• No results found

The effect of a competitive environment on cost stickiness

N/A
N/A
Protected

Academic year: 2021

Share "The effect of a competitive environment on cost stickiness"

Copied!
45
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The effect of a competitive environment on cost stickiness

Name: Gijs-Bertus Jan (Stefan) Kas Student number: 11418427

Thesis supervisor: dr. A. Sikalidis Date: June 25, 2018

Word count: 13,922

MSc Accountancy & Control, specialization Control

(2)

Statement of Originality

This document is written by student Gijsbertus Jan Kas who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is 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.

(3)

Abstract

Previous literature showed the existence of cost stickiness and its firm-specific determinants. This paper however, focuses on an external factor that could influence the degree of cost stickiness, namely a competitive environment. Based on an ordinary least squares regression, I find strong evidence that cost stickiness is present at S&P 1500 firms between 2007 and 2017. More importantly, this study finds evidence that a competitive environment increases the degree of cost stickiness. Furthermore, no evidence is found that degree of cost stickiness differs between a leading or a following company in an industry. However, this study finds evidence that industries differ from each other in the degree of cost stickiness.

(4)

Contents 1 Introduction ... 1 2 Literature review ... 3 2.1 Cost stickiness ... 3 2.2 Competitive environment ... 5 2.3 Hypothesis development ... 7

3 Sample and research ... 10

3.1 Sample selection ... 10 3.2 Measurement of Variables ... 12 3.2.1 Cost stickiness ... 12 3.2.2 Competitive environment ... 13 3.2.3 Control variables ... 14 3.3 Empirical model ... 14 4 Results... 17 4.1 Descriptive statistics ... 17 4.2 Correlation test ... 20 4.3 Multivariate analysis ... 20 4.3.1 Cost stickiness ... 20

4.3.2 Competitive environment effect on cost stickiness ... 22

4.3.3 Leading or following company effect on cost stickiness ... 24

4.3.4 Industry effect on cost stickiness ... 26

4.4 Robustness check ... 29

5 Conclusion ... 34

References ... 37

Appendices ... 41

(5)

1

1 Introduction

Costs in the management accounting literature are an important element in the different analysis. Therefore, understanding their behaviour is crucial. In for example, every profit analysis or budgeting processes costs are an important part. Alongside, of these cost, also the development of them are important for the analysts. It gives the analyst a better forward-looking understanding of the cost and a better understanding how the company deals with them. This is relevant because investors, internal managers, auditors and analysts trust on the cost accounting data for their analysis (Baumgarten, 2012).

Most widely known is the difference between fixed and variable costs. Fixed costs are: a cost that does not vary in the short run with a specified activity (Atkinson, Kaplan, Mutsumura, Young 2012 p. 90) and variable cost are: a cost that increases proportionally with changes in the activity level of some variable (Atkinson et al., 2012 p. 88). However, numerous studies deviate from this symmetrical cost behaviour idea. They found that the cost decrease less with a negative change in activity than they do with positive change in activity (Calleja, et al. 2006; Subramaniam, Weidenmier 2003). This is known as the assymetrical cost behaviour, or nowadys as the theory about cost stickiness (Anderson et al. 2003).

The presence of cost stickiness does have an effect on multiple things. Weiss (2010) found in his research that cost stickiness influences analyst earnings forecasts. Firms with more cost stickiness got a less accurate analysis then firms with less stickier costs. This risk of imperfect forecast could be decreased if the analysts incorporates the issue of cost stickiness (Banker, Chen 2006), but therefore understanding of this topic is necessary.

Most of the previous literature on cost stickiness have focused on cost asymmetry determinants within a firm (Anderson et al. 2003; Banker et al. 2014; Chen et al. 2012). Two more recent papers, however, focuses more on external factors. Banker et al. (201) investigated the stickiness of cost in highly strict countries regarding the employment protection legislation provision. The other study researched how a fixed-price regulation increases a firms’ cost elasticity and decrease cost asymmetry (Holzhacker, et al. 2015). This study focuses on another external factor, namely a competitive environment.

Wu et al. (2015) find that companies within a competitive industry have different strategies from one another. From this strategy managers make decisions about future investment and their market share. Also, a competitive environment has its effects on the productivity of a company, which increases the resources (Akdoğu, MacKay 2008). This could result in different degrees of cost stickiness, which makes it an interesting subject to investigate.

(6)

2 The results of this thesis should be of relevance to analyst, investors, managers and auditors, who will get a better understanding of the external factors on cost stickiness. Also, this paper contributes to the knowledge about cost stickiness, because it will give new insides in this phenomenon. It widens the view on the potential determinants of cost stickiness. It specifically broadens the view of external factors on the stickiness of cost.

This paper is structured as follows: the second section concerns the literature review and the hypothesis development. Afterwards, the empirical model and the used sample is described in section three. Next, in section four, the regression results are described and lastly, in section five, the conclusion and implications can be found.

(7)

3

2 Literature review

2.1 Cost stickiness

This study is primarily focused on the theory about cost stickiness, also known as the asymmetric cost phenomenon. The thoughts about costs have changed. In the past, everybody assumed that the traditional cost system was the correct one. In this system, they expect that the variable costs are flexible and react the same with an increase and a decrease in volume (Noreen 1991). In other words, variable costs change proportionally with the change in volume. However, around the turn of the century, researchers found that this is not always the case. Cooper and Kaplan (1998) and Noreen and Soderstrom (1997) introduce a different cost behaviour model. They find that with an activity volume increase cost increase more, but with a decrease they decrease less. This is then known as the asymmetric cost model. The researchers who introduced the term “cost stickiness” were Anderson, Banker and Janakiraman (2003, p.48) and they defined it as: “Costs are sticky if the magnitude of the increase in costs associated with an increase in volume is greater than the magnitude of the decrease in costs associated with an equivalent decrease in volume”.

Anderson et al. (2003) were the first who investigated this subject using a large sample. Their sample consisted of 7,629 firms. They found that selling, general and administrative (SG&A) costs increase with 0.55% per 1% increase of sales revenue, which is a proxy for operational activities, while there would be a decrease of only 0.35% per 1% decrease of sales revenue. They chose for SG&A costs because these costs can be logically linked with the revenue of a company.

After the study of Anderson et al. (2003), the research on cost stickiness became broader; also, the stickiness of other costs is investigated. In 2003 the existence of cost stickiness was proved using the total costs of a company. The total costs consist of the SG&A costs and the Cost of goods Sold (COGS). Subramaniam and Weidenmier (2003) found that an increase of 1% in revenues leads to increase in total costs of 0.93%, but at a decrease of 1% of revenues leads to a decrease of only 0.85%. This asymmetrical behaviour only remains when the revenue changes are higher than 10%. With small revenue changes the authors did not find cost stickiness. This remain for both costs.

In 2006 the study of Calleja, Steliaros and Thomas investigated the asymmetric cost behaviour in four countries; United Kingdom, United States of America, France and Germany. In this research the stickiness of operating cost was investigated. Their results show differences between the cost stickiness in the four countries, but their overall conclusion is that an increase

(8)

4 of 1% in revenue leads to an average increase of operating cost by around 0.97%, but a decrease of 1% in revenue leads to an average decrease of operating cost by 0.91%.

Dierynck, Landsman and Renders (2012) look at labor costs in private Belgian firms. These costs should have a more variable characteristic than SG&A costs. Their findings are that an increase or decrease of 1% in revenue leads to a respective increase and decrease of 0.60% and 0.34%. This asymmetric behaviour becomes less when firms just beat the zero-earning benchmark. Still the results indicate that both variable and fixed cost show forms of cost stickiness.

So far, all the studies looked at the costs for the organization as a whole, but the

stickiness also differs across all the departments of an organization (Balakrishnan, Gurca 2008). This difference occurs because some departments are more important for the core business than others and will wait longer before they cut their costs.

