying the use of the fixed effects model (StataCorp, n.d.-a).

Second, Wald tests were performed to test whether or not time fixed effects should be incorporated in the fixed effect models (StataCorp, n.d.-b). For each model, (Model 2 (ROA), Model 2 (ROE), Model 2 (NIM), Model 3 (2000– 2006) and Model 3 (2009–2017)), the null hypothesis that the coefficients for all year dummies jointly equal to zero was rejected (p-value<0.001) (see Appendix A); justifying the use of time fixed effects in the fixed effects models.

5. Data

This section describes the dataset, shows descriptive statistics and discusses the correlation coefficients of the vari-ables. Two databases were consulted. The inflation rates and indices were obtained from the OECD (2018). The data for all other variables were obtained from the Compustat database in WRDS (Banks Daily, Bank Fundamentals Annu-al). As stated before, this study focusses on large U.S. commercial banks, with “large” defined as having an amount of total assets greater than or equal to $1 billion (following Feng and Wang (2018)). All observations with negative values for asset or equity measures or with obvious wrong values for any of the variables were excluded from the dataset (Degl’Innocenti et al., 2016; Dietrich & Wanzenried, 2014; Feng & Wang, 2018). Finally, banks with only one obser-vation were excluded as well. The sample is an unbalanced panel dataset of 525 U.S. commercial banks and 4,758 bank-year observations over the period 2000–2017. Table 2 provides descriptive statistics for each variable. Appendix B shows how the profit measures and control variables were computed.

Now, the correlation matrix in Appendix C is discussed to examine the basic relationship between each variable and to examine if the AER is correlated to any of the control variables. Using the guide of Evans (1996) to describe the strength of correlations verbally, it can be stated that the correlations between the independent variables are mostly very weak (0.00–0.19) or weak (0.20–0.39) and that only a few are strong (0.60–0.79). The independent variables are (overall) higher correlated to ROA than to ROE and NIM, which is (considered to be) in line with the view that ROA is

Table 2

Descriptive Statistics

Variable Symbol Mean Median Standard deviation Observations

Return on assets ROA 0.724 0.908 1.106 4,758

Return on equity ROE 6.111 9.255 30.01 4,758

Net interest margin NIM 3.762 3.720 0.759 4,703

Asset employee ratio AER 8.439 8.398 0.436 4,543

Capital ratio CR 9.775 9.530 2.610 4,758

Cost-to-income ratio CTIR 81.41 77.21 22.50 4,758

Loan loss provisions LLP 0.685 0.307 1.263 4,758

Deposits to assets DTA 0.762 0.778 0.087 4,758

Bank size BS 8.432 7.994 1.423 4,758

Interest income share IIS 79.22 81.27 34.36 4,758

Funding costs FC 2.003 1.555 1.701 4,758

Non-performing assets NPA 0.013 0.007 0.018 4,715

the most comprehensive profitability measure (Dietrich & Wanzenried, 2014). The AER, on the other hand, is nega-tively correlated with ROA and NIM (and not correlated with ROE), which is against the expectation that the AER is positively related to profitability.

The last correlation remark is about multicollinearity. (Imperfect) multicollinearity occurs when independent vari-ables are very highly correlated and could result in imprecise estimates of the corresponding regression coefficients (Stock & Watson, p. 199). In any case, however, multicollinearity between control variables is considered to be not problematic because the interpretation of their estimated coefficients is not relevant to this paper. The highest correla-tion with AER is with bank size and equals only 0.24. Therefore, no correlacorrela-tion coefficient is considered to have indi-cated any problem.

6. Results

Sections 6.1, 6.2 and 6.3 show and explain the results for Model 1, Model 2 and Model 3, respectively. Section 6.4

explains what tests were performed to check the robustness of the results. Section 6.5 discusses what results persist.

6.1. Results for the analysis of Model 1

The regression results for Model 1 are presented in Table 3. H1a is supported because the sum of #1 and #2 is

posi-tive. H1b is not supported because #2 is not significantly positive. So, this means that there is evidence for a positive

trend in the inflation-adjusted AER in the period 2009–2017, but that this trend is not more positive than it was in the period 2000–2008. The F-test indicates that the coefficients of FY and !FY are jointly significantly different from zero (p-value<0.001).

