Performance-based executive compensation and its influence on firm volatility in the non-financial sector

(1)

1

University of Amsterdam Name: Joris Boelens

Bsc Economics and Business Student number: 10002131

Specialization: Finance and Organisation Supervisor: Andro Rilović Field: Organizational Economics Submission date: 29 June 2015

Performance-based executive

compensation and its influence on firm

volatility in the non-financial sector

(2)

2

Abstract

In this thesis the effect of performance-based executive compensation on firm volatility is examined. Studies on, particularly, financial firms show divergent results of the effects of performance-based compensation. While most of the earlier studies focused on the financial sector, this thesis focuses on the non-financial sector. Firm volatility in this study is defined in two different ways, the yearly standard deviation of the share prices and the standard deviation of the profits of the firm. The results of this research show that the relative size of performance-based compensation has a small positive effect on the both measures of volatility. This is concluded after an empirical study for the period from 2006 until 2014 for all American S&P 1500 firms. However the results also show that it is only a very small explanation of the volatilities. So the performance-based compensation is only a partly explanation of the firms volatilities.

Keywords: Executive compensation, firm performance volatility, stock prices, profits

Statement of Originality

This document is written by student Joris Boelens 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)

3

1. Introduction

Since the start of the recent financial crisis, the system of performance-based compensation in the banking industry has been strongly criticized in the political circles. Politicians are asking for legal restrictions and also the public opinion became more and more critical about this system of executive compensation. Although the discussion mainly focusses on the total size of the salary of the executives of the banks there are also questions about the performance-based part specifically. And what the influence of this would be on the banks.

Because the generally accepted assumption after several studies is that performance-based compensation induces risk-taking, opposition against this kind of remuneration increased. For the banking industry this has been studied by Chen, Steiner and Whyte (2005) for the period before the financial crisis. They concluded that the structure of executive compensation induces risk-taking behaviour because of the stock option-based wealth of the executives. And in general most other studies about this relationship (Chen & Ma, 2011 ; Coughlan & Schmidt, 1985) have similar conclusions.

Opponents of the performance-based executive compensation say that this kind of remuneration creates short-term incentives for the executives (Efing, Hau, Kampkötter & Steinbrecher, 2014). They blame this incentive misalignment as one of the main factors causing the global financial crisis of 2008.

However a study of Gregg, Jewell and Tonks (2012), researching the period of the financial crisis of 2007-2008, concludes that it is unlikely that these types of incentives structures could be held responsible for a focus on short-term results. Also another study from that same period tells us that higher bonus payments are not related to higher risk-taking (Gehrig, Lütje and Menkhoff, 2009).

So the main question to answer in this thesis is if rewarding that is linked to performance-based measures does indeed create wrong incentives for the executives of the firms, which in turn causes an increase in firm volatility. As described above, there have already been several studies for the financial sector to examine this. But this way of remuneration is also frequently used in other sectors. The performance-based pay in these sectors faces almost no opposition from the public opinion or the politicians. In principle, the same type of remuneration structure should create the same type of incentives for the executives of these

(4)

4 firms. That’s why I would like to research the influence of the remuneration policy on the volatility of firms in the non-financial sector.

Firm volatility in this study is defined in two different ways. The first measure is the yearly standard deviation of the share prices of the firm. As a second way of defining the firm volatility, the standard deviation of the profits is used. The effect on both variables will be studied in this thesis.

I will do this for all US S&P 1500 firms that are not in a financial sector. The period I will use in this thesis is from 2006 until 2014. Choosing this period, which coincidences by and large financial crisis, has two reasons. First of all because I wanted to have the most recent data that tell something about how firms and executives nowadays operate in economic times of instability. But also the practical reason, that the way of reporting the executive compensations changed in 2006 to the new standards of reporting according to the SEC-fillings. This makes it not possible to compare data from before 2006 with the data afterwards. Choosing this period also brings some adverse effects with it, which will be discussed later on in this thesis in the discussion part.

1.1. Research Question

Taking the above considerations together brings me to the following research question that will be discussed in this thesis:

“What is the influence of the relative size of the performance-based compensation of CEO’s on the volatility of firms performance in the non-financial sector?”

1.2. Hypothesis

As said before in the current literature there are studies that say different things about the relationship of the research question. After reading all the different above mentioned studies with its conclusions pro and against the effect of the compensation structure on the volatility of firms brings me to the expectation that there will probably be a (small) positive effect on the volatility.

H0: The relative size of the performance-based compensation has no effect on the volatility

of firms in the non-financial sector.