From all the previous studies about cost stickiness researchers have found enough evidence for the existence of this phenomenon. The differences occur, because the cost stickiness has multiple determinants and drivers that affect the asymmetrical cost behaviour.

First of all, the research of Anderson and Lanen (2009). They looked at different industries and the effect on different firms. They found that the cost stickiness differs. In other words, they found that the cost behaviour differs through the firm type, geographical situation, cost account, and industry.

After their findings Anderson et al. (2003) have given explanations why this asymmetric behaviour occurs. An explanation is that managers need to make adjustments costs when cutting their overall costs. These adjustments costs happen right now and in the future. An example of these cost could be paying firing costs when an employee is dismissed or extra hiring cost when the company is looking for new employees. Therefore, managers are always looking for the decisions with the best cost efficiency. These decisions are based on previous knowledge and sales demand. Banker et al. (2014) stated that in period with high previous sales managers are more positive about the future. Even when the sales in the current period decrease, they will not cut their costs.

In 2011 Yasukata also studied the managers optimism and the effect on cost stickiness. They found that the decisions managers make, have an effect on cost stickiness. To test this, they used the sales forecast and found that the likelihood of improved sales, has an effect. When the forecast states that future sales increase managers keep resources at the same level, although current sales decrease. Otherwise, there would not be enough resources in the future to succeed in the increasing sales demand. The result is an increase in cost stickiness. This happens, because

(9)

5 managers focus on the profit for the long term. The long term is more important, causing

managers to be reluctant in cost cutting. The effect is more pronounced for the SG&A costs. Another determinant is the decisions made in the past about technology questions. It is not solely about the managers decisions, also choices made in the past about technology have an impact on the asymmetrical cost behaviour. These decisions are made in advance and are most of the times based on forecasts. When such choices are made, this will influence the costs stickiness in the future. When starting an investment or project, companies assume it will have positive influence in the future, therefore, they keep investing in this project. The previous costs are not seen as sunk costs. It is not always the case that technology investment result in a higher cost stickiness. The findings are less obvious when a manager needs to meet an earnings target. Then the cost stickiness will be lower (Kama, Weiss 2010; Diernynck, Landsman, Renders 2012).

In 2014 Cannon investigated the link between the selling price and the cost stickiness. Therefore, the author looked at the airline industry in America. The results show that when the demand differs in a period, managers change the selling price until it matches the demand, rather than cut down the capacity. In other words, when the demand decreases, the selling price will be cutting down by the manager until the return of the lower selling price is higher than if the capacity would be reduced. The same idea holds for an increasing selling price. Managers then will add capacity until the return of those extra sales is higher than the profit from the increasing selling price. So, managers need to consider how they should react when the demand differs.

In 2013 Banker, Byzalov and Chen investigated if cost stickiness is also influenced by external factors. For their research the look at the employment protection legislation in multiple countries. They find that companies who are settled in a country with strict legislation have a higher degree of cost stickiness than companies who are settled in less strict countries. Their findings are in line with the adjustment cost argument, because the decision of the manager depends on the extra cost the legislation could have for the company.

2.2 Competitive environment

As stated above external factors can harm the stickiness of cost. Another external factor almost all companies have to face is competition within the market. The definition of competition which nowadays is still used comes from Porter (1979). It refers to the intensity of the rivals within in an industry or a market. According to his model, this intensity is one of the main forces that shapes such a market. Before Porter, Stigler carried out a study in 1958 about the effects competition has on a company. Even then the results showed that such an environment has an effect on the prices and on the economic efficiency of a company.

(10)

6 Jan Boone (2001) finds in his research that companies within a competitive environment have more incentives to invest more in R&D, even with a low demand. A competitive market arises when products become more substitutable and there is an open market. When this is the case the prices will fall, new companies will arise and others will exit or emerge (Sutton 1991).

More recent studies have also investigated the effects of competition on a company. A competitive environment has a big effect on the decisions management will make now and in the future (Dhaliwal, Huang, Khurana, Pereira 2014). This is in line with the findings of Xu (2012) who stated that it also influences the choices management makes with regard to the capital structure of the company.

In 1983 Hart developed a model about the product market competition. The results of this study show that more competition results in lower managerial slack. They assume that managers will do everything to meet their given profit target. Within a competitive environment the chance of dropping prices is big. So, for meeting their profit target managers have to focus on cost reduction. Moreover, previous studies also found that competition can help to improve the alignment between the interests of the shareholder and the managers (Giroud, Mueller 2011). The shareholders would like to improve sales and reduce costs, to get a maximal firm value. Managers on the other hand, could chase their own goals. These goals may not always be aligned with the shareholders. This is known as the agency problem (Jensen and Meckling 1976). Competition makes it for the shareholders easier to monitor the actions of managers (Giroud, Mueller 2011). The monitoring mechanism will be more efficient, because in a monopolistic competition a benchmark is available. This benchmark consists of all the other competitors. Therefore, shareholders can make a better judgment about the performance of the managers, because they can compare it with companies in the same industry (Holmstrom 1982). Furthermore, Defond and Park (1999) support these findings. According to them competition improves the usefulness of the performance evaluation of the managers.

Schmidt (1997) found in his study that when an industry faces more competition in a period, the company with the highest cost will be the first company that becomes unprofitable. This could result in liquidation of this company or immediately cutting the costs, because otherwise they will run out of resources. Furthermore, Chhaochharia, Grullon and Grinstein (2009) found that companies who operate in a non-competitive environment are less efficient. Therefore, the idea holds that companies operating in industry with competition are and need to be more efficient. Valta (2015) partly contradicts this view. In his study the results show that the cost of bank debt

(11)

7 is systematically higher for firms that operate in competitive product markets. In other words, managers still have enough resources in periods of decreasing demand.

The investment strategy of a company differs when facing a more competitive environment. When a company is active in a market where competition is absent, management will postpone investments until the net present value is sufficient enough (McDonald, Siegel 1986). Grenadier (2002) finds that management will invest earlier in a project with a net present value close to zero, when it is operating in a competitive market. The company’s investment strategy is affected by their competitors. Furthermore, this increases when the competitors already implemented new investments. The company then will also invest in the same new investment.

2.3 Hypothesis development

The literature review about a competitive environment shows that in previous cases competition effects the shareholders, management decisions, the evaluation of managers and other employees and the efficiency of the entire company. Hence, this research assumes that all these factors will affect the SG&A costs of a company.

As described in the literature review, a lot of empirical research has been done about the behaviour of costs. They found that the costs are sticky, which means that they decrease less with a 1% decrease of sales than they increase with a 1% increase of sales (Anderson et al., 2003; Subramaniam, Weidenmier 2003; Calleja 2006; Dierynck et al. 2012). Firstly, this study investigates if this behaviour of the SG&A costs is still asymmetrical, within the used sample. So, based on existing research, the following hypothesis is tested:

H1: The relative magnitude of an increase in costs for an increase in sales revenue is greater than a decrease in costs for a decrease in sales revenue.

Most of the previous literature on cost stickiness have focused on cost asymmetry factors within a firm (Anderson et al. 2003; Banker et al. 2014). Two more recent papers, however, focus more on external factors. Banker, Byzalov and Chen (2013) investigated the stickiness of cost in highly strict countries regarding the employment protection legislation provision. The other study researched how a fixed-price regulation increases firms’ cost elasticity and decreases cost asymmetry (Holtzhacker, Krishnan, and Mahlendorf 2015).

This research will investigate another aspect of a firms’ external effect, namely the effect of a competitive environment on a firm asymmetric cost behaviour. In other words, how do the firms deal with other rival firms, regarding their cost behaviour. This is in line with previous economic literature, where the effect of a competitive market on a firm’s investment strategies is

(12)

8 examined (Pindyck 1988). Grenadier showed in 2002 that firms will invest faster in an investment with a threshold close to zero NPV, when there is a lot of competition.