The model suggests that, on average, the AER increased with approximately 2.9% per year over the period 2000– 2008, and with approximately 2.0% per year over the period 2009–2017. A 2% increase in the AER could mean that a firm with $10 billion in total assets and 10,000 employees would lay off about 200 of them, for example.

6.2. Results for the analysis of Model 2

The regression results for Model 2 are presented in Table 4. Based on ROA, H2a is not supported because the sum of #1 and #2 is not positive, and H2b is supported because #2 is significantly positive (p-value<0.001). Based on ROE

and NIM, H2a and H2b are both not supported. So, the AER does not have a positive impact on profitability in the pe-riod 2009–2017 although its impact is more positive than for the pepe-riod 2000–2008 (when ROA is used to measure profitability). All three models have significant F-statistics (p-value<0.001).

The model for ROA suggests that, on average, the return on assets decreases with approximately 0.00317 percent-age points (pp) in the period 2000–2017 and with 0.00298 pp in the period 2009–2017 when the non-logarithmic

Table 3

Summary of Regression Analysis for Model 1, Predicting the Inflation-Adjusted AER

Variable Coefficient (Standard error)

FY 0.029*** (0.003) FY × ! -0.009*** (0.002) (constant) 8.272*** (0.014) Number of observations 4,543 Number of banks 525 F-Test 147.06*** R2 _0.0594

Note. Robust standard errors. FY equals the corresponding fiscal year minus 2000 (meaning FY=0 for the fiscal year 2000, FY=1 for 2001, etc.). The inflation-adjusted AER is the natural logarithm of the untransformed equivalent. H0: #1 = 0 vs. H1: #1 ≠ 0. H0: #2 = 0 vs. H1: #2 ≠ 0.

* Indicates statistical significance at the 5% level. ** Indicates statistical significance at the 1% level. *** Indicates statistical significance at the 0.1% level.

equivalent of the AER increases by 1%. A 1% increase in the AER could mean that a firm with $10 billion in total 15

assets and 10,000 employees would lay off about 100 of them. A decrease in ROA of 0.00298 pp instead of 0.00317 pp could mean that the net income of a bank with $10 billion in total assets would drop from $100,000,000 to $99,702,000, instead of to $99,683,000. This suggests that after the financial crisis, a 1% increase in the AER reduces net income by about $19,000 less than it would have in the period 2000–2008 (elaborating on the same example).

6.3. Results for the analysis of Model 3

The regression results for Model 3 are presented in Table 5. H3a is supported because #1 is significantly negative

(p-value<0.001) (and therefore not significantly positive). H3b is supported because #1 is significantly positive

(p-val-ue<0.05) and H3c is supported because the difference between #1(2000–2006) and #1(2009–2017) is significantly positive as

well (p-value<0.01). Both models (3(1) and 3(2)) have a significant F-statistic (p-value<0.001).16

The results for Model 3 suggest that an above average percentage change in the AER for bank i between year t-1 and year t would have resulted in a change in ROA of -0.035 pp in the period 2000–2006 and of +0.050 pp in the peri-od 2009–2017. This could mean that, as a result, the net income of a bank with $10 billion in total assets would have changed from $100,000,000 to $96,500,000 in the period 2000–2006, but to $105,000,000 in the period 2009–2017.

Table 4

Summary of Regression Analysis for Model 2, Predicting ROA, ROE and NIM ROA (1) ROE (2) NIM (3)

Variable Coefficient (Standard error) Coefficient (Standard error) Coefficient (Standard error)

AER -0.317*** (0.082) -8.744* (3.630) -1.047*** (0.095) AER × ! 0.019*** (0.005) 0.160 (0.181) -0.016** (0.005) CR 0.039*** (0.007) 1.960** (0.595) 0.048*** (0.008) CTIR -0.048*** (0.003) -0.605*** (0.062) -0.010*** (0.002) LLP -0.009 (0.041) -2.306 (1.562) 0.148*** (0.029) DTA -0.512* (0.211) -24.90 (14.68) 0.236 (0.360) BS -0.261*** (0.045) -3.575* (1.558) 0.041 (0.051) IIS 0.001** (0.000) 0.014*** (0.003) 0.000 (0.001) FC 0.078*** (0.013) 1.381*** (0.347) -0.033** (0.010) NPA 2.537* (1.186) -166.1* (76.29) -3.736** (1.124) (constant) 9.127*** (0.570) 158.3*** (28.10) 12.49*** (0.845) Number of observations 4,526 4,526 4,523 Number of banks 507 507 506 F-Test 348.11*** 36.72*** 44.47*** R2 _{0.7427 0.3360 0.2490}

Note. Robust standard errors. R2 _{is the “overall” variant of the in Stata reported R-squareds. The asset employee ratio and bank size are the natural logarithm}

of the untransformed equivalent. H0: #1 = 0 vs. H1: #1 ≠ 0. H0: #2 = 0 vs. H1: #2 ≠ 0.

