CEOoOverconfidence and cost stickiness

(1)

Asna Warsha Mohan (10546464) 22-06-2014

CEO Overconfidence and Cost

Stickiness

Faculty of Economic and Business, University of Amsterdam MSc in Accountancy and Control, track Accountancy

Thesis supervisor: dr. Bo Qin

Co-assessor: dr. Alexandros Sikalidis

Abstract:

In this research we propose CEO overconfidence as a behavioral explanation of SG&A cost stickiness. Building on the psychology literature, we predict that overconfident CEOs are more likely to overestimate future demand and therefore less likely to cut SG&A costs when sales decline. Using a UK sample of 1628 firm-years between 2002 and 2009 we document that SG&A cost stickiness increases in the degree of CEO overconfidence. We document that our results are consistent when we perform additional analysis and find that cost stickiness showed up in firms with overconfident CEOs. In further analysis, we restricted overconfident CEOs to the top quartile of overconfidence in the sample and found the same results. Overall, the results show that cost stickiness is mainly associated with overconfident CEOs rather than less confident CEOs. By providing a cost stickiness determinant at executive level, our results provide strong support for the role of managerial discretion in cost management.

(2)

TABLE OF CONTENT

1. Introduction ... 3

1.1 Background of the study ... 3

1.2 Problem definition ... 4

1.3 Research question ... 5

1.4 Key findings ... 6

1.5 Academic and managerial relevance ... 7

1.6 Structure of the thesis ... 7

2. Theoretical framework ... 8

2.1 Cost stickiness ... 8

2.1.1 The theory of asymmetric cost behavior ... 8

2.1.2 Streams SG&A costs stickiness... 9

2.1.3 Previous findings related to cost stickiness ... 10

2.2 CEO overconfidence ... 12

2.2.1 Defining and conceptualizing overconfidence ... 12

2.2.2 Previous findings related to CEO overconfidence ... 15

2.3 Gaps identified ... 16

2.4 Hypotheses development ... 17

2.5 Conceptual model ... 18

3. Methodology... 19

3.1 Data and sample collection ... 19

3.2 Measurement variables ... 20

3.2.1 Measurement cost stickiness ... 20

3.2.2 Measures of CEO overconfidence ... 20

3.4 Empirical model ... 23 3.3 Control variables ... 24 4 Results ... 25 4.1 Descriptive results ... 25 4.2.3 PCA analysis ... 27 4.2 Main results ... 27 4.2.1 Cost stickiness ... 27

4.2.2 Regression results hypothesis ... 28

4.2.3 Additional analyses ... 34

Conclusion ... 35

Reference List ... 37

Appendix ... 42

Appendix 1 Variable definitions ... 42

Appendix 2 Screeplot PCA ... 43

(3)

CEO Overconfidence and Cost Stickiness | 1. Introduction 3

1. INTRODUCTION

This chapter contains an introduction of the thesis. Firstly it will provide some background information on the problem definition. Secondly, the research question will be defined. Next, the academic relevance will be outlined. This chapter finalizes with the structure of the thesis.

1.1 BACKGROUND OF THE STUDY

Understanding cost behavior is a fundamental element of cost accounting and the management of a firm. Conventional costs - and management accounting techniques assume that costs vary proportionally with the activity. Various accounting management techniques, such as budgeting build on the assumption that the costs change proportionally if the activity changes regardless of the direction of the change. Deviating from this traditional assumption of symmetric cost behavior, numerous studies show that costs are sticky, that is, they decrease less when sales fall than they increase when sales rise (Balakrishnan, et al, 2004; Noreen and Soderstrom ,1997 and Calleja, et al, 2006). Specifically, Anderson et al, 2003 conclude that changes in the cost not only depend on the size, but also on the direction of the change in the activity. They conclude that Selling, General and Administrative costs (SG&A) increase by 0.55% per 1% increase in sales, but only decrease by 0.35% per 1% decline in sales. This phenomenon is known as asymmetrical cost behavior and is referred to as ‘cost stickiness’ in the management accounting literature (Balakrishnan, et al, 2004; Noreen and Soderstrom ,1997 and Calleja, et al, 2006).

Different studies have investigated the asymmetrical behaviour of costs (Andersen et al, 2003; Subramaniam & Weidenmier, 2003; Calleja et al, 2006). All of these studies concluded that the most frequent costs that behaves sticky, are Selling, General and Administrative costs. According to Needles et al, 2008, SG&A costs are “expenses other than those related to the production that are needed to support sales and overall operations during an accounting period”. SG&A expenses occur in a period, not directly related to the production of goods. While selling expenses are related to the company’s effort to sell products, general and administrative costs are expensed for the general administration within the company (Poston & Grabksi, 2001). Examples of SG&A costs are salaries expenses, advertising expenses and other administrative expenses. According to the literature, SG&A costs are adjustable by managers and therefore influenced by managerial behavior (Dalla et al, 2013). Under certain circumstances and management decisions, SG&A costs behave in an

(4)

asymmetrical way regarding sales activity changes because they are exposed to adjustment costs. On the contrary, resources like direct materials can easily be adjusted without incurring adjustment costs (Banker et al, 2011). Consistent with prior research, the focus for the cost stickiness analysis will be on SG&A costs in this thesis.

Prior literature focuses on economic and agency explanations for the cross-sectional variation in cost stickiness. The economic explanation suggests that the decision to cut or keep SG&A resources when sales decline depends on the trade-off between the managers’ expectations about the persistence of the demand decline and the magnitude of the adjustment costs associated with cutting SG&A resources in the short term and replacing such resources when demand is restored in the future. Managers will be more inclined to keep excess resources if they expect future demand to restore sufficiently fast and if adjustment costs related to cutting resources and restoring them when demand rebounds are sufficiently high (Anderson en at, 2013). The agency explanation provides two predictions for cost stickiness. On the one hand, empire building incentives will motivate managers to keep excess resources, leading to greater cost stickiness (Chen et al, 2012). On the other hand, earnings management incentives will motivate managers to cut excess SG&A resources in order to meet earnings targets, resulting in lower cost stickiness (Kama and Weiss, 2012).

Cost stickiness i.e. the managers’ decision to bear the cost of unutilized resources in the case of falling sales affects current earnings as well as the expectation of future earnings. We can distinguish between efficient and inefficient cost sticky firms. A cost sticky firm is efficient if current sales fall but future sales are expected to rebound, while it is inefficient if sales are expected to decline permanently. Irrespective of whether the cost sticky firm is efficient or inefficient, cost stickiness has a negative effect on current earnings because the drop in sales is not compensated by an equivalent drop in costs (Homburg and Nasev, 2009).

1.2 PROBLEM DEFINITION

Prior literature focuses on economic and agency explanations for the cross-sectional variation in the degree of cost stickiness (Anderson, et al., 2007; Banker, et al., 2010,2011; Chen, et al., 2012; Dierynck, et al., 2012; Kama & Weiss, 2012). There is not much research yet conducted in the field of cost stickiness from a behavioral perspective in the accounting literature. In this thesis we try to fill in this gap by proposing “CEO overconfidence” as a behavioral explanation for cost stickiness.