In the research of Anderson et al. (2003) they find that firms want to maintain their competitive position, and therefore, regularly spend resources. This means that even in bad economic times, companies need to make investments to keep up with their competitors. Secondly, if management finds out that the sales fall they will do their best to reduce the fall. This result in additional expenditure, to get back at their position. The problem arises that most of the times managers are too optimistic about the demand (Grenadier, 2002).

This view is partly disagreed by Ammann, Oesrch and Schmid (2013) who state that a high level of competition will result in more incentives for managers to keep profits stable in order to survive in the market. In other words, managers have targets they must meet, hence they will reduce unutilized cost into optimal levels. They have more incentives to reduce slack resources.

Wu, Gao and Gu (2015) find that companies within a competitive industry have different strategies from one another. These different strategies all have an impact on the real earnings management of the company. Companies within the same industry, follow different strategies which impact the results of the company

The size of the market could also affect the asymmetric cost behaviour. In 2007 Karuna finds that the size of the market reflects the density of customers. In line with these findings it also has an effect on expected future sales or possible demand for a company’s products. If the chance of increased future sales is higher, managers are less likely to cut costs when sales decline in the current period (Anderson et al., 2003).

If companies face a small profit, they will cut their cost better than companies who have a higher profit (Dierynck et al., 2012). This is in line with the paper of Dhaliwal, et al. (2014). They tested if the product market competition has an effect on the conditional conservatism. In their result they found a difference between a leading company within the industry and a following company. The position in the market has an effect on economic decisions, which also should be the case in the cost behaviour of the company.

According to the theory, it is a bit vague what would be the appropriate reaction of the cost regarding a competitive environment. Some assume the cost stickiness will be higher, because managers are most of the times too optimistic about the future demand. Causing them to hold their resources, because this is necessary for the future. Researchers contradict this view, because managers need to cut their managerial slack, will they be competitive in the future. Still, using the findings of Valta (2015), who stated that companies have enough resources through debt financing.

(13)

9 Therefore, this research use assumes an increase of cost stickiness. This leads to the following hypothesis:

H2: Companies within a competitive environment will have a higher degree of cost stickiness.

Nevertheless, both views can be justified and occur in the same market. It depends on different companies within the industry. Market leaders, assume their demand to be stable and do not need to cut their costs very fast, because they are the biggest supplier. For following companies, it is important to remain profitable, because the chance of liquidation is bigger in a competitive market. They need to cut their costs earlier. This expected result is tested through the following hypothesis.

H3: Leading companies in a competitive environment will have a higher degree of cost stickiness than following companies.

Different industries are likely to have an effect on the asymmetric cost behaviour, because every industry has its own characteristics. The differences could occur in production level, technology development, R&D and regulatory environments. According to the research of Ely (1991), these differences within industries effect the accounting variables of the different companies. Industries have different levels of assets, property, plant and equipment, level of employees or need to have different operating ratios (Ely, 1991; Lazere, 1995).

Bugeja, Lu and Shan (2015) examined the sticky behaviour of costs within different industries. They conducted their research in Australia, using the Australian Securities Exchange (ASX) and looked at six sorts of industries; which are manufacturing, retail, resources, services, construction and others. They found that the specific industry has an effect on the stickiness of the costs. Therefore, this study will also investigate if a competitive environment, within different industries differs from the results in the first hypothesis. The difference between the previous research is that this study focuses on the American markets. I assume the outcomes could differ, because the economic structure, corporate governance of the two countries vary. Calleja et al. (2006) also found differences in cost stickiness between countries. Furthermore, Via and Perego (2014) studied and found that small Italian companies have a different level of cost stickiness than large and public firms. Moreover, on average U.S. firms have higher sales revenue than Australian firms (Bugeja et al., 2015). Therefore, the following hypothesis is formulated and tested:

(14)

10

3 Sample and research

3.1 Sample selection

The quantitative data is required from the databases within the Wharton Research Data Services (WRDS). The data is obtained from COMPUSTAT Annual updates – fundamentals Annual, for the period between 2007 and 2017. This period is chosen, because it gives a good view of the current situation, because only topical data is used. This dataset provides complete financial reports and other financial data of Standard & Poors 1500 index (S&P 1500) firms.

In table 1, the initial sample is stated and after screening of the data some data is dropped. In the first column the reason why this data is dropped can be found, and in the second column the effect on the sample. Firstly, all the missing values regarding SG&A, assets, employees, sales and capital expenditures are deleted. Furthermore, when the SG&A costs are greater than the sales and if the variables had negative values. Secondly, according to the prior research (Kama and Weis 2010; Banker et al. 2011) financial institutions and public utility must be excluded from the dataset, because they react differently than other for-profit organizations. In the initial database, these institutions can be found using the SIC codes 6000 till 6999 and public utility are SIC 4900-4999. The top 1% and bottom 1% of the sales and SG&A costs are deleted, because these extreme observations can heavily influence the data. Lastly, the observations who have a fiscal year of 2006 or 2017 are deleted. These observations are deleted, because the fiscal year depends on the date the financial data is delivered. Some companies use different financial year runs. This heavily influences the competition number, because almost no data is available in 2006 or 2017. As a result, their market share is almost 1. As a result, the sample consist of 39,131 observations.

(15)

11 Table 1. Sample selection

The database COMPUSTAT Annual updates – fundamentals Annual will be used to calculate the cost stickiness, through the SG&A costs. Therefore, the following data is necessary to take out of the database. The first one is “XSGA -- Selling, general and administrative expense” in this field the SG&A cost can be found for a company. Also, the data “REV – Revenue – total and SALE” is going to be used. This item represents the gross income from all the divisions of the company. Thirdly, the item “SALE – Sales / Turnover (Net)”. This item represents the gross sales, the number of actual billings to costumers for regular sales completed during the period, reduced by the different discounts. Finally, the item “Company Name” will be included. This one is necessary for linking the industry with the companies.

For calculating the HHI both a few different items are necessary as some similar for calculating the stickiness of costs. The same items are: “REV – Revenue – total and SALE” and “SALE – Sales / Turnover (Net)”. In addition to the sales, for this subject also the total sales in the industry are required. These can be calculated through looking at the different industry codes, which will be discussed next, and then adding up the different sales number within the industry. This results in the total sales of the specific industry. After that, the HHI can be calculated.

For defining the firm’s industry, I will use the classification of Fama and French (1997). They classified 48 industries in a classification scheme. The firms are assigned to their industry by linking the SIC codes of COMPUSTAT with the classified industries of Fama and French. The classification can be found in appendix 1. The definition of a SIC code is: Standard Industrial Classification (SIC) codes are four-digit numerical codes assigned by the U.S. government to business establishments

Number of observations

Initial sample 132,579

Less: Missing values 69,984

Less: financial institutions (6000 - 6999) 10,058

Less: Public utility (4900 - 4999) 722

Less: Negative values 5,844

Less: Sales < SG&A 4,234

Less: Top 1% of SG&A 417

Less: Top 1% of Sale 182

Less: Bottom 1% of SG&A 418

Less: bottom 1% of Sale 200

Less: year 2006 613

Less: year 2017 776

(16)

12 to identify the primary business of the establishment (Siccode, 2018). Hence, these codes can be used for this study.

3.2 Measurement of Variables 3.2.1 Cost stickiness

For the measurement of cost stickiness, the original model of Anderson et al. (2003) is used. This model uses two main variables, namely the SG&A costs and the net sales revenue.