* Indicates statistical significance at the 5% level. ** Indicates statistical significance at the 1% level. *** Indicates statistical significance at the 0.1% level.

In Model 2, the AER is the natural logarithm of the quotient of total assets over employees. The example, however, discusses the effect of an increase in the non-logarith

15

-mic equivalent of the AER (i.e., the quotient of total assets over employees).

H3c is supported with a significance level of at least 1% because the 99% confidence intervals of Z for the period 2000–2006 and 2009–2017, respectively, do not overlap.

6.4. Robustness checks

Two robustness checks regarding the regression results of Model 1, Model 2, and Model 3 were carried out. Be-cause of the focus on U.S. (commercial) banks only, and beBe-cause of the use of time fixed effects in the fixed effects models, macroeconomic factors that differ only across countries and not within (the U.S.) can not have biased the re-sults to the extent that the banks in the sample are active in the U.S. only. Extreme values in the data are also a poten-tial source of bias. Because the largest banks are considered most likely to be active in countries other than the U.S. and also to have the most extreme data values, all regression analyses were performed again while excluding bank-year observations that belong to the category that represents the 1% highest values of total assets. Bank-year observations that belong to the category that represents the 1% lowest values of net income were excluded as well: banks with heavy losses might have been in financial distress in the corresponding year and are therefore considered to may have had extreme values for other variables as well. The results persist and seem robust for every model (see Appendix D, Table D1).

A final check for all three models was performed, where, additionally to the exclusions of the previous analysis, only banks for which data for every year in the period of the corresponding model was available were included; creat-ing balanced panels. The results for Model 1 persist. The findcreat-ing of Model 2 that the AER has had a more positive im-pact on ROA after the financial crisis does not persist. The results for Model 3 persist to some extent: the imim-pact of above average percentage changes in the AER on profitability is still negative and significant for the period 2000– 2006, but not significantly positive for the period 2009–2017; the finding the impact is more positive after the financial crisis than before persists (see Appendix D, Table D2).

Table 5

Summary of Regression Analysis for Model 3, Predicting ROA 2000–2006

(1)

2009–2017 (2)

Variable Coefficient (Standard error) Coefficient (Standard error)

Z -0.035*** (0.010) 0.050* (0.020) CR 0.021*** (0.005) 0.060*** (0.014) CTIR -0.041*** (0.003) -0.040*** (0.003) LLP -0.063 (0.063) -0.077* (0.038) DTA 0.367 (0.205) -0.188 (0.411) BS -0.192*** (0.028) -0.490*** (0.048) IIS -0.010*** (0.002) 0.002*** (0.000) FC 0.084*** (0.012) 0.082 (0.044) NPA 1.355 (3.394) 0.431 (1.242) (constant) 5.999*** (0.358) 7.326*** (0.603) Number of observations 1,694 2,493 Number of banks 349 396 F-Test 28.95*** 213.83*** R2 _{0.5156 0.5875}

Note. Robust standard errors. R2 _{is the “overall” variant of the in Stata reported R-squareds. Bank size is the natural logarithm of the untransformed}

equivalent. H0: #1 = 0 vs. H1: #1 ≠ 0.

* Indicates statistical significance at the 5% level. ** Indicates statistical significance at the 1% level. *** Indicates statistical significance at the 0.1% level.

6.5. Discussion of the results

This section discusses the results of the initial analyses and the robustness checks. All three analyses of Model 1 indicate that there is a positive trend in the AER that is not steeper after 2008. To provide better insight into the trend in the AER: Appendix E shows the values for the yearly averages of the AER and the inflation-adjusted AER. While the average inflation-adjusted AER sharply increased after 2011, it decreased between 2008 and 2011, which could have resulted in a less steep slope for the period 2009–2017 overall.