(5)

There are many manifestations of overconfidence in the psychology literature. Firstly, people are miscalibrated (they overestimate the probability that their judgments are correct; Alpert & Raiffa, 1982) and suffer from positive illusions (Taylor & Brown, 1988) such as illusion of control (i.e. they overestimate their control over events; Langer & Roth, 1975), unrealistic optimism (i.e., they consider that bad events are more likely to happen to others than to themselves; Weinstein, 1980). Secondly, there is the ‘‘better-than-average’’ effect; this is when people overestimate their achievements and abilities relative to others (Svenson, 1981).

The above literature in psychology has been cited by behavioral finance researchers to support the claim that people are overconfident, and that this is likely to lead them to make errors in financial markets (Barber & Odean, 2000, 2001; Odean, 1998). Research has documented numerous costs of this overconfidence. The study of Carmerer & Lovallo (1999) indicated that managers take excessive risks in new ventures because they think they are more capable and knowledgeable than other managers. Furthermore, Malmendier & Tate (2005) documented CEOs to engage in too many acquisitions. Research of Goel and Thakor (2008) conclude that increases in managerial overconfidence lead to higher firm value but only up to a point. Their main result was that overconfidence destroyed firm value.

1.3 RESEARCH QUESTION

Prior findings indicate that cost stickiness has a negative effect on the current earnings of a firm because the drop in sales is not compensated by an equivalent drop in costs (Homburg and Nasev, 2009). But what if a firm has a CEO who is overly positive about his impact on restoring sales demand? What happens if a CEO overestimates the accuracy of his assessment of future demand? How will this overconfidence affect cost stickiness? Our study builds on the assumption that cost stickiness is influenced by overconfident CEOs. CEO’s overconfidence is argued to make them keep excess SG&A resources resulting in greater cost stickiness. Therefore, recognizing one’s limitations would help people set more realistic goals and select strategies that facilitate success. The overall objective of this thesis is therefore to examine the effect of CEO overconfidence on cost stickiness.

(6)

1.4 KEY FINDINGS

In this research we expect CEO overconfidence to increase the degree of SG&A cost stickiness. We argue that both overconfidence mechanisms; the better-than-average effect and miscalibration should affect the CEO’s assessment of future demand. Following prior literature we measure CEO overconfidence based on five different measures (Schrand and Zechman, 2012). These five measurements of overconfidence are: the CEO’s photo in the annual report, his/her cash as well as non-cash compensation relative to the average cash and non-cash compensation of the other executives of the firm, excess investments made and mergers and acquisitions where Malmendier and Tate (2008) found that overconfident executives are both more likely to overpay and to engage in value destroying acquisitions. Since we used different measurements for overconfidence we tested the hypothesis on each individual measure as well as on two composite measures of overconfidence. We tested our hypothesis based on a sample of 1628 observations for the fiscal years 2002-2009 in accordance with the cost stickiness base model of Anderson et al. (2003) and found that cost stickiness also existed in our sample. Afterwards we extended the baseline model with the aforementioned overconfidence measures in order to find out if cost stickiness increased with CEO overconfidence. Our results indicated that cost stickiness increased with CEO overconfidence when we tested for this relationship based on the CEO’s photo1 in the annual report which we used as an individual overconfidence measure in our sample. After we tested the baseline model which showed the presence of cost stickiness in our sample, we also tested the same model based on the overconfidence median cut2 and found that cost stickiness only showed up in firms with overconfident CEOs. Afterwards, we restricted overconfident CEOs to the top quartile of overconfidence in the sample and found the same results. Overall, the results show that cost stickiness is mainly associated with overconfident CEOs rather than less confident CEOs.

1_{The overconfident CEO will try to make himself stand out with his photograph in the annual report compared}

other executives as a strong declaration that he is more important, skilled and knowledgeable than other executives in the firm.

2_{We split our sample into two sub-samples whereas one sample included observations larger than the}

overconfidence median whereas the second sample included observations smaller than the overconfidence median.

(7)

1.5 ACADEMIC AND MANAGER IAL RELEVANCE

With this thesis we contribute to two streams of accounting literature. First, we extend the cost stickiness literature by providing a behavioral explanation for cost stickiness. This behavioral explanation differs fundamentally from the economic explanations suggested in prior research. While economic explanations focus on managers’ trade-off between costs and benefits associated with keeping excess SG&A resources, and assume unbiased managerial expectations, overconfidence in this thesis reflects a persistent managerial characteristic that indicates a positive bias in the CEOs expectations. Our explanation also differs from the agency-based explanations documented in prior literature. While CEOs driven by agency considerations to keep or cut excess resources for opportunistic reasons (to build empires or to manage earnings), the notion of overconfident CEOs in this thesis, is that these CEOs keep excess resources because they believe they are acting in the best interest of the shareholders.

Second, our study contributes to the accounting literature on overconfidence (Schrand and Zechman, 2012; Hribar and Yang, 2011; Hilary and Hsu, 2011; Libby and Rennekamp, 2012; Ahmed and Deullman, 2012). Our study extends the emerging literature on cost stickiness by documenting the effect of overconfidence on cost behavior and cost management by CEOs.

Furthermore, this study is not only important for academic researchers but also for professionals in companies who want to gain more knowledge in their cost behavior. Therefore with the results of this thesis we try to give companies more insights on the effects of overconfident CEOs on cost stickiness as this phenomenon is argued to negatively affect the current earnings of a company. Recognizing CEO’s limitations would help set more realistic goals and select strategies that facilitate success.

1.6 STRUCTURE OF THE THESIS

The second chapter contains the theoretical framework together with the hypotheses of this research. A conceptual model is proposed based on the relevant theories and gaps. The third chapter reveals the research design. The fourth chapter sets out the results of the different analyses used. The thesis ends with a final conclusion in which the study limitations and recommendations are discussed.

(8)

CEO Overconfidence and Cost Stickiness | 2.Theoretical framework 8

2. THEORETICAL FRAMEWORK

In this chapter the theoretical framework will be presented in which firstly most of the relevant literature for this research will be discussed in order to find out what is already known about the different variables that that will be used in this thesis. In the following part, the conceptual model will be presented based on the gaps and related opportunities from prior research. In the final part, theoretical models are used in order to present the hypotheses derived from the conceptual model.

2.1 COST STICKINESS

2.1.1 THE THEORY OF ASYMMETRIC COST BEHAVIOR

Traditional managerial accounting literature assumes that cost can be distinguished between fixed and variable, where variable costs are proportional to the volume change. The proportionality and symmetry between costs and activity implies that a 1% increase in activity results in a 1% increase in costs, and a 1% decrease in activity results in a 1% decrease in costs. In other words, the relation between variable costs and volume is symmetric for both volume increase and decrease. That costs might not be linear and proportional is now acknowledged in the literature. Some authors like Cooper and Kaplan (1998) state that costs raise more along with the increase in activity volume, than they fall when volume is decreased. Cooper & Kaplan (1998) render the traditional model obsolete, since the traditional model ignores the adjustment costs giving inaccurate results. Firms need to incur adjustment costs when they need to increase or reduce committed resources. For example, in the context of a firm’s decision to reduce its labor force, adjustment costs may include the costs of administering a layoff program, severance pay, and even litigation in case of a legal battle with the labor union. In addition, there are potential costs of ramping up the labor force again if the demand rebounds. A firm adding to its labor force faces adjustment costs that may take the form of recruitment of qualified workers, training of new employees, and organizational changes to accommodate the additions. To the extent managers perceive expected adjustment costs to be higher for a downward adjustment than for an upward adjustment, activity costs are likely to exhibit “sticky” behavior.