The following formula will be used (Anderson, et al. 2003):

𝐿𝑜𝑔 𝑆𝐺&𝐴, 𝑆𝐺&𝐴, = 𝛽 + 𝛽 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, + 𝛽 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒_𝐷𝑢𝑚𝑚𝑦, ∗ 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ) + 𝜀,

The part log (𝑆𝐺&𝐴, / 𝑆𝐺&𝐴, ) and log (𝑆𝑎𝑙𝑒𝑠, /𝑆𝑎𝑙𝑒𝑠, ) describes change in cost

and the activity level of company i in year t, relative to year t-1. This number is displayed as a percentage. They used logarithms to create a better normal distribution. For measuring the activity level, the sales are used because the differences in activities within a company is hard to measure (Anderson, et al. 2003). The 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒_𝐷𝑢𝑚𝑚𝑦, represents a dummy variable which is equal to

1 if the sales of company i decrease in year t and 0 when the sales increase in year t. The slope of 𝛽 assumes a sales increase, which is larger than the slope for a decrease (𝛽 + 𝛽 ). Resulting in a 𝛽 smaller than zero. This than results in the cost stickiness phenomenon. Finally, the 𝜀, part

displays the error term.

This study looks at the SG&A costs, although the asymmetrical cost behaviour literature found that more costs show signs of stickiness, SG&A is most widely used. Moreover, in 2011 Banker, Huang and Natarajan find that on average, the SG&A costs to total ratio is 27 percent, compared to the research and development (R&D) to total assets ratio of 3 percent. Also, these costs can be logical linked with the revenue of a company. The main cause of these SG&A costs is the demand in sales (Anderson et al. 2003). Furthermore, professionals look closely at the SG&A spending and try to control these costs, because of their importance (Chen, Lu, Sougiannis 2012). Which makes it a good measurement for measuring the costs stickiness in a firm.

(17)

13 3.2.2 Competitive environment

For the measurement of a competitive environment, the Herfindahl-Hirschman Index (HHI) will be used. This is one of the most common used indicators to detect competitive behaviour in an industry (Matsumoto, Merlone, Szudarovszky 2012). This index is also used by the U.S Department of Justice and Federal Trade Commission. They all use HHI because this index takes in consideration the amount of companies in the industry and their size. HHI is defined as the sum of the squared markets shares of all companies in the market. With this approach, it is possible to combine the amount of companies and size of the industry in the distribution. An index of 1, corresponds to a low competition. In this case the industry will be a monopolized one. Whereas an outcome of 0 means that the competition intensity is high.

To measure the competition of the sample, first the mean is calculated. After that, I divide companies who are active in the same industry into subgroups. These subgroups are sorted by quartiles, according to their market shares These quartiles are used to divine four possible ways of competition. The quartile below 25% demonstrates a high high competition, between 25 and 50% a high competition, between 50% and 100% a low competition (Giroud, Meuller 2011; Li 2010). The model that will be used to measure a competitive market, is the model of Rhoades (1993):

𝐻𝐻𝐼 = (𝑀𝑆𝑖)

Whereas, the part (𝑀𝑆𝑖) measures the market share of the companies using the sales of a specific company and compare that with the entire sales of the industry. The i is about the industry the company is active in and lastly, the j which refers to the specific year.

For the measurement of leading and following companies the HHI index is also used. Companies who have more or equal to 75% of the market share in their industry, are divined as leading. All the companies who have a market share of less than 75% are following companies (Dhaliwal et al. 2014). This is calculated by taking the square root of the HHI outcome. For this measurement a dummy variable is created, which takes a 1 if it is a leading company and 0 if it is a following company.

In all the previous measurements the SIC codes are used for the calculation of a competitive environment. For the fourth hypothesis a different industry classification is used, namely the of Fama and French (1997). They divide all the SIC codes into 48 industries. The HHI is now calculated over this classification, rather than all the possible 378 sic industries.

(18)

14 3.2.3 Control variables

To measure the effect of a competitive environment on cost stickiness, control variables have to be incorporated. This avoid that variables may unfairly influence the effect of a competitive environment on cost stickiness. In this thesis, I am not interested in other effects and therefore with these control variables I reduce other relationships.

The first control variables that are used in this study are asset intensity and employee intensity (Anderson et al. 2003; Bugeja et al. 2015). Both researchers found in their study that these two variables result in a higher degree of cost stickiness. One and the other have this effect, through the adjustment costs. Both are measured using a logarithm, where the asset intensity is measured by dividing the assets with the sales, and the employee intensity by dividing number of employees with the sales.

Secondly, according to Chen, Lu and Sougiannis (2012) another variable that should be controlled for is a decrease of sales for two consecutive periods. Managers will be cynical about the future, when the sales decreases for multiple periods. Therefore, this will influence manager to invest less in capacity of holding less slack in the capacity. The researchers state that a decrease of sales for two consecutive periods, leads to lower cost stickiness. This control variable is measured using a dummy variable, which is 1 if the sales from the current period is below the previous period and the previous is lower than the year before and 0 if this is not the case.

Another control variable is about the agency problem. When a company has a high level of free cash flows (FCF), this allows managers to spend more on SG&A costs when the demand is increasing and suspend SG&A cutting when the demand decreases. Causing, the stickiness to be higher when the FCF is higher (Chen et al. 2012). This variable is measured through the following formula:

The last variable that will be controlled for is the financial crisis. This data used in this study starts in 2007 and lasts until 2017. Within this period also the crisis started and ended. (Campello, Graham, Harvey 2010). This crisis also has its effect on the cost stickiness of the company, because in these periods companies needed to cut their cost better. Otherwise, liquidation could occur. Hence, I will control for this variable.

3.3 Empirical model

In this study four hypothesis are tested. With the use of the different empirical models, the relationship between the cost stickiness and a competitive environment is tested. As stated above,

(19)

15 the original model of Anderson et al. (2003) is used for calculating the first hypothesis Resulting in the first model:

Model 1: 𝐿𝑜𝑔 𝑆𝐺&𝐴, 𝑆𝐺&𝐴, = 𝛽 + 𝛽 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, + 𝛽 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒_𝐷𝑢𝑚𝑚𝑦, ∗ 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ) + 𝛽 𝐴𝑆𝑆𝐸𝑇𝐼𝑁𝑇 + 𝛽 𝐸𝑀𝑃𝐿𝑂𝑌𝐸𝐸𝐼𝑁𝑇 + 𝛽 𝐶𝑂𝑁𝑆𝐸𝐷𝐸𝐶 + 𝛽 𝐹𝐶𝐹 + 𝛽 𝐹𝐼𝑁𝐶𝑅𝐼𝑆 + 𝜀,

The second model is used for calculating H2, where the effect of a competitive

environment on cost stickiness is measured. In the formula the HHI part is changed in HHI. For

the formula see page 17.