Also, it seems to be the case that the AER and above average percentage changes in the AER have had a negative impact on profitability before the financial crisis. A more positive impact of the AER on profitability after the financial crisis persists through the first robustness check, but not through the second. A positive impact of above average per-centage changes in the AER after the financial crisis also persists through the first robustness check, but not through the second. Finally, a more positive impact of above average percentage changes in the AER on profitability after the financial crisis as compared to before persists through both robustness checks.

7. Conclusion

In this paper, I have answered the following research question: “Regarding large commercial banks, and comparing a period before to a period after the financial crisis; to what extent is there a decreasing trend in the amount of labor that is used by banks (compared to size), and to what extent can (1) this amount of labor and (2) large job cuts serve as predictors of profitability?”. This was investigated by using a sample of 525 large U.S. commercial banks over the pe-riod 2000–2017 and by specifying the amount of labor that banks use, compared to their size, as the asset employee ratio (AER) (i.e., total assets over employees).

The findings suggest that there is a decreasing trend in the amount of labor that is used by banks (compared to size), but that there is no stronger decrease after the financial crisis. It also seems that there is a positive relationship 17

between the amount of labor that banks use (compared to size) and profitability and that this relationship is less

posi-tive after the financial crisis. Lastly, the findings suggest that the relationship between making job cuts that are (rela18

-tively) above average in size and profitability is negative before the financial crisis, but positive after the financial cri-sis.19

I see the following limitations of my research. From Section 2.1 it seems that IT investment is correlated with em-ployment and may have a positive impact on profitability (Beccalli, 2007; Fung, 2006; Hauswald & Marquez, 2003; Hung et al., 2018; Mithas et al., 2012). From the introduction it seems that IT investment is positively correlated with both the AER and profitability, especially for more recent years (e.g., Arnold, 2017;Nelson, 2016). Therefore, not in-cluding IT investments in the regression models could have resulted in OLS estimators that have omitted variable bias (Stock & Watson, p. 180). IT investment was omitted because of a lack of available data.

A second factor that may be correlated with both the AER and profitability is competition. Competition seems to have a negative impact on profitability and (as banks try to cut costs) a positive impact on the AER (Hung et al., 2018; ING, 2018; Pollari, 2017). Competition measures that are country-specific were not applicable because the values of these measures only differ across time and not across entities, and are therefore already controlled for by using time fixed effects in the fixed effects models. Non-structural competition measures that can be applied to individual firms (such as the Lerner index) were not used because computing such measures for each bank was beyond the scope of this research (Dietrich & Wanzenried (2014)). However, I would like to suggest a way to apply country-specific competi-tion measures in a single-country setting.

To do so, one could divide the country into multiple regions and use the geographical footprint of each bank to determine in what regions each bank is active. Next, for each region, the competition measures can be computed by 20

only including those banks that are active in this region in the calculation. Then, for each bank separately, the value of each competition measure that applies to the geographical footprint of the bank can be computed by taking the average of the corresponding competition measure values of all regions in which the bank is active.21

A third limitation of this paper is the potential mismatch between fiscal years and calendar years. While the infla-tion rates used in Model 1 are based on calendar years, the data on total assets and employees are reported for fiscal years. Also the “after 2008” dummies in Model 1 and Model 2, and the periods for which each model was tested are (in

I.e., a positive trend in the AER that does not have a steeper slope after the financial crisis.

17

I.e., a negative relationship between the AER and profitability that is less negative after the financial crisis.

18

The findings discussed in this paragraph are the findings of the initial analyses. The findings for the robustness checks are discussed in Sections 6.4 and 6.5.

19

The geographical footprint of a bank shows the locations of its branches or customers for example.

20

Here follows an example. Banks A and B are active in region X. Banks A and C are active in region Y. Competition measure I for region X is based on banks A and B.

21

Competition measure I for region Y is based on banks A and C. The value of competition measure I that applies to bank A (i.e., to its geographical footprint) is the average of the value of competition measure 1 for region X and the value of competition measure 1 for region Y (because bank A is active in regions X and Y). Note that increasing the amount of regions increases the accuracy of the bank-specific competition measures.

terms of interpretation) based on calendar years (that might differ up to a year from a bank’s fiscal year). This issue could have had implications for the results when the average fiscal year period of the sample does not equal a calendar year, for instance. However, because almost all banks in the sample have a fiscal year that does equal a calendar year, I suppose that no bias has arisen from this third limitation.