Anderson, Banker and Janakiraman (2003), hereinafter referred to as ABJ, provided the first evidence coming from a large sample showing that Selling, General and Administrative costs behave asymmetrically, contrary to the traditional model. They investigated the behavior of SG&A costs in relation to sales revenue activity. Using a sample

(9)

of 7,629 firms over a 20 year period, they hypothesized that cost stickiness is lower in the second year of consecutive sales declines and greater in periods of macroeconomic growth, and for firms with high intensity of assets or employees relative to sales. Their empirical evidence was consistent with the general proposition that the degree of cost stickiness is driven by the expected level of adjustment costs that vary systematically across firms and over time. They documented that the percentage increase in SG&A costs for an increase in sales revenue is larger than the percentage decrease in SG&A costs for an equivalent decrease in sales revenue. Specifically, their sample showed SG&A costs increase 0.55% per 1% increase in revenue but fell only 0.35% per 1% decrease in revenue. Motivated by the same resource adjustment based theoretical framework of cost behavior, Subramaniam and Weidenmier (2003) also confirm cost stickiness, finding that total costs increase 0.93% per 1% increase in revenues but decrease by 0.85% per 1% decrease in revenues. Furthermore, Calleja et al. (2006) have found that when sales revenue increase with 1%, SG&A costs go up with 0.97%. On the contrary, when sales revenue declines with 1%, SG&A costs go down by only 0.91%. This asymmetrical cost behavior is known as ‘cost stickiness’.

2.1.2 STREAMS SG&A COSTS STICKINESS

Prior studies on cost stickiness fall into two streams. The first stream of studies focuses on the impact of economic factors on the degree of cost stickiness. The economic explanation suggests that the decision to cut or keep SG&A resources when the sales level falls depends on the trade-off between managers’ expectations about the persistence of the demand decline and the magnitude of the adjustment costs associated with cutting SG&A resources in the short term and replacing such resources when demand is restored in the future. Managers will be more likely to keep excess resources if they expect future demand to restore sufficiently fast and if adjustment costs related to cutting resources and restoring them when demand rebounds are high (Anderson et al, 2003). If the fall in demand is perceived as temporary then one can expect higher cost stickiness since cost of adjustment might be higher than costs of unused capacity. Specifically, elimination of the resources (due to decline in sales) and then again their reacquisition (when sales are recovered) may result in higher costs. In this case managers will decide to retain unutilized resources rather than incur adjustment costs resulting in sticky costs (Anderson et al. 2003).

The second stream of studies focuses on the agency problem and investigates the impact of managerial incentives on cost stickiness. Managers that are not or not completely

(10)

monitored by their shareholders, make self-maximizing decisions that are not in the best interest of the shareholders (Anderson et al, 2003). This self maximizing behavior is known as “empire building” and is widely known as one of the agency costs, which are a result of the relation between shareholders that are the owners of the company, and managers that work for the company and need to be controlled (Eisenhardt, 1989). This empire building behavior includes decision-making by managers that aggressively grow the company above its optimal point and therefore destroying firm value and reduces firm performance (Hope & Thomas, 2008). In the case of empire building, managers hold excessive resources under their control, to increase their power, compensation, prestige and status within the company (Jensen, 1986). For example, when sales increase, managers will expand their staff3 in order to increase these sales furthermore and get appreciation for this higher performance. Because the manager holds more resources under his control, he is able to increase the sales and earn a higher compensation (Jensen, 1986). Empire building contributes to the stickiness of costs, because managers do not behave in resource disposing activities to avoid personal consequences like the downsizing of their division. Chen et al, 2012 provide evidence that empire building incentives, as captured by free cash flow, shifts SG&A cost asymmetry away from its optimal level. On the other hand, using data respectively from private Belgium firms and US public firms, Dierynck et al (2012) and Kama and Weiss (2012) both show that earnings management incentives will motivate managers to cut excess SG&A resources to meet earnings targets, resulting in lower cost stickiness.

2.1.3 PREVIOUS FINDINGS RELATED TO COST STICKINESS

Following ABJ, numerous studies have documented sticky costs in various contexts (e.g., Weiss 2010; Chen et al. 2012; Dierynck et al. 2012; Kama and Weiss 2013), and ABJ’s theory of cost stickiness has become dominant in research on cost behavior and its implications. Subsequent research has demonstrated that cost stickiness is pervasive across different cost categories and datasets, and has explored the implications of sticky costs for both financial and cost accounting. In order to get a view of what is already known about the main variable of interest in this thesis, cost stickiness, we provide an overview of the findings of prior research in table 1.

(11)

Table1 Previous findings related to Cost Stickiness Study Findings

Calleja et al (2006)

Authors conducted the research for US, UK, French and German companies during the 1988-2004 period. By applying ABJ’s methodology, authors found that: (1) operating costs are sticky in all four countries: operating costs increase, on average, by around 0.97% per 1% increase in revenues, but decrease by only 0.91% per 1% decrease in revenues; result is attributable to differences in corporate governance systems (common law vs. civil law) and managerial oversight; (2) in longer horizon (two year period) stickiness declined for US, UK and French companies, while increased for German companies; (3) stickiness is less pronounced for high revenue changes than for low revenue changes.

Medeiros & Costa (2005)

Analysis of costs stickiness for 198 Brazilian publicly listed companies in period 1986-2003. By replicating Anderson et al. methodology authors found that SA&G costs for sampled Brazilian companies were sticky; the authors found that SG&A costs increased by 0.59% per 1% increase in the sales but that the SGA costs decreased only by 0.32% per 1% decrease in sales. Surprising finding was the fact that cost stickiness increased when data was aggregated for two, three and four years, which means that cost stickiness gets worse in longer periods. Hypothesis on lagged adjustment of SG&A costs was rejected, while partial reversion hypothesis of stickiness was accepted.

Balakrishan et al (2004)

Research focus was on capacity utilization. Empirical analysis was done on the sample of 49 physical therapy clinics during the period 1994-1997. The authors proved that respond to decrease of activity should be higher than response to increase of activity if company is having excess capacity. Based on this finding, they concluded that ABJ’s study on cost stickiness should be interpreted with caution since cost stickiness may be feature only for the firms with strained current capacities.

Banker and Chen (2006)

Cross-country variations in the stickiness of operating costs are significantly associated with labor market characteristics (i.e. bargaining power of trade unions increases adjustment costs and is associated with higher cost stickiness).