H1: The relative size of the performance-based compensation has positive effect on the

(5)

5

2. Data and methodology

The independent variable in our regression will be the executive compensation. To define the firm volatility as the dependent variable we have two variables which will be studied. The standard deviation of the share prices and the standard deviation of profits of the firms. In the following three paragraphs all three variables will be explained in more detail. In the fourth paragraph the methodology used for the data analysis in Stata will be explained.

2.1 Executive Compensation

The data on the executive compensation I will collect from the ‘Compustat ExecuComp’ database. This database covers data on the compensation of the executives of all US companies in the S&P 500, S&P MidCap 400 and S&P SmallCap 600 indexes. Another name for these group of companies is the S&P 1500, this term I will also use from now on in this thesis. This database covers data from 1992 until 2014 of in total 3410 companies.

But I will not use all data from this database, because these raw data must be cleaned up first. For the period of time on which the data are available I will only use the data from 2006 until 2014, so a period of 9 years. This, as mentioned in the introduction, because the measurement standards for compensation changed in 2006 which makes it impossible to compare data from before 2006 with the data in the period of 2006 until 2014. This leaves data on 2.341 companies with in total 90.754 observations. But still not all of these data are useful.

Next part to clean up for, is the number of executives listed in the database per firm. This database covers data of the top 5 executives of all firms. For this thesis I will only use the data of the CEO’s of every firm. This for the case of simplicity but also because not all firms have an equal amount of executives in the database covered so that would create a skewed image of reality. By dropping all non-CEO executives from the firms (using the ‘pceo’ variable) all non-CEO executives are dropped. This will drop 76.610 observations.

I will also drop the data on the firms that are in the financial sector. As described in the research question I am only interested in the companies in the non-financial sectors. The variables ‘spindex’ (Industry Group) and ‘inddesc’ (Industry Group Description) give the information about which firms are in which sector. In the appendix (7.1) you will find an overview of all sectors from which the firms are dropped. This will drop 423 financial firms, which are 2.634 observations. Checking the data revealed that one firm had a missing values

(6)

6 for the ‘spindex’ variable. This appeared to be the Hess Corporation which is an Oil company so these observations were given an index code (‘spindex’) of 1010.

So after these main mutations of the database the cleaned up dataset on executive compensation still consists of 11.510 observations from 1.918 firms. An overview of the summarized statistics of the variables on executive compensation is listed below in Table 1. When you are looking at the number of observations you see that not all observations include data on the compensation (11.510 observation versus 11.331 data on the compensation). These observations with missing values are dropped afterwards which is described in the methodology paragraph (2.3).

Table 1 – Data on Executive Compensation from ‘Compustat ExecuComp’

Variable Obs Mean Std. Dev Min Max

coname 11510

year 11326 2006 2014

stock_awards 11331 1836.504 4623.374 -1845.69 376180

option_awards 11331 1118.227 3016.642 -0.011 90693.4

total_sec 11331 5397.762 7393.518 0 377996.5

2.2 Volatility of Share Prices

The first measure for the firm volatility as dependent variable is the standard deviation of share prices per year. To collect these data I use the ‘Thomson Reuters Datastream’ database. From this database we can collect data of all daily stock prices (‘P: Price (Adjusted -

Default)’) for all S&P 1500 firms (‘LSPSUP’) in the period of 2006 until 2014. First I split

these daily stock prices for every fiscal year. After having this done I calculated for every firm their yearly standard deviation of the share prices. This variable I named

‘SD_SharePrice’. An advantage that makes the stock prices a very appropriate variable to

define firm volatility is that it contains daily data which makes it possible to calculate very accurate standard deviations covering all fluctuations of the firms. The dataset consists now in total 12.879 observations from 1.498 firms. In Table 2 below you see an overview of the summarized statistics of this dataset.

Table 2 –Data on Share Price Volatility from ‘Datastream’

Name 12879

Year 12879 2006 2014

(7)

7

2.3 Volatility of Firm Profits

The second measure to define firm volatility as dependent variable is the standard deviation of the profits of the firms. The yearly data on profits I collect from the ‘Compustat Noth America’ database. The data on the firms profit are measured by the Earnings Before Interest and Taxes (‘ebit’) variable. In the methodology will be explained how I edited the dataset to create standard deviations from these data to compare them to the compensation data of ExecuComp. In Table 3 below you find an overview of the raw dataset of the Compustat database.

Table 3 – Data on Firm Profits from ‘Compustat North America’

coname 99083

fyear 99083 2006 2014

ebit 78332 437.4272 2765.728 -80053 130622

2.4 Methodology

After the first cleanup of the data as described above, the statistical analyses in Stata can start. For all three datasets we created a new identifying variable by concatenating the ‘ticker’ and

‘year’ variables to an new ‘TickerYearCode’ variable. With this new variable as a key

variable, the databases can be merged into one new dataset to run the regressions. All data that cannot be merged because there are no data on that specific company in that year in both databases are being dropped. Also all observations for which there are missing values for the executive compensation are dropped.