A fourth limitation is the omission of two profitability determinants. The profitability models (Model 2 and Model 3) are based on the article of Dietrich and Wanzenried (2014), who also use dummy variables that indicate state- and foreign ownership. In this paper, these variables were omitted because of a lack of available data. However, because I 22

do not see (or could find literature that indicates) why foreign-ownership would be (highly) correlated with the AER, I do not think that omitting this variable could have (drastically) biased the results.

Omitting state-ownership could be more problematic if banks that are state-owned have less incentive to become cost-efficient or adapt to technological change and have therefore a lower AER. Although Iannotta et al. (2007) found a significant impact of state-ownership on profitability for large European banks over the period 1999–2004, state-own-ership did not have a significant impact on profitability for high-income countries in the paper of Dietrich and Wanzen-ried (2014) (that examines commercial banks over period 1998–2012), which could mean that omitting “state-owner-ship” is not that problematic.

Furthermore, an effective way to account for mergers and acquisitions might enable more sophisticated ways to test the effect of the AER on profitability. For example, incorporating ROAA and delayed effects of the AER, large job cuts and IT investments might lead to different results and more insights. Also, it could be that increasing the AER is foremost relevant for very large banks and that focussing only on those banks yields different results as well. Lastly, because the average (inflation-adjusted) AER decreased between 2008 and 2011, but sharply increased thereafter (see

Appendix E), it could be that the job cuts that were made to cut costs were less large than the job cuts that were the result of digitalization. Therefore, comparing two periods after the financial crisis may yield more understanding 23

about the relationships between the AER and the underlying factors of job cuts.

While this paper has provided some insights into the labor developments in the banking sector and their relation-ship with profitability, more research is needed to draw conclusions about the causal relationrelation-ship between the AER and profitability. Finally, I think that this topic deserves more investigation as the job cuts that have yet occurred may have been only the tip of the iceberg (Smith, 2018; Noonan, 2018a).

Acknowledgements

The author wishes to thank Sander Onderstal for his guidance and valuable comments. The author also wishes to thank two anonymous referees for their helpful feedback. All errors remain those of the author.

Appendix A

Table A1

Testing the Adequacy of Fixed Effects Models and Time Fixed Effects

Test Model 2 (ROA) Model 2 (ROE) Model 2 (NIM) Model 3 (2000–2006) Model 3 (2009–2017)

Hausman ($2_{) 238.29*** 43.05*** 275.78*** 135.16*** 177.40***}

Wald (F) 12.59*** 3.33*** 23.30*** 7.61*** 6.47***

Note. The corresponding test statistic is shown in the first column (within parentheses). *** Indicates statistical significance at the 0.1% level.

Indicating if more than 50% a bank’s shares are held by the state or by foreign shareholders, respectively.

22

Section 1 suggests that the job cuts after, but close to the financial crisis were made to cut costs while the more recent job cuts were (mainly) the result of digitalization.

Appendix B

Appendix C

Table B1

Variable Formulas

Variable Formula

Return on assets (Net Income / Total Assets) × 100

Return on equity (Net Income / (Common/Ordinary Equity Total + Preferred/Preference Stock)) × 100

Net interest margin Net Interest Margin

Capital ratio ((Common/Ordinary Equity Total + Preferred/Preference Stock) / Total Assets) × 100 Cost-to-income ratio (Total Current Operating Expenses / Total Current Operating Revenue) × 100 Loan loss provisions (Provision for Loan/Asset Losses / Net of Total Allowance for Loan Losses) × 100

Deposits to assets (Total Deposits / Total Assets) × 100

Bank size log(Total Assets)

Interest income share (Interest and Related Income / (Interest and Related Income + Total Non-Interest Income)) × 100

Funding costs (Interest and Related Expenses / Total Deposits) × 100

Non-performing assets (Non-Performing Assets / Total Assets) × 100

Note. The variables used in the formulas are all from the Compustat database in WRDS (Banks Daily, Bank Fundamentals Annual).