Banker et al (2011)

Strong support to sticky costs framework across a wide range of countries, ranging from low- income developing economies (e.g. Indonesia and India) and middle- income economies (e.g. Malaysia and Thailand) to high- income developed economies, operating costs are on average sticky, degree of cost stickiness varies systematically with asset and employee intensity.

Banker et al (2010)

Long-termed trends in firm’s sales are mostly positive, thus managers tend to be optimistic about future activity and retain underutilized resources in case of current sales decreases. This optimism will manifest in cost stickiness.

Anderson et al. (2003)

Study also discovered that cost stickiness was: lower when company had successive revenue decrease (in time t and t-1); higher in years with growth of GDP; and higher in companies with higher assets and labor intensity. Cost stickiness is associated with both asset and employee-intensity.

Subramaniam &

Weidenmier (2003)

Authors explore weather cost stickiness is related with different ranges of activity changes. The use of ABJ model has resulted with finding that SG&A costs were stickier than COGS. Also, authors found that "sticky parameters" are not negative or significant for revenues change less than 10%, but beyond 10% change almost all parameters were negative and significant, cost stickiness is associated with both asset and employee-intensity.

Hamermesh & Pfann, 1996

The higher downward adjustment costs for capital are exacerbated by the lack of secondary markets for many capital goods, making cutting resources in response to a sales decline costlier than increasing resources by the same amount in an equivalent sales’ increase. Thus a manager will prefer to cut resources when sales decline than to increase resources when sales increase, leading to cost stickiness.

(12)

Table 1 Continued

Balkrishnan & Gruca (2008)

Use departments of Ontario hospitals as sample. Methodology was aimed to reveal differences in (operating) cost stickiness in one organization. Results: sticky costs in direct patient services exceed those in ancillary and support services, both because of the differential in the costs of adjusting capacity and because of the higher visibility of patient care related expenses. Therefore, authors conclude that core competences influence costs stickiness.

Kama and Weiss (2010)

In explaining costs stickiness phenomenon, the authors put focus on managers' intention to meet earnings target. Research was done on the sample of listed US companies for period 1979-2006 and the obtained results suggested that the incentives to meet earnings targets (to avoid losses and/or avoid earnings decreases) lead to deliberate resource adjustments that diminish cost stickiness.

Balakrishnan et al (2011)

Show that smaller companies tend to be less sticky than the larger ones. It is understood that the resources industry in the US in the event of a revenue decrease will demonstrate higher stickiness in costs than the average US firms’ stickiness in other industries, evidence support that cost stickiness is associated with asset- intensive firms and with high employ intensity.

He et al, 2010 Cost stickiness of SG&A costs for Japanese listed companies was explored for period 1975-2000. Results confirmed that stickiness reverses in subsequent periods and stickiness decreases with length of the data aggregation period. Research model also included dummy variable for Japan post bubble economy (1992-2000), which revealed that SG&A costs have become much less sticky in the post bubble economy era. Porporato and

Werbin, 2010

Application of cost stickiness behavior in banking sector was analyzed by the authors. Research was done on the sample of banks from Argentina, Brazil and Canada in period 2004-2009 and by replicating Anderson et al. model authors revealed cost stickiness in all three countries.

2.2 CEO OVERCONFIDENCE

2.2.1 DEFINING AND CONCEPTUALIZING OVERCONFIDENCE

The theory of overconfidence is a behavioral one. Overconfidence is a characteristic not only observed in the business world, but everywhere in everyday life. Prior research of Alicke & Govorun (2005) and Dunning et al (2004) documented that overconfident individuals assume they are better than other individuals, even when this is not the case. Generally, overconfidence is defined as inaccurate, overly positive perceptions of one’s abilities or knowledge (Moore & Healy, 2008). Overconﬁdence is conceptualized in different ways in the psychology literature. One is the ‘better-than-average’ effect. This is the tendency of individuals to think of themselves as ‘above average’ on positive characteristics (e.g., Svenson 1981; Alicke et al. 1985, Kruger 1999; Chen, 2013) and the other one is

miscalibration; when individuals overestimate themselves without comparison to others

(Moore, 1977; Chen, 2013). An outline of both overconfidence mechanisms can be found below.

(13)

Better-than-average effect

Overconfidence can arise through the ‘better-than-average’ effect; this is when individuals overestimate their ability or skills relative to others. An extensive experimental literature documents the tendency of individuals to consider themselves ‘above average’ on positive characteristics (e.g. Alicke et al., 1995; Kruger, 1999; Svenson, 1981; Alicke, 1985). Svenson (1981) for example, demonstrates that the majority of subjects rate their driving skills as ‘above average’. Svenson’s finding has been replicated numerous times in various countries and with respect to various IQ- or skill related outcomes other than driving. For example, when asking a sample of entrepreneurs about their chances of success, Cooper et al. (1988) found that 81% answered between 0 and 30% (with 33% attaching exactly zero probability to failure). However, when asked the odds of any business like theirs failing, only 39% of them answered between 0 and 30%.

The better-than-average effect is further reinforced by evidence on attribution bias, the self-serving inclination to attribute good outcomes to one’s performance and skills and bad outcomes to task difficulty or bad luck (Miller and Ross, 1975; Feather and Simon 1971). Larwood and Whittaker (1977) find that corporate executives are particularly prone to this form of self-serving bias. Because individuals expect their behavior to produce success, they attribute outcomes to their actions when they succeed and to bad luck when they fail (Feather and Simon, 1971; Miller and Ross, 1975).

The ‘better than average’ effect is particularly likely to apply to high-rank managers for a number of reasons. Firstly, Kruger (1999) and Camerer & Lovallo (1999) show that the effect is especially strong among highly skilled individuals, possibly due to insufficient weighting of the comparison group. If CEOs compare themselves to the average manager rather than other CEOs, they may conclude they are better than average at picking investment projects or merger targets. According to Malmendier and Tate (2005), the better-than-average effect is reinforced when highly skilled CEO’s compare themselves with other managers in the company, instead of other CEO’s, because these other internal managers are often in fact also really less skilled. This gives CEO’s the feeling that they are better than average.

Secondly, the effect tends to be strongest for outcomes that are abstractly defined

rather than in a one-to-one comparison with other people (Moore and Kim, 2003). CEOs will rarely have a direct comparison. Overestimation of themselves is often very difficult to measure for CEO’s, because of their complex decisions and overestimation is therefore difficult to compare with CEO’s of other firms. Decisions such as large-scale investments are

(14)

naturally complex and hard to compare across firms, making it hard to detect overestimation in contrast to salespeople who often learn of each other’s quarterly performance, allowing them to compare each other’s level of confidence to actual performance.