Model 2: 𝐿𝑜𝑔 𝑆𝐺&𝐴, 𝑆𝐺&𝐴, = 𝛽 + 𝛽 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, + 𝛽 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒_𝐷𝑢𝑚𝑚𝑦, ∗ 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ) + 𝛽 𝐻𝐻𝐼 + 𝛽 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ∗ 𝐻𝐻𝐼 + 𝛽 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒 , ∗ 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ) ∗ 𝐻𝐻𝐼 + 𝛽 𝐴𝑆𝑆𝐸𝑇𝐼𝑁𝑇 + 𝛽 𝐸𝑀𝑃𝐿𝑂𝑌𝐸𝐸𝐼𝑁𝑇 + 𝛽 𝐶𝑂𝑁𝑆𝐸𝐷𝐸𝐶 + 𝛽 𝐹𝐶𝐹 + 𝛽 𝐹𝐼𝑁𝐶𝑅𝐼𝑆 + 𝜀,

The third model measures the competition between leading and following companies. This is measured by the dummy variable “LORF” Resulting in the following model:

(20)

16 Model 3: 𝐿𝑜𝑔 𝑆𝐺&𝐴, 𝑆𝐺&𝐴, = 𝛽 + 𝛽 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, + 𝛽 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒_𝐷𝑢𝑚𝑚𝑦, ∗ 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ) + 𝛽 𝐿𝑂𝑅𝐹 + 𝛽 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ∗ 𝐿𝑂𝑅𝐹 + 𝛽 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒 , ∗ 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ) ∗ 𝐿𝑂𝑅𝐹 + 𝛽 𝐴𝑆𝑆𝐸𝑇𝐼𝑁𝑇 + 𝛽 𝐸𝑀𝑃𝐿𝑂𝑌𝐸𝐸𝐼𝑁𝑇 + 𝛽 𝐶𝑂𝑁𝑆𝐸𝐷𝐸𝐶 + 𝛽 𝐹𝐶𝐹 + 𝛽 𝐹𝐼𝑁𝐶𝑅𝐼𝑆 + 𝜀,

For testing H4, this study looks at the effect different industries could have on the stickiness of costs. The new variable is called “HIIFFI”, and can have 48 possible outcomes. This model is tested without the use of control variables. This has been done because, the outcomes should indicate if there is a difference between different industries. The determinants are in this case not important yet. It is expected that through these different choices companies in a industry need to make, result to a difference in cost stickiness. Therefore, it is measured without control variables. This new parts lead to empirical model 4:

Model 4: 𝐿𝑜𝑔 𝑆𝐺&𝐴, 𝑆𝐺&𝐴, = 𝛽 + 𝛽 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, + 𝛽 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒_𝐷𝑢𝑚𝑚𝑦, ∗ 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ) + 𝛽 𝐻𝐻𝐼𝐹𝐹𝐼 + 𝛽 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ∗ 𝐻𝐻𝐼𝐹𝐹𝐼 + 𝛽 𝐷𝑒𝑐𝑟𝑒𝑎𝑠𝑒 , ∗ 𝐿𝑜𝑔 𝑆𝑎𝑙𝑒𝑠, 𝑆𝑎𝑙𝑒𝑠, ) ∗ 𝐻𝐻𝐼𝐹𝐹𝐼

With the use of the four different models, this thesis tries to measure and explain the effect of competition on the market as a whole, within industries and if it differs from leading and following companies. The entire study focuses on one country, namely the United States of America.

(21)

17

4 Results

4.1 Descriptive statistics

In table 2 all the descriptive statistics of the used variables in my analysis can be found. As said before the sample consists of 39,131 fiscal firm year observations. These firm year observations are generated by 6,614 specific firms. As a result, each firm has an average of 5.9 financial observations within 2007 and 2017.

Table 2. Descriptive statistics

Variables Mean Median St. dev Minimum p25 p75 Maximum

Sales ($mill) 2671.316 413.542 6854.374 2.832 90.401 1850.623 76534

SG&A ($mill) 289.912 80.042 1340.66 1.052 20.23 309.363 14726.43

Log (Sales) 6.042 6.027 2.051 1.343 4.515 7.524 11.246

Log (SG&A) 4.484 4.394 1.855 0.719 3.055 5.738 8.914

Log (Sales i,t/Sales i,t-1) 0.054 0.048 0.292 -3.672 -0.056 0.159 3.895

Log ( SG&A i,t/SG&A i,t-1) 0.060 0.046 0.254 -3.309 -0.046 0.148 3.400

HHI (sic) 0.248 0.196 0.206 0.031 0.091 0.329 1

HHI (ffi) 0.062 0.044 0.060 0.011 0.026 0.071 0.546

Leading or following (dummy) 0.082 0 0.275 0 0 0 1

Asset intensity 1.665 1.134 2.141 0.016 0.710 1.850 88.405

Employee intensity 0.0055 0.0036 0.0092 0.000001 0.0021 0.0059 0.3707

Free cash flows -0.048 0.081 1.289 -115.107 -0.070 0.211 12.610

Successive Decrease (dummy) 0.122 0 0.327 0 0 0 1

Note: The table provides summary statistics for the main variables for firm-year observations from fiscal years 2007 - 2016. This panel reports summary statistics for the main variables for the initial sample (n = 39,131). Only the Log of Sales and SG&A use a different sample due to missing t-1 values (n = 32,800).

(22)

18 On average, firms in this sample have an annual sale of $ 2,671 million (a median of $ 414 million) and an annual SG&A cost of $ 290 million (a median of $ 80 million). As a percentage, the mean SG&A costs are 11% of the mean sales. Looking at the median numbers, this percentage is 19%. The mean and median from these two variables differ a lot, suggesting that some firms have a very high annual sale or SG&A cost, influencing the mean upwards. Hence, a logarithm is taken of these two variables to see if the variables then are normally distributed. Subsequently, the mean and median are at both variables nearly identical and are normally distributed.

For calculating a competitive environment, two variables are created. The first one is the HHI(sic) and the second one is the HHI(ffi). Both values differ a lot, because of the quantity of firms active in the industry. The sic method consist of more than 360 industries, while the ffi only has 48. The mean of HHI(sic) is 0.25 and the mean of HHI(ffi) is 0.062. Indicating that the industries within HHI(ffi) are more competitive. More important are p25 and p75, because these values are used for calculating high competition and low competition. At the HHI(sic) method observation firms with a value lower than 0.090 are active in a high competitive market and firms with a value higher than 0.329 are active in a low competitive market. For HHI(ffi) the values are respectively 0.011 and 0.076. So, the height of the competition is calculated with the specific market numbers. Furthermore, on average 8% of the observations are leading companies. This can be seen by the variable leading or following, which has a mean of 0.082. Hence, the median is zero, because 92% of the observations are following companies and therefor this value occurs most often.

The last 4 rows of the table show the descriptive statistics of the control variables. Asset intensity has a mean of $ 1.67 million (a median of $ 1.13), meaning that on average $ 1.67 million of assets are needed to generate one million of sales. Employee intensity has a mean of 0.0055. So, for a million dollars of annual sales, on average, 55 employees are necessary. Thirdly, the control variable free cash flow, indicate how much the free cash flow is compared to the current assets. In this sample that is -0.048. In other words, the free cash flow accounts for minus 4.8% of the total current assets. Lastly, the control variable successive decrease has a median of 0, meaning that less firms have experienced a sales decrease for two successive years than firms that did. On average, only 12% experienced a sales decrease for two years (mean of 0.122).

(23)

19 Table 3. Pearson and Spearman correlation matrix

Log(Sales

i,t/Sales i,t-1)

Log(SG&A i,t/SG&A

i,t-1)

Sales

decrease HHI (sic) HHI (ffi) intensity Asset Employee intensity FCF Leading or following Successive decrease

Log(Salesi,t/Sales i,t-1) 1 0.6424*** -0.8338*** -0.0740*** -0.0551*** 0.0347*** -0.0440*** 0.0424*** -0.0310*** -0.4635***

Log(SG&A i,t/SG&A i,t-1) 0.6007*** 1 -0.5244*** -0.0698*** -0.0556*** 0.0664*** -0.0201*** 0.0196*** -0.0246*** -0.3457***

Sales decrease -0.6183*** -0.3946*** 1 0.0308*** 0.0457*** 0.0097* 0.0154*** -0.0998*** 0.0230*** 0.5450*** HHI (sic) -0.0480*** -0.0430*** 0.0343*** 1 0.3709*** -0.2915*** 0.0854*** 0.0706*** 0.4817*** 0.0232*** HHI (ffi) -0.0302*** -0.0397*** 0.0337*** 0.2915*** 1 -0.0136** 0.0163*** -0.0034 0.1024*** 0.0235*** Asset intensity -0.0309*** 0.0313*** 0.0118** -0.1634*** -0.0235*** 1 -0.1190*** -0.0470*** -0.1123*** -0.0023 Employee intensity 0.0031 0.0219*** -0.0097* -0.0165*** -0.0145*** -0.0126** 1 0.0313*** 0.0173*** 0.0332*** FCF -0.0485*** -0.0372*** 0.0098* 0.0644*** 0.0153*** -0.2311*** 0.0217*** 1 0.0018 -0.0727*** Leading or following -0.0228*** -0.0161*** 0.0256*** 0.7573*** 0.1635*** -0.0690*** -0.0167*** 0.0230*** 1 0.0246*** Successive decrease -0.3460*** -0.2650*** 0.5620*** 0.0280*** 0.0173*** 0.0097* 0.0073 0.0061 0.0254*** 1