2.4.1 Regression 1: standard deviation of share prices as dependent variable (SD_SharePrice) For the first regression of the standard deviation of the share prices on the relative size of performance-based compensation, the ExecuComp and the Datastream datasets must be merged. After this merge there are 8.146 observations left from 1.178 firms. Both datasets consist of respectively 11.510 and 12.879 observations. So that means that over 3.000 observations are missing after the merging process. That the merging process drops so many observations seems strange since both datasets contain data on the same group of companies (S&P 1500) for the same period of time (2006-2014).

(8)

8 A probable explanation for this could be the difference in covered companies in both databases. You already see a difference in the amount of firms both databases cover data for in their raw data overviews (1.918 versus 1.498 firms). First cause for this is that the Datastream database still got the financial firms in its raw data described above. Because I already dropped the financial firms out in the ExecuComp database it was not necessary to also do this manually for the Datastream database because after the merge they would also drop out. That’s the main cause for the loss in amount of observations. Another possible cause is the different approach to which companies are covered in the dataset. The Datastream dataset only consists of firms that are still in the S&P 1500 at the moment in 2014. For example for 2006 it only has data on 1345 firms. ExecuComp covers data on all companies that were in the S&P 1500 for that year. This creates missing values if you are merging both databases.

After the merging process is finished we have one large set of data. With the known data on the executive compensation we can now generate new variables on the total amount of performance based awards (‘perf_based_awards’). By comparing this new generated variable to the total amount of compensation (‘total_sec’) makes it possible to generate a new variable on the relative amount of performance based awards (‘RPBA’).

With the standard deviation of the share prices (‘SD_SharePrice’) and the relative size of performance-based compensation (‘RPBA’) we do now have both main variables we were looking for to be able to run the first regression.

2.4.2 Regression 2: standard deviation of profits as dependent variable (SD_EBIT)

For this second regression the ‘Compustat North America’ database with the data on the yearly profits must be merged with the ‘ExecuComp’ database. In the database of the profits all listed companies in the US and Canada are covered so after the merge a lot of observations from firms that are not in the S&P 1500 are dropped. With the profit (‘ebit’) data of the Compustat database I can calculate the standard deviation of the profits per firm (‘SD_EBIT’). For every firm this is calculated over the total period of years of which the data are available for that firm. For this same period of time I calculate the average percentage of performance-based compensation (‘Mean_RPBA’). Since I only use now one observation per firm, the amount of observations is considerably lower with 1732 observations.

(9)

9 Since every firm only got one observation the Compustat dataset has only 1732 observations. These are for 100% merged with the ExecuComp dataset without any observation dropped, so after the merge we have 1732 observations left.

With the standard deviation of firm profit (‘SD_EBIT’) and mean of the relative size of performance-based compensation (‘Mean_RPBA’) we do now have both main variables to be able to run the second regression.

2.4.3 Control variables

However, since the fluctuations in firm volatility are influenced by many other factors we should also try to find a way to control for these influences as good as possible.

To control for industry specific risk there are also dummies created. Since we are studying a very broad range of firms from all kinds of industries there are several different industry specific effects on the volatilities of the firms that should be controlled for. Some industries have happen to be less volatile than others. In the ExecuComp database all firms have a different ‘spindex’ code which assigns all companies to one of the 20 different industries. This creates the following dummy variables: DIndustry i | i ∈{1,…,20}. In the Appendix (7.2)

you can find which different industry corresponds to which dummy variable number.

For instance for the instable economic environment in which the firms have operated over the last few years. In most severe years of the crisis almost all firms were exposed to higher volatilities than in the years before. This was not caused by higher risk-taking by the executives because of their performance-based compensation. This would give a skewed image of reality. To control for this we add dummies for all average fluctuations in the share prices per year separately. So we add the following dummy variables: Dj | i ∈{2006,…,2014}

to control for these possible different effects of the financial crisis. 2.4.4 Statistical Models

Now we do have all main variables and all controlling dummy variables we can compose both models we want to study:

Regression 1: SD_SharePrice = β0 + β1RPBA + β1+iDIndustry i + βj-1984Dj

[ DIndustry i | i ∈{1,…,20} and Dj | j ∈{2006,…,2014} ]

Regression 2: SD_EBIT = β0 + β1Mean_RPBA + β1+iDIndustry i + β j-1984Dj

(10)

10

3. Results

Running these both regressions gives the following results as listed below in Table 4a (2.4.1 – ‘SD_SharePrice’) and Table 4b (2.4.2 – ‘SD_EBIT’):