Thirdly, previous literature documents the tendency of individuals to be too optimistic

about their own future prospects (Weinstein, 1980; Kunda, 1987; Weinstein and Klein, 2002). Individuals are the most optimistic about outcomes which they believe are under their control (Langer, 1975). And individuals are more prone to overestimate outcomes to which they are highly committed (Weinstein, 1980). Top corporate managers are likely to satisfy both of these pre-conditions. First, a CEO has the ultimate say about his firm’s big strategic decisions and decides whether or not a large scale investment or a merger goes ahead. Such a position may induce the CEO to believe that he or she can also control the outcome, and thus to underestimate the likelihood of failure (March and Shapira, 1987). Second, a large portion of CEO compensation (stocks and options) depends on how well the company is doing. Similarly, the value of a CEO’s human capital (probability of firing, outside options) is tightly related to company returns. So, for compensation and career reasons alone, we would expect top executives to be highly committed to the outcome of their corporate decisions.

Miscalibration

Overconfidence can also arise through miscalibration; this is when individuals overestimate themselves without comparison to others (Moore, 1977; Chen, 2013). That is, they tend to overestimate the probability that their judgments are correct (Alpert & Raiffa, 1982). People are overconfident in the way they calibrate probabilities; they show excessive confidence in the precision of their estimates of probabilities. The calibration theory has demonstrated that people are overconfident in answering moderate to extremely difficult questions, and people are under-confident when they have to answer easy questions (Gervais, et al, 2003). People also overestimate their own skills and knowledge, and overestimate the precision of their information. Ben-David et al (2007) use the psychological finding of miscalibration to explain the behavior of corporate managers and their corporate policies. In their study, Ben-David et al. (2007) find that CFO’s of companies tend to hugely overestimate the precision with which they would be able to predict the future stock market performance of their company. This huge miscalibration by CFO’s proves the pervasive overconfidence at top level corporate decision makers.

(15)

2.2.2 PREVIOUS FINDINGS RELATED TO CEO OVERCONFIDENCE

In the finance and accounting literature, an overconfident manager is viewed as a manager who systematically overestimates future returns from the firm’s projects or equivalently systematically overestimates the likelihood and impact of favorable events on his/her firm’s cash flows and/or underestimates the likelihood and impact of negative (adverse) events on his/her firm’s cash flows (Chen et al, 2013; Heaton, 2002; Malmendier and Tate, 2005). A large number of studies in finance document the influence of CEO overconfidence on corporate and accounting decisions4. Doukas and Petmezas (2007) argue that overconfident CEOs feel that they have superior decision-making abilities and are more capable than their peers. The presence of these cognitive biases encourages CEOs to emphasize their own judgment in decision making and to engage in highly complex transactions such as diversifying acquisitions. Because of their overconfidence, these CEOs tend to underestimate the risks associated with a merger or overestimate the possible synergy gains from a business combination. Malmendier and Tate (2008) examine the extent to which overconfidence can help to explain merger decisions and various characteristics of the deal itself. They find that overconfident CEOs are more likely to pursue acquisitions when their firms have abundant internal resources. They further report that overconfident CEOs are significantly more likely than other CEOs to undertake a diversifying merger. Finally, they observe that overconfident CEOs use cash to finance their mergers more often than other CEOs. Recent accounting studies document that overconfidence affects Accounting and Auditing Enforcement Releases (Schrand and Zechman, 2012), management earnings forecasts (Libby and Rennekamp, 2012; Hribar and Yang, 2011), accounting conservatism (Ahmed and Duellman, 2012) and analyst earnings forecasts (Wong and Zhang, 2009). For example, using an abstract experiment and a survey of experienced financial managers, Libby and Rennekamp (2012) show overconfident managers are more likely to issue earnings forecasts because they are overly optimistic about firm performance and overconfident about their ability to predict future firm performance. Furthermore, Hribar and Yang (2011) provide empirical evidence that CEO overconfidence increases the likelihood of issuing a forecast, increases the amount of optimism in management forecasts, and increases the form and precision of forecasts.

4

Malmendier and Tate (2005) indicated that managerial overconfidence heightens the sensitivity of corporate investments to cash flow, especially for equity-dependent firms. Ben-David et al. (2007) documented that firms with overconfident CEOs made more investments, were more sensitive to the investment–cash relation, and had a lower cash dividend payout and higher long-term leverage Malmendier et al. (2010) found that overconfident managers were less likely to use external finance and issued less equity than other managers.

(16)

2.3 GAPS IDENTIFIED

Prior literature focuses on economic and agency explanations for the cross-sectional variation in the degree of cost stickiness. There is not much research yet conducted in the field of cost stickiness from a behavioral perspective in the accounting literature. In this thesis we try to fill in this gap by conceptualizing overconfidence according to the psychology literature, more specifically by the two overconfidence mechanisms; the better-than-average-effect and miscalibration as described in the previous paragraphs.

Our overconfidence explanation differs from the economic explanation provided in prior research. For example, Banker et al (2010) studied cost stickiness which was caused by managerial optimism. In their research, they expected cost stickiness to be dependent on a prior increase in sales while cost anti-stickiness to be dependent on a prior decrease in sales. These anticipations are reflections of optimal decisions with adjustment costs and the influence of previous changes in sales on the expectations of the manager regarding the changes in future sales. However, our notion of cost stickiness in this thesis differs from the aforementioned authors. Banker et al (2010) propose in their research managerial demand expectations to be informed by rational signals such as prior sales changes. While this economic explanation assumes unbiased management demand expectations, overconfidence in this thesis reflects a positive bias in CEO’s expectations because they believe that their behavior will produces success, resulting in the overestimation of returns from keeping excess SG&A resources leading to a higher degree of cost stickiness.

Our overconfidence explanation is also distinct from the agency explanation. The agency explanation states that on the one hand, managers with empire building incentives will prefer to keep excess SG&A resources to maximize private benefits from size related to power, status and prestige (Jensen, 1986), resulting in greater cost stickiness. On the other hand, managers with earnings management incentives will cut excess SG&A resources too quickly in order to meet earnings benchmarks, leading to lower cost stickiness (Dierynck et al, 2012; Kama and Weis, 2012). In this thesis cost stickiness is distinct from the agency explanation. Although both, empire builders and overconfident CEOs, have the tendency to avoid cutting excess resources, unlike empire builders who maintain excess resources for opportunistic reasons, overconfident CEOs, as defined in this thesis, believe they are acting in the best interest of the shareholders and therefore maintain excess resources. These CEOs also believe that their actions will produce positive results because they show excessive confidence in the precision of their estimates of probabilities.

(17)

2.4 HYPOTHESES DEVELOPMENT

Cost stickiness depends on the CEO’s expectations about the persistence of the

demand decline and the magnitude of the adjustment costs associated with cutting SG&A

resources in the short term and replacing such resources when demand is restored in the future.

The decision to cut or keep excess SG&A resources when sales decline depends on a large part on the CEO’s expectation about future demand. CEOs will be more inclined to keep excess resources if they expect future demand to restore sufficiently fast. Drawing on the psychology literature on overconfidence, we expect CEO overconfidence to increase the degree of cost stickiness. We argue that both overconfidence mechanisms, the better-than-average-effect and miscalibration, should affect managers’ assessment of future demand. The better-than-average-effect implies that overconfident CEOs will be overly positive about their impact on restoring sales demand. Consequently, they will overestimate the likelihood of a sales rebound in the near future, which will motivate them to retain excess SG&A resources when sales decline, leading to greater cost stickiness. In addition, the miscalibration implies that overconfident CEOs will overestimate the accuracy of their assessment of future demand, which will also increase the probability of retaining excess SG&A resources, also resulting in greater cost stickiness. Taken together, both behavioral mechanisms will bias managers’ expectations about the likelihood of a future sales rebound upward. Hence, overconfident CEOs should be more likely to retain excess SG&A resources, resulting in greater SG&A cost stickiness.