Notes: This table provides pair-wise correlations for firm-year observations from fiscal years 2007-2017 for the main variables for the initial sample (n=39,131). It shows the correlations for companies in a competitive environment using two methods. Top right shows Spearman and bottom left Pearson Correlations. The fixed

effects of years are excluded from the correlation matrix. *** indicate significance at the 1% level

** indicate significance at the 5% level * indicate significance at the 10% level

(24)

20

4.2 Correlation test

To investigate if my main variables are correlated with each other, I conducted two correlation tests. The tests show if there is a relationship between two variables. If the outcome is between 0 and 1, it indicates a positive relationship between does two variables. A value between -1 and 0 indicates a negative relationship between does two variables.

Furthermore, for this study also a test is computed for multicollinearity. No problems are found within this test. According to Belsey, Kuh and Welsch (2005) the correlation should not above 0.7, otherwise it could complicate the regression analysis because of the direct relationship between the two variables. For the formal test of multicollinearity, the VIF values should not be above 10.

Table 3 shows the result of the correlation tests. A lot of correlations between the variables are significant. One value has an outcome which is higher than 0.7, namely “HHI(sic) with leading and following”. This could harm the regression of the third hypothesis, because the two variables are highly correlated. On the other, it does make sense because the variable “leading or following” is computed by taking the square root of “HHI (sic). It would be stranger if these two variables did not correlate. Except from this correlation, no other correlation shows a relationship that could harm the regression analysis.

The logarithm of the change in sales and SG&A shows a positive significant relation between each other (0.6007), which is as expected by the first hypothesis. This outcome means that they move in the same direction, however they still have a different pattern. Furthermore, the correlation between the logarithm of the change in SG&A and the HHI index is 0.0430 and -0.0397 and both significant. It looks like the competition within the market indeed influences the SG&A costs. Moreover, asset and employee intensity have a low positive correlation with the dependent variable SG&A cost. In contrast, free cash flow and successive decrease have a low negative correlation. Meaning that asset and employee intensity move in the same direction and free cash flow and successive decrease in the opposite direction.

4.3 Multivariate analysis 4.3.1 Cost stickiness

First, hypothesis 1 is tested. This hypothesis is created to see if the asymmetrical cost behaviour is also present in this sample. In table 4 the different coefficients with regard to the formula of Anderson et al. (2003) can be found. No control variables are added, because for this hypothesis

(25)

21 this study only looked if cost stickiness is present in this sample. In this hypothesis, in contrast to the others, the causes and effect linkages between the variables are not important yet.

The first β is significant and has a value of 0.60. This value indicates that if the sales increases with 1% the SG&A costs increase with circa 0.60%. The β2 value looks at the height of the SG&A costs when there is a period of sale decrease. The coefficient is significant and has a value of -0.184, which means that if the sales decrease with 1% the SG&A costs only decreased with approximately 0.42%. This is calculated through adding β1 and β2 with each other. These findings are in line with the findings of Anderson et al. (2003) who found that the costs increased with 0.55% and only decreased with 0.35%. The findings of this study are almost identical to the outcomes of that study. The cost increase percentage is a bit a higher, but so is the height of the cost at a sales decrease.

Table 4. Results regression analysis model 1 Dependent variable

∆Log(SG&A i,t/SG&A i,t-1)

Explanatory variables: CF

β0 Intercept 0.015***

(10.39)

β1 Log(Salesi,t/Sales i,t-1) 0.6***

(106.62) β2 Sales decrease * Log(Sales i,t/Sales i,t-1) H1 (-) -0.184*** (18.39) Year NO N 32,800 Adjusted R-squared 0.3673 Prob > X-squared 0.000 F-statistic 9520.57

Note: This table provides the results of the OLS regression using Log(SG&A i,t/SG&A i,t-1) as the dependent variable for firm-year observations from fiscal years 2007-2016. Significance level is two-tailed. *** indicate significance at the 1% level

** indicate significance at the 5% level * indicate significance at the 10% level

Both the betas have a significance level of <1%, therefore strong evidence has been found that the relative magnitude of an increase in costs for an increase in sales revenue is greater than a decrease in costs for a decrease in sales revenue, is also present in this sample. The adjusted

(26)

r-22 squared has a value of 0.3749. So, the changes in SG&A costs are for 37% explained by this model. Lastly, the F-statistic (2, 32797) is 9520.57, below α and significant, meaning the model itself is sufficient applied. According to these findings, strong support has been found for the first hypothesis.

4.3.2 Competitive environment effect on cost stickiness

The second hypothesis predicts that the asymmetrical costs behaviour, here the SG&A costs, is positively associated with a competitive environment. In other words, firms who are active in a competitive environment face more cost stickiness. Therefore, the coefficient relating to the fifth beta should be negative and significant. The results of the regression with model 2 can be found in table 5.

The left column shows the results of the regression without taking the control variables in to account. So, here the results for β5 could have been influenced by other variables. Therefore, in the right column the control variables are added in the model. The most important coefficient of the variables in this model is β5, which indicates the three-way interaction of the variables (Sales decrease * Log (Sales i,t/Sales i,t-1) * HHI(sic)). In other words, it is the percentage change in SG&A costs within a period of sales decrease for a company who is active in the most competitive environment relative to a company who is active in a non-competitive environment.

The coefficient of interest has a value of -0.098 The negative value is in line with the hypothesis. It shows that in a decreasing period of sales, a company who faces high competition cuts their SG&A costs by 0.098 percentage point less than a company who faces low competition. Using the mean descriptives of the Sales ($2,761 million) and the SG&A costs ($290 million) this means that companies active in a competitive environment facing a sales decrease of $276.1 million (1% of $2,761 million) cuts their cost by $284,200 (0.098% of $290 million) less than a firm active in a low competitive environment. The difference is statistically significant at a 1% level.

All the control variables in the model are significant. The coefficient of employee intensity is positive, meaning that employee intensive companies have a lower degree of cost stickiness. This is not in line with previous research, who found that these companies will have a higher degree of cost stickiness (Anderson et al. 2003; Bugeja et al. 2015). A possible explanation is that this sample consists of less employee intensive firms. The mean in this sample is only 0.0055, which is lower than at the other studies. The asset intensity is in line with previous studies. The coefficient is -0.068 and significant, meaning that asset intensive companies have a higher degree of cost stickiness. Two successive years of decrease leads to a higher degree of cost stickiness, which is not as expected. A possible explanation could be that after the financial crisis, mangers even in

(27)

23 successive years of sales decrease expect the sales to increase in the future because the economy is getting better.