Table C1

Correlation Matrix

ROA ROE NIM AER CR CTIR LLP DTA BS IIS FC NPA

ROA 1 ROE 0.66*** 1 NIM 0.25*** 0.14*** 1 AER -0.08*** -0.02 -0.36*** 1 CR 0.21*** 0.16*** 0.20*** 0.13*** 1 CTIR -0.90*** -0.58*** -0.21*** -0.08*** -0.25*** 1 LLP -0.63*** -0.47*** 0.00 -0.01 -0.12*** 0.71*** 1 DTA -0.09*** -0.08*** 0.20*** -0.05*** 0.00 0.02 -0.02 1 BS 0.08*** 0.05*** -0.23*** 0.24*** 0.15*** -0.08*** 0.07*** -0.42*** 1 IIS -0.02 -0.01 0.08*** 0.07*** -0.04* 0.04** 0.00 0.05*** -0.15*** 1 FC -0.04** -0.01 -0.04** -0.22*** -0.35*** 0.20*** 0.14*** -0.54*** 0.00 0.08*** 1 NPA -0.56*** -0.44*** -0.09*** 0.02 -0.08*** 0.61*** 0.63*** 0.14*** -0.11*** 0.04* -0.01 1

Note. Pearson correlation coefficients. The variable names are showed in Table 2. The AER is the natural logarithm of the untransformed equivalent. * Indicates statistical significance at the 5% level.

** Indicates statistical significance at the 1% level. *** Indicates statistical significance at the 0.1% level.

Appendix D

See Table D1 and Table D2.

Table D1

Robustness Check of Model 1, Model 2 and Model 3 (Outliers in Total Assets and Net Income Excluded)

Model 1 Model 2 (1) Model 2 (2) Model 2 (3) Model 3 (1) Model 3 (2)

Dependent variable AER ROA ROE NIM ROA ROA

Period 2000-2017 2000-2017 2000-2017 2000-2017 2000-2006 2009-2017 FY 0.028*** 0.029*** - - - - -FY × ! _-0.009***-0.008** - - - - -AER - -0.305*** -0.317*** -6.979** -8.744* -1.009*** -1.047*** - -AER × ! - 0.017*** 0.019*** 0.181 0.160 -0.017** -0.016** - -Z - - - - -0.036*** -0.035*** 0.043* 0.050* Number of observations 4,453 4,543 4,436 4,526 4,436 4,526 4,433 4,523 1,677 1,694 2,438 2,493 Number of banks 524 525 506 507 506 507 505 506 348 349 391 396 F-Test 143.50*** 147.06*** 318.71*** 348.11*** 29.57*** 36.72*** 45.49*** 44.47*** 31.10*** 28.95*** 174.60*** 213.83*** R2 0.0595 0.0594 0.7271 0.7427 0.3140 0.3360 0.2694 0.2490 0.5119 0.5156 0.5947 0.5875

Note. The values of the estimated coefficients and statistics of the original analyses are shown in grey. * Indicates statistical significance at the 5% level.

** Indicates statistical significance at the 1% level. *** Indicates statistical significance at the 0.1% level.

Appendix E

Table D2

Robustness Check of Model 1, Model 2 and Model 3 (Outliers in Total Assets and Net Income Excluded; Balanced Panels)

Model 1 Model 2 (1) Model 2 (2) Model 2 (3) Model 3 (1) Model 3 (2)

Dependent variable AER ROA ROE NIM ROA ROA

Period 2000-2017 2000-2017 2000-2017 2000-2017 2000-2006 2009-2017 FY 0.019*** 0.029*** - - - - -FY × ! _-0.009***0.000 - - - - -AER - -0.412*** -0.317*** -4.319** -8.744* -1.328*** -1.047*** - -AER × ! - -0.007 0.019*** -0.115 0.160 -0.012 -0.016** - -Z - - - - -0.031** -0.035*** 0.007 0.050* Number of observations 1,273 4,543 1,267 4,526 1,267 4,526 1,267 4,523 954 1,694 1,260 2,493 Number of banks 72 525 72 507 72 507 72 506 137 349 140 396 F-Test 50.20*** 147.06*** 103.19*** 348.11*** 89.54*** 36.72*** 44.75*** 44.47*** 18.03*** 28.95*** 195.91*** 213.83*** R2 0.0727 0.0594 0.7955 0.7427 0.6916 0.3360 0.4444 0.2490 0.5161 0.5156 0.4791 0.5875

Note. The values of the estimated coefficients and statistics of the original analyses are shown in grey. * Indicates statistical significance at the 5% level.

** Indicates statistical significance at the 1% level. *** Indicates statistical significance at the 0.1% level.

A ss et s P er E m pl oye e (i n U S D m il li ons ) 3 4 5 6 7 8 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017

Average AER Average Inflation-Adjusted AER

Note. 2010 was used as base year to adjust total assets for inflation. The (inflation-adjusted) AER values that were used to calcualte the yearly averages were not logarithmically transformed.