CEOs will also be more inclined to keep excess resources if adjustment costs related to cutting resources and restoring them when demand rebounds are sufficiently high. Adjustment costs are mostly determined by external factors such as the difficulty of hiring or laying-off employees and the difficulty of acquiring or disposing of equipment. We cannot make assumptions about whether overconfident and non-overconfident CEOs assess adjustment costs differently because there is little theory to guide that argument. If CEOs strongly underestimate adjustment costs, we would find lower stickiness. However, in this thesis we argue in the same vein of previous research of Anderson et al (2013) that if the fall in demand is perceived as temporary then one can expect higher cost stickiness since cost of adjustment might be higher than costs of unused capacity. As mentioned before, overconfident CEOs are positive about their impact on restoring sales demand. In this case we argue that CEOs are

(18)

motivated to retain the unutilized resources rather than incurring adjustment costs resulting in cost stickiness.

H1: SG&A cost stickiness increases with CEO overconfidence.

2.5 CONCEPTUAL MODEL

By combining all the aforementioned gaps and related opportunities of prior research a conceptual model can be developed. We predict that CEO overconfidence will positively affect cost stickiness. The decision to cut or keep excess SG&A resources when sales decline depends on a large part on the CEO’s expectation about future demand. We expect that overconfident CEOs will be overly positive about their impact on restoring sales demand and/or they will overestimate the accuracy of their assessment of future demand; that is, they tend to overestimate the probability that their judgments are correct (Alpert & Raiffa, 1982). Consequently, they will overestimate the likelihood of a sales rebound in the near future, which will motivate them to retain excess SG&A resources when sales decline, leading to greater cost stickiness. CEOs will also be more inclined to keep excess resources if adjustment costs related to cutting resources and restoring them when demand rebounds are sufficiently high. In this case we argue that CEOs are motivated to retain the unutilized resources rather than incurring adjustment costs resulting in cost stickiness. See figure 1 for the conceptual model.

CEO Overconfidence

Cost Stickiness + H1

(19)

CEO Overconfidence and Cost Stickiness | 3.Methodology 19

3. METHODOLOGY

This chapter gives an overview of the research method that is used in this study. The first part comprises the data and sample collection. The second part describes the measurements of the dependent and independent variables followed by the plan of analysis. The final part outlines the model specification.

3.1 DATA AND SAMPLE COLLECTION

Data Collection

The sample for this study consists of 257 UK listed firms (including 217 FTSE 350 and 40 SmallCap firms). The data set consists of UK firms because previous studies have extensively focused on US firms regarding the cost stickiness researches. The starting sample exists of 1840 firm-year observations. The examination period covers the fiscal years 2002-2009. Data was collected from several sources. The first source was a unique UK data set provided by the thesis supervisor. The second source was Compustat and the third source was Orbis. See table 2 for a specification of sources where the different variables were derived.

Sample Collection

The sample used in this thesis covers the period 2002-2009 in accordance with the unique data set provided by the thesis supervisor. We start with an initial sample of 1840 firm-year observations for the measurement of the dependent variable and independent variable. Next, we combine the unique data set with the data set obtained for the control variables from Compustat resulting and drop unmatched data, resulting in 1748 observations. Afterwards we dropped missing data which resulted in only 557 observations. Missing data arise in almost all serious statistical analyses. In order to save the sample size we therefore made use of mean imputation. Mean imputation is the replacement of a missing observation with the mean of the non-missing observations for that variable and were therefore able to keep the 1748 observations. Furthermore, working with the paper of Anderson and Lanen, 2007, we also require SGA costs to be smaller than sales resulting in 1693 observations. Required for the log Table 2 Data Sources

Variables Unique data set Compustat Orbis Dependent: Cost Stickiness X X Independent: CEO Overconfidence X X X Control variables X X

(20)

transformation, is to delete observations with non-positive amounts for either sales revenue or SG&A. In order to minimize the effects of outliers, we wonsorized top and bottom 1% of the observations with extreme values in the SG&A costs and sales revenue resulting in a final data set of 1628 observations. Table 3 details the sample procedure.

Table 3 Sample selection

Procedure Observations deleted

Starting data set fiscal year 2002-2009 1840

Less: unmachted data after merging initial data set with data set from Compustat

(92) Less: observations for which sales revenue is smaller than SG&A

costs in current year

(55) Less: observations with non-positive amounts for either sales revenue

or SG&A.

(65)

Final sample: 1628

3.2 MEASUREMENT VARIABLES

3.2.1 MEASUREMENT COST STICKINESS

The original model developed by Anderson et al. (2003) will be used for the measurement of cost stickiness:

ΔlnSGAi,t = 0 + 1ΔlnSalesi,t + 2DecrDumi,t + 3DecrDumi,t*ΔlnSalesi,t + Ɛi,t (1)

where i represents a firm index and t a time index, ΔlnSGAi,t = ln(SGAi,t/SGAit-1), ΔlnSalesi,t = ln(Salesi,t/Salesit-1), SG&A is Selling, General and Administrative costs (#189 Compustat), Sales is net sales (#12 Compustat), DecrDum is one if sales in t are lower than sales in t-1 and zero otherwise. According to Anderson et al. (2003), the log specification of ln(SGAi,t/SGAit-1)

and ln(Salesi,t/Salesit-1) improves comparability of the variables across firms and alleviates

potential hetroskedasticy (unequal variances of residuals).

The coefficient ß1 measures the percentage increase in SG&A costs with a 1% increase in sales. Since the value of DecrDum is 1 when revenue decreases, the sum of the coefficients (ß1+ ß3) measures the percentage decrease in SG&A costs with a 1% decrease in sales. A positive and significant coefficient ß1 and a significantly negative coefficient ß3 are consistent with cost stickiness, indicating a smaller cost reaction when sales decline.

3.2.2 MEASURES OF CEO OVERCONFIDENCE

For the main analysis we use five overconfidence measures. The choice for these measures was based on prior research and data availability (Schrand and Zechman, 2012; Jia

(21)

et al, 2013; Malmendier and Tate, 2008). An outline of these the five overconfidence measures can be found below.

Two measures of relative compensation

The most revealing manifestation of CEOs overconfidence is their relative compensation to other executives in the firm. CEOs are known to have considerable influence in the setting of their own compensation and they have even more control over the compensation of other executives (Hambrick and Chatterjee, 2007; Bebchuk and Fried, 2004; Tosi and Gomez Mejia 1989). As mentioned previously, the overconfident CEO believes that he is far more skilled and knowledgeable relative to others and therefore more valuable than anyone else in the firm, and this then becomes reflected in the CEO’s compensation relative to others. A large pay gap reveals the CEO’s beliefs that executives vary widely in their contributions (Hayward and Hambrick, 1997). Higher relative pay could be a source of overconfidence causing the CEO to believe that he is better than average (Schrand and Zechman, 2012).