Table 5.Results regression analysis model 2 Dependent variable

∆Log(SG&A i,t/SG&A i,t-1)

Explanatory variables: CF CF

β0 Intercept 0.014*** 0.015*

(8.77) (1.77)

β1 Log(Salesi,t/Sales i,t-1) 0.617*** 0.600***

(87.76) (83.50) β2 Sales decrease * Log(Sales i,t/Sales i,t-1) -0.147*** -0.16*** (11.99) (12.84)

β3 HHI(sic) 0.004 -0.001

(1.25) (0.18) β4 Log(Sales i,t/Sales i,t-1) * HHI(sic) -0.049*** -0.045***

(4.15) (3.76) β5 Sales decrease * Log(Sales i,t/Sales i,t-1) *

HHI(sic) H2(-) -0.106*** -0.098*** (4.96) (4.56) β6 Employee Intensity 0.028*** (4.94) β7 Asset intensity -0.068*** (10.56)

β8 Free cash flow 0.003*

(2.77) β9 Successive decrease -0.050*** (14.58) Year NO YES N 32,800 32,800 Adjusted R-squared 0.3706 0.3874 Prob > X-squared 0.000 0.000 F-statistic 3861.7 2208.75

Note: This table provides the results of the OLS regression using Log(SG&A i,t/SG&A i,t-1) as the dependent variable for firm-year observations from fiscal years 2007-2016. Significance level is two-tailed. *** indicate significance at the 1% level

** indicate significance at the 5% level * indicate significance at the 10% level

(28)

24 Concluding, the coefficient of interest has a significance level of <1%, therefore strong evidence has been found a competitive environment has a positive effect on cost stickiness. The adjusted r-squared has a value of 0.3874. So, the changes in SG&A costs are for 39% explained by this model. Lastly, the F-statistic is significant, meaning the model itself is sufficient applied. According to these findings, strong support has been found for the second hypothesis.

4.3.3 Leading or following company effect on cost stickiness

The third hypothesis predicts that the asymmetrical costs behaviour, here the SG&A costs, is positively associated with a leading company in comparison to a following company. Leading companies will have a higher degree of cost stickiness than following companies. The results of the regression with model 3 can be found in table 6.

The left column of table 6 shows the results of the regression without taking the control variables in to account. So, here the results for β5 could have been influenced by other variables. Therefore, in the right column the control variables are added in the model. The most important coefficient of the variables in this model is β5, which indicates the three-way interaction of the variables (Sales decrease * Log (Sales i,t/Sales i,t-1) *LorF). In other words, it is the percentage change in SG&A costs within a period of sales decrease for a leading company higher than for a following company.

The coefficient of interest has a value of 0.005 The positive coefficient is not in line with the formulated hypothesis. It shows that in a decreasing period of sales, a leading company cuts their SG&A costs by 0.005 percentage point more than a following company. Using the mean descriptives of the Sales ($2,761 million) and the SG&A costs ($290 million) this means that leading companies facing a sales decrease of $276.1 million (1% of $2,761 million) cuts their cost by $14,500 (0.005% of $290 million) more than a following company does. The problem arises that the founded coefficient is not significant. The p-value is 0.14 and that is even below a p-value of 10%.

All the control variables in the model are significant. The control values have the exact same direction as in model 2 and are almost the same. Therefore, the same explanation holds as described in section 4.3.2.

Concluding, the coefficient of interest has a significance level of >10%, therefore no evidence has been found that a leading company will have a higher degree of cost stickiness than a following company. The adjusted r-squared has a value of 0.3779. So, the changes in SG&A costs are for 38% explained by this model. Lastly, the F-statistic is significant, meaning the model

(29)

25 itself is sufficient applied. So, according to these results of the regression analysis, no support has been found for the second hypothesis.

Table 6. Results regression analysis model 3 Dependent variable

∆Log(SG&A i,t/SG&A i,t-1)

Explanatory variables: CF CF

β0 Intercept 0.015*** 0.015***

(9.92) (9.92)

β1 Log(Salesi,t/Sales i,t-1) 0.6*** 0.580***

(102.60) (95.65) β2 Sales decrease * Log(Sales i,t/Sales i,t-1) -0.185*** -0.191*** (17.81) (17.61)

β3 LorF 0.000 0.003

(0.06) (0.68) β4 Log(Sales i,t/Sales i,t-1) * LorF 0.006 0.001 (0.26) (0.04) β4 Sales decrease * Log(Sales i,t/Sales i,t-1) * LorF H3(-) 0.016 0.005 (0.42) (0.14)

β5 Employee Intensity 0.046***

(7.62)

β6 Asset intensity -0.078***

(12.65)

β7 Free cash flow 0.004***

(2.90) β8 Successive decrease -0.052*** (15.07) Year NO YES N 32,800 32,800 Adjusted R-squared 0.3673 0.3779 Prob > X-squared 0.000 0.000 F-statistic 3808.17 1106.11

Note: This table provides the results of the OLS regression using Log(SG&A i,t/SG&A i,t-1) as the dependent variable for firm-year observations from fiscal years 2007-2016. Significance level is two-tailed. *** indicate significance at the 1% level

** indicate significance at the 5% level * indicate significance at the 10% level

(30)

26 4.3.4 Industry effect on cost stickiness

The fourth and final hypothesis predicts that the asymmetrical costs behaviour, here the SG&A costs, is affected by the industry companies are active in. The results of the regression with model 4 can be found in table 7. Eventually, 42 industries are investigated instead of the 48 described earlier. This happened because the financial institutions and utilities were deleted; see section 3.1 for the explanation.

For this hypothesis I am interested in the differences between β1 and β2 for a specific industry. If there is a difference between those values within different industries, this could mean that some industries face a higher degree of cost stickiness than others. The industry codes are here mentioned by the Fama and French code classification. The meaning of a specific industry code can be found in appendix 1. Also, for this specific regression is chosen to not include the control variables. This choice has been made, because the industry is the cause of all the other effects, here the control variables. An industry has different characteristics which are normal for that environment (Ely, 1991; Lazere, 1995). In some industries it is necessary to have a high amount of cash-flows or assets. This study is therefore interested in the cause of cost stickiness through the different industries. Does the industry a company is active in really influence the stickiness.

The different reactions of the SG&A cost to a 1% increase and decrease per industry can be found by looking at β1 and β2. The overall stickiness in this sample is 0.6% by -0.184%, see section 4.3.1 The results of this regression indeed show big differences between industries. For example, industry 10 (apparel) has the highest cost increase with a 1% sales increase, namely 1,06% and the cost only decrease with 0.509% (β1 - β2). Which shows that this industry is very sticky. The same happens at industry 8 (printing and publishing) were the cost increase with 0.984% at a 1% increase but only decrease with 0.517%. Lastly, also industry 35 (electronic equipment) has a high degree of cost stickiness. The difference between an increase and decrease of 1% sales with the SG&A cost is -0.480%.

In contrast, industry 41 (wholesale) has a low degree of cost stickiness. The difference between β1 and β2, the degree of cost stickiness, is here only -0.083%. In other words, the cost increase and decrease almost the same with an increase and decrease of the sales. To a lesser extent, industry 22 (electrical equipment) also shows a lower degree of cost stickiness compared to the other industries.