Table 4a – Regression 1 (Share Prices) Table 4b – Regression 2 (Profits)

SD_SharePrice Coef. P>|t| SD_EBIT Coef. P>|t|

RPBA 0.939896 (0.001) Mean_RPBA 397.386 (0.000) DIndustry1 1.880703 (0.004) DIndustry1 265.312 (0.178) DIndustry2 1.226611 (0.059) DIndustry2 – 108.622 (0.580) DIndustry3 1.184423 (0.064) DIndustry3 – 148.750 (0.441) DIndustry4 – 0.289566 (0.674) DIndustry4 – 309.502 (0.141) DIndustry5 – 0.025194 (0.972) DIndustry5 – 130.008 (0.558) DIndustry6 – 0.434080 (0.613) DIndustry6 238.711 (0.335) DIndustry7 2.584117 (0.000) DIndustry7 – 193.994 (0.341) DIndustry8 3.664939 (0.000) DIndustry8 – 277.229 (0.174) DIndustry9 – 0.211045 (0.779) DIndustry9 – 61.719 (0.777) DIndustry10 1.875651 (0.004) DIndustry10 – 239.909 (0.224)

DIndustry11 0 (omitted) DIndustry11 0 (omitted)

DIndustry12 – 0.224817 (0.745) DIndustry12 – 121.742 (0.562) DIndustry13 0.083567 (0.926) DIndustry13 – 264.486 (0.324) DIndustry14 0.710401 (0.272) DIndustry14 – 279.465 (0.151) DIndustry15 0.883881 (0.204) DIndustry15 – 179.396 (0.375) DIndustry16 0.110348 (0.865) DIndustry16 – 255.810 (0.183) DIndustry17 0.005697 (0.993) DIndustry17 – 100.383 (0.608) DIndustry18 – 0.835104 (0.224) DIndustry18 – 298.268 (0.143) DIndustry19 – 0.920066 (0.278) DIndustry19 166.334 (0.490) DIndustry20 – 1.207403 (0.072) DIndustry20 – 161.783 (0.431) D2006 0 (omitted) D2006 0 (omitted) D2007 0.862212 (0.006) D2007 13.788 (0.887) D2008 2.954943 (0.000) D2008 38.139 (0.693) D2009 0.942371 (0.002) D2009 18.084 (0.854) D2010 0.300525 (0.323) D2010 – 20.563 (0.836) D2011 0.691102 (0.022) D2011 48.305 (0.623) D2012 0.240214 (0.422) D2012 105.031 (0.276) D2013 1.709569 (0.000) D2013 – 7.006 (0.942) D2014 1.527060 (0.000) D2014 – 18.943 (0.879) _cons 2.120405 (0.001) _cons 180.187 (0.377)

(11)

11 From the results as listed in the two tables above we find that both relative performance-based compensation (RPBA) variables do indeed have a positive effect on the volatility of the firm. Both of these coefficient are also significant with p-values of respectively 0.001 and 0.000.

As you can see both coefficients have a positive value. Although the coefficient of the first regression on the standard deviation of the share prices seems low compared to the coefficient of the second regression this is a bit of a distorted picture since you see in the data overview tables that the standard deviations of the profits are generally much higher than those of the share prices. Relatively they are somehow equal and both relatively small coefficients. Something that was also already shortly memorized in the hypothesis paragraph (1.2).

Another thing that must be taken into account in the consideration of the value of the proposed model are the values of the statistics on the R2 and the F-statistic. The outcomes of these statistics are listed below in Table 5a and 5b:

Table 5a – Regression 1 (Share Prices) Table 5b– Regression 2 (Profits)

Observations 8046 Observations 1731

F (28, 8017) 18.50 F (28, 1702) 2.72

Prob > F 0.0000 Prob > F 0.0000

R2 0.0607 R2 0.0428

Adj R2 0.0574 Adj R2 0.0270

As you can see in the tables above both regressions have low R-squared values. This indicates that the fluctuations in the firm performances are only for a very small part explained by the relative performance-based compensation. So only a very small percentage of the volatility of the firms (6% and 4% respectively) is explained by model. The reason for this is that there are probably many more factors that influence the firm volatility which are not yet considered and controlled for in this thesis. These shortcomings of this study will be discussed later on in the next chapter.

That there is a relationship between the ‘RPBA’ and the volatility of the firm performances can in both regression considered to be true since both models have a probability of the F-statistic of 0.000. This means that there is certainly some relationship between the ‘RPBA’ and the both volatility measures. But this influence is as calculated by the R2 quite small.

(12)

12 The graphs of both regressions with all observations are shown in Graphs 1a and 1b below: Graph 1a – Regression 1 (SD_SharePrice, including all outliers)

Graph 