In this thesis we use two measures of CEO’s relative compensation. Relative cash compensation (REL_CASH) equals the CEO’s cash compensation (salary and bonus) divided by the average cash compensation of all other executives in the firm. Relative non-cash compensation (REL_NONCASH) is the CEO’s non-cash compensation (equity linked compensation) divided by the average non-cash compensation of the other executives in the firm. A large pay gap between the CEO and the other executives in the firm confirms the CEO to belief that he is better than average.

Prominence of the CEO’s photograph

The second measurement of CEO overconfidence is based on the prominence of the CEO’s photograph in the annual report (PHOTOSCORE). The annual report is a source for the CEO to present himself or herself as the leader of the firm. Hambrick and Chatterjee (2007) documented that CEOs paid a lot of attention to the design as well as the content of the annual report; in particular CEOs were very controlling over how they were presented in the annual report. We can expect that the overconfident CEO will try to make himself/ herself stand out or more visible than other executives in the annual report as a strong declaration that he/she is more important, skilled and knowledgeable than other executives in the firm. CEOs with a single photo should arguably be more overconfident than those with a collective photo. The greater the visibility of the CEO’s photographs in the annual report the more confident

(22)

the CEO is about himself. We rated this indicator of overconfidence as follows (Schrand and Zechman, 2012):

Table 4 Prominence CEO photograph

Label Rate

At least half a page & no other individuals 4 Less than half a page & no other individuals 3 There are other individuals with the CEO 2

No CEO photo 1

Excess- Investment

The third measurement of CEO overconfidence is the industry-adjusted excess investment, which is the firm’s residual from a regression of total asset growth on sales growth less the industry median residual (XSINVEST_INDADJ). Ben-David et al. (2007) document excess investment on average by overconfident executives consistent with the prediction that overconfident managers overestimate the cash flows of an investment project and/or underestimate the risk of the payoffs. Furthermore, prior research indicated that executives who are overconfident invest more when there are sufficient internal resources available to fund the investment but that these overconfident executives limit the investment when external resources are necessary. If excess investment is greater than the industry median (i.e. XSINVEST_INDADJ greater than zero) for that year, this indicates CEO overconfidence.

We perform the following steps in order to obtain XSINVEST_INDADJ:

 The annual firm asset growth rate (TAgrowth), is calculated using the year-on-year percentage change in total assets (#6 Compustat) and is calculated as below:

= TAt (Data6) – Tat-1(Data6) TAt-1(Data6)

 The annual sales growth (SALESG) is calculated in the same way as the asset growth by using data item 12 in Compustat.

 Afterwards we run an OLS regression that included the one year % change in total assets (total asset growth) on the one year % change in sales (sales growth). The regression was run by year and included all firms in all industries.

 For each firm we obtained its estimated residual from the OLS regression.

 We then computed the median of all the residuals for firms in the industry. Industry is defined at the 2-digit level.

 Lastly, we subtracted the industry median residual from the firms residual. TAgrowtht =

(23)

Acquisitions

The fourth and final measurement used in this thesis for overconfidence is the industry-adjusted net dollars of acquisitions made by the firm, obtained from the statement of cash flows (ACQUIRE_INDADJ). With respect to merger and acquisition activity, Malmendier and Tate (2008) find that overconfident executives are both more likely to overpay and to engage in value destroying acquisitions. Acquisitions (#129 Compustat) by the firm in excess of the industry median for the year suggest overconfidence.

We perform the following steps in order to obtain ACQUIRE_INDADJ:

 We obtain the data regarding the acquisitions made by the firm which was obtained from the statement of cash flows.

 Next we obtain the industry-adjusted net dollars of acquisitions by subtracting the median industry net dollars acquisitions from the firm net dollars acquisitions. Industry is defined at the 2-digit level.

The aforementioned five overconfidence variables differ in measurement. Since we have multiple dimensions of CEO overconfidence we will standardize each overconfidence dimension and combine the equally weighted 5 standardized individual overconfidence measures into one composite measure: zOC = zREL_CASH + zREL_NONCASH + zPHOTOSCORE + zXSINVEST_INDADJ + zACQUIRE_INDADJ.

We will make an alternative composite measure of CEO overconfidence by performing Principal Component Analysis (PCA) to determine if the aforementioned items fall into one composite measure. We will also include the results of this alternative composite measure in the main analysis.

3.4 EMPIRICAL MODEL

To test whether CEO overconfidence increases cost stickiness the original model of Anderson et al. (2003) is extended by including the corresponding three-way interaction term (OC* DecrDumi,t* ΔlnSalesi,t):

ΔlnSGAi,t = o + 1ΔlnSalesi,t+ 2DecrDumi,t + + 3DecrDumi,t*ΔlnSalesi,t+ 4OC + 5OC* ΔlnSalesi,t + 6OC* DecrDumi,t + 7OC* DecrDumi,t* ΔlnSalesi,t + Ɛi,t (2)

(24)

We will perform regression analysis to test whether CEO overconfidence increases cost stickiness. The Beta of interest in model 2 is 7 which represents the three-way interaction effect (OC* DecrDumi,t* ΔlnSalesi,t).

3.3 CONTROL VARIABLES

In this thesis we follow prior literature and include two sets of control variables. These two sets are: Economic and Agency variables: We control for two economic factors that may affect SG&A cost asymmetry (Anderson et al, 2003). We control for employee and asset intensity. As proxies for adjustment costs, both should increase the degree of cost stickiness. Employee intensity (EMPLINT) is calculated as the number of employees (#29 Compustat) divided by sales. Asset intensity (ASSETINT) is calculated as total assets (#06 Compustat) divided by sales.

We control for the agency factor free cash flow (FCF), which is calculated as cash flow from operating activities (#308 Compustat) in year t less capital expenditures (#128 Compustat) scaled by current assets (#4 Compustat). High levels of FCF allow CEOs to overinvest in SG&A when demand increases and to postpone SG&A cost cuts when demand decreases. Hence, higher levels of FCF should increase cost stickiness (Chen et al, 2012). When we include the economic and agency factors we get the following equation:

ΔlnSGAi,t = 0 + 1ΔlnSalesi,t + 2DecrDumi,t + + 3DecrDumi,t*ΔlnSalesi,t+ 4OC + 5OC*

ΔlnSalesi,t + 6OC*DecrDumi,t + 7OC*DecrDumi,t*ΔlnSalesi,t + 8EMPLINT + 9EMPLINT*

ΔlnSalesi,t + 10EMPLINT*DecrDumi,t + 11EMPLINT*DecrDumi,t*ΔlnSalesi,t + 12ASSETINT + 13ASSETINT*ΔlnSalesi,t + 14ASSETINT*DecrDumi,t + 15ASSETINT*DecrDumi,t*ΔlnSalesi,t + 16FCF + 17FCF* ΔlnSalesi,t + 18FCF* DecrDumi,t + 19FCF*DecrDumi,t* ΔlnSalesi,t + Ɛi,t (3)

(25)

CEO Overconfidence and Cost Stickiness | 4 Results 25

4 RESULTS

In this chapter the results of this study are presented. In the first section we will provide the descriptive results. The second section will outline the result of the hypothesis test, followed by the last section which will outline the robustness test.