(31)

27

Table 7. Results regression analysis model 4 Dependent

variable ∆Log(SG&A i,t/SG&A i,t-1)

Industry code β0 β1 β2 Year N R-squared Prob >X-squared F-statistic 1 0.056 0.529*** -0.341 NO 158 0.1057 9.16 0.0002 (1.44) (3.77) (1.30) 2 0.011 0.687*** -0.291*** NO 747 0.3807 230.33 0.0000 (1.29) (18.51) (4.08) 3 0.002 0.865*** -0.253* NO 165 0.6174 133.35 0.0000 (0.15) (7.19) (1.71) 4 0.004 0.896*** -0.197** NO 199 0.6799 211.31 0.0000 (0.45) (13.42) (1.96) 5 0.013 0.758*** 0.237 NO 48 0.7797 84.15 0.0000 (0.80) (7.63) (1.56) 6 0.005 0.449*** 0.076 NO 239 0.448 96.35 0,0000 (0.45) (8.98) (0.84) 7 0.018* 0.573*** -0.091 NO 524 0.3666 152.38 0.0000 (1.84) (14.65) (1.00) 8 -0.012** 0.984*** -0.467*** NO 208 0.7704 348.28 0.0000 (2.04) (20.76) (6.36) 9 0.006 0.777*** -0.271*** NO 484 0.6136 384.53 0.0000 (0.87) (22.30) (4.30) 10 0.007 1.060*** - 0.551*** NO 494 0.5112 258.78 0.0000 (0.75) (17.45) (5.75) 11 -0.011 0.802*** -0.363*** NO 589 0.5550 367.73 0.0000 (1.44) (23.51) (5.44) 12 0.014* 0.689*** -0.048 NO 1,065 0.5091 552.81 0.0000 (2.59) (22.35) (0.91) 13 0.019 0.670*** -0.183** NO 1,209 0.4025 407.84 0.0000 (2.48) (22.96) (3.05) 14 0.011 0.639*** -0.331*** NO 839 0.2253 121.57 0.0000 (0.93) (12.90) (3.82) 15 0.012 0.638*** -0.116 NO 268 0.4258 100.00 0.0000 (1.21) (9.30) (1.00) 16 0.014 0.568*** 0.122 NO 96 0.5719 64.46 0.0000 (0.86) (8.59) (0.77) 17 -0.007 0.772*** -0.277*** NO 838 0.4029 283.37 0.0000 (0.85) (15.05) (3.76) 18 0.041*** 0.490*** 0.060 NO 339 0.4947 166.64 0.0000 (3.52) (13.93) (0.72) 19 0.008 0.589*** -0.253*** NO 521 0.3750 157.00 0.0000 (0.57) (14.85) (3.30)

(32)

28 20 -0.007 0.911*** -0.455** NO 101 0.4541 42.59 0.000 (0.24) (6.54) (2.05) 21 0.026*** 0.504*** 0.011 NO 1,261 0.5122 662.49 0.0000 (4.65) (22.24) (0.30) 22 0.023** 0.538*** -0.134* NO 637 0.3234 153.01 0.0000 (2.04) (12.46) (1.85) 23 0.021** 0.499*** -0.049 NO 619 0.3745 186.01 0.0000 (2.45) (11.51) (0.71) 24 0.021** 0.476*** -0.122 NO 219 0.3181 51.85 0.0000 (1.97) (8.07) (1.02) 25 0.072** 0.099 0.230 NO 82 0.3382 21.69 0.0000 (2.90) (1.15) (2.41) 26 0.043 0.729 0.091 NO 64 0.0588 2.97 0.0000 (0.55) (1.57) (0.09) 27 0.024 0292*** 0.089 NO 555 0.1199 38.73 0.0000 (1.13) (5.81) (0.88) 28 0.011 0.176** 0.157 NO 436 0.0764 18.99 0.0000 (0.46) (2.71) (1.40) 29 0.041 0.399*** 0.059 NO 153 0.2116 21.40 0.0000 (1.24) (3.94) (0.32) 30 0.026*** 0.444*** -0.144*** NO 2,701 0.2686 496.77 0.0000 (2.90) (22.12) (3.93) 32 -0.015** 0.806*** -0.219*** NO 1,309 0.4355 505.48 0.0000 (2.24) (26.15) (3.53) 33 -0.008 0.940*** -0.233*** NO 475 0.7349 657.84 0.0000 (1.10) (28.05) (3.75) 34 0.007** 0.769*** -0.288*** NO 4,764 0.5262 2595.77 0.0000 (2.17) (59.00) (11.42) 35 -0.010 0.799*** -0.480*** NO 1,216 0.4764 533.64 0.0000 (1.45) (27.95) (8.81) 36 0.013*** 0.608*** -0.222*** NO 2,629 0.4126 924.04 0.0000 (2.91) (32.04) (6.87) 37 0.012** 0.517*** -0.207*** NO 732 0.3687 214.45 0.0000 (1.76) (15.11) (3.80) 38 0.017** 0.758*** -0.036 NO 482 0.6119 380.18 0.0000 (2.37) (18.21) (0.54) 39 0.009 0.816*** -0.147 NO 108 0.5745 73.23 0.0000 (0.75) (10.11) (0.65) 40 0.015** 0.717 -0.364 NO 1,110 0.3665 321.86 0.0000 (1.75) (20.35) (6.52) 41 0.024*** 0.629*** -0.083** NO 1.272 0.4908 613.56 0.0000 (4.23) (25.50) (1.92) 42 0.007** 0.889*** -0.236*** NO 1,963 0.5769 1338.54 0.0000 (1.95) (43.69) (5.27)

(33)

29 43 0.001 0.678*** -0.261** NO 680 0.2675 125.00 0.0000

(0.11) (13.84) (2.48)

48 -0.024 0.799*** -0.328*** NO 292 0.5830 204.44 0.0000

(1.40) (16.07) (4.14)

Note: This table provides the results of the OLS regression using Log(SG&A i,t/SG&A i,t-1) as the dependent variable for firm-year observations from fiscal years 2007-2016. β0 is the intercept; β1 is Log(Sales i,t/Sales i,t-1) and β2 stands for Sales decrease * Log(Sales i,t/Sales i,t-1). The value in parentheses are the appropriate t-values. See appendix 1 for the meaning of the industry codes. Significance level is two-tailed. *** indicate significance at the 1% level

** indicate significance at the 5% level * indicate significance at the 10% level

Concluding, all the coefficients discussed above have a significance level of <1%, and therefore strong support has been found that the industry indeed effects the cost stickiness of a company. Beside the coefficients discussed a lot of other industries also differ a lot from each other and had a significant which was lower than the 10% level chosen. This gives even more evidence for this hypothesis. The adjusted r-squared of the models differs from one another, but all are bigger than 0., which indicates that the change in change in SG&A can indeed be

explained by the change in sales. Lastly, the F-statistic is in all occasions significant, meaning the models are all sufficient applied. So, according to these results of the regression analysis, strong support has been found for the fourth hypothesis.

4.4 Robustness check

To secure the validity of the results, also some extra robustness tests are executed. Three sensitivity analysis are performed. First, another measurement of a competitive environment is used to see if the results are still the same. Secondly, the hypothesis with regard to a leading or a following company is retested and finally this study looks at the effect of the financial crisis on the results. Even though, the fixed effect of the year is already looked at, the years within and after the crisis could give some extra explanation.

The first robustness check is to test whether the results found in the second hypothesis are consistent when the measurement of a competitive environment changes. Previously, for measuring the HHI the SIC codes where used. For this test the 48-industry code of Fama and French is used. This is in line with other studies who also used this method to measure the competition within a market (Irvine, Pontiff 2009). A new dummy variable is created, which gets a 1 if the HHI(ffi) value is higher than the upper quartile and gets a 0 if it is below the highest quartile. Model 2 is also used here, but the dummy variable is replaced by the new one. So, also

Referenties

GERELATEERDE DOCUMENTEN

Even though economic literature provides a broad perspective on the impact of board size on performance and value of different types of organizations, it is still unclear which result

The present study compared go-signal response times (GSRT) and stop-signal response times (SSRT) of behavioral inhibition, which is estimated by the stop signal task,

Mean- while, nodes keep track of coordinates that are either (1) included in messages carrying sensor data, or (2) are explicitly transmitted. Using the received coordinate

The strength of the relationship between human capital and value creation, as well as the mediating effects of organizational and social capital are expected to

Common-law employment contract; labour legislation; good faith; fairness; implied duty of trust and confidence; implied duty of fair dealing; constitutional development of the

A contribution to the literature can be made to describe which business model is most suited for a network company, how a network company should design their marketing and

The business phenomenon in this research is that the networks of management accountants are likely to differ between a management accountant operating in a bean

The implementation failure of the cost-to-serve method (excellerate) is caused by as well “technical” as “organizational &amp; behavioral” factors. The technical factors for