4.1 DESCRIPTIVE RESULTS

Panel A of table 5 provides descriptive statistics on annual sales revenue and SG&A costs for the firms within the sample. These firms have a mean value of £2730.93 billion in annual sales revenue, where the median is £863.60 billion. On average, the SG&A costs for these firms are £455.15 billion, with a median of £136.69 billion. The average sales ratio (sales of year t divided by sales of year t-1) is 1.48 and that of the SG&A ratio is 9.16. SG&A costs as a percentage of sales, averages at 22.21 % with a median of 19.05%.

Panel B describes the statistics related to the overconfidence variables. The CEO photo score has a mean value of 2.67 and a median of 3. On average, the relative compensation which equals the CEO’s cash compensation divided by the average cash compensation of all other executives in the firm has a value of £7.80 and the relative non-cash compensation has a value of £6.49 on average.

See the Appendix 1 for the variable definitions

Table 6 summarizes the sample composition by industry based on a classification by two-digit SIC codes. The manufacturing (34.59%) and services (21.68%) sectors were the most heavily represented industries in our sample.

Table 5 Descriptive statistics

Panel A: Sales and SGA costs Mean SD 10% 50% 90%

Sales (£bill) 2730.93 4975.94 128.07 863.60 7797.70

SGA costs (£bill) 455.15 1051.59 17.97 136.69 795.60

Salesi,t/Salesit-1 1.48 4.10 0.64 1.08 1.47

SGAi,t/SGAit-1 9.16 167.57 0.58 1.06 1.83

SGAi,t/ Salesi,t (%) 22.21 17.03 3.96 19.05 44.66

Panel B: Overconfidence variables PHOTOSCORE 2.67 0.96 1 3 4 REL_CASH 7.80 48.90 0.480 1.41 7.80 REL_NONCASH 6.49 36.30 0.47 1.39 6.45 XSINVEST_INDADJ 0.02 252.30 −73.80 −11.14 15.75 ACQUIRE_INDADJ 0.00 1044.52 −197.53 0.73 90.68

(26)

Table 6: Sample Distribution by Industry (N = 1628)

Industry SIC N % Agriculture-Forestry-Fishing-Hunting 00–09 3 0.18 Mining-Construction 10-19 224 13.76 Manufacturing 20-39 563 34.59 Transportation 40-49 191 11.73 Wholesale-Retail trade 50-59 291 17.88 Finance-Insurance-Real estate 60-67 3 0.18 Services 70-89 353 21.68 Total 1628 100

Table 7 represents the Pearson Correlations coefficient. We will discuss the correlations between the main variables of interest in this research section. Bold coefficients indicate significance at or below the 5% level. As expected, a significant positive correlation is present between the natural logarithm of the change in sales revenue and the natural logarithm of the change in SGA costs (0.55), suggesting that changes in sales revenue and changes in SGA costs move in the same direction. Furthermore, a significant positive correlation (0.18) is present between the natural logarithm of the change in sales revenue and overconfidence, suggesting that changes in sales revenue and CEO overconfidence move in the same direction. In other words, we find a (small) strength of association between CEO overconfidence and the change in lnSales.

Lastly, visual inspection of the correlations does not seem to warrant any concerns with respect to multicollinearity. Typically, correlations above 0.7 complicate a regression analysis. However, as a more formal test for multicollinearity, we conducted multicollinearity analysis for the independent and control variables. The tolerance of each of these variables, defined as the inverse of the variance inflation factor (1/vif) is higher than 0.1, so we can conclude that multicollinearity is not a concern in our regression models (Belsley, Kuh, & Welsch, 1980).

Bold coefficients indicate significance at or below the 5% level. See the Appendix 1 for the variable definitions.

Table 7 Pearson Correlation matrix

Variable V1 V2 V3 V4 V5 V6 V1: ΔlnSalesi,t 1.0000 V2: ΔlnSGAi,t 0.5556 1.0000 V3. zOC 0.1892 0.1141 1.0000 V4. EMPLINT −0.0613 −0.0180 −0.0687 1.0000 V5. ASSETINT −0.0176 −0.0019 −0.0212 0.0354 1.0000 V6. FCF 0.0334 0.0442 −0.0231 −0.0494 −0.1610 1.0000

(27)

4.2.3 PCA ANALYSIS

In this section we perform PCA analysis5. We do this in order to discover or to reduce the dimensionality of the data set and too identify new meaningful underlying variables. We performed PCA by working with the five independent variables that measure CEO overconfidence (REL_CASH, REL_NONCASH, PHOTOSCORE, XSINVEST_INDADJ, and ACQUIRE_INDADJ). Based on the results of the PCA and the screeplot 6 we based one component on the variables REL_CASH and REL_NONCASH and the remaining variables (PHOTOSCORE, XSINVEST_INDADJ, and ACQUIRE_INDADJ) are no longer are part of the component. We did this based on table 8 which indicates that the coefficients of the aforementioned three variables did not load well on the component compared to the coefficients of the two variables REL_CASH and REL_NONCASH which did load well on the component (>0.3). We also looked at the KMO7 results. As can be seen from table 9 the KMO has an overall value of 0.5007. The literature states as a rule of thumb, that this value should be above 0.500 in order to justify the principal component analysis. As the KMO is not below 0.500 we will continue to work with the new composite measure of CEO overconfidence and perform the regression analysis again in order to see in cost stickiness increases with CEO overconfidence.

Table 8 Principal Components

Variable Comp1 Unexplained

CEOPHOTO −0.0423 0.0073

REL_CASH 0.7055 0.2405

REL_NONCASH 0.7061 0.2393

XSINVEST_INDADJ −0.0385 0.9977

ACQUIRE_INDADJ −0.0222 0.9992

See the Appendix 1 for the variable definitions

4.2 MAIN RESULTS

4.2.1 COST STICKINESS

Essential for the analysis of the hypotheses is the presence of cost stickiness. If firms exhibit sticky cost behavior of SG&A costs, ß1 and ß3 will be positive and negative respectively and both significant as well. The regression results in table 10 indicate a

5_{Principal component analysis (PCA) involves a mathematical procedure that transforms a number of (possibly)}

correlated variables into a (smaller) number of uncorrelated variables called principal components.

6_{See appendix 2}

7

Kaiser-Meyer-Olkin measure of sampling adequacy

Table 9 KMO Variable kmo CEOPHOTO 0.5117 REL_CASH 0.5004 REL_NONCASH 0.5004 XSINVEST_INDADJ 0.5688 ACQUIRE_INDADJ 0.5084 Overall 0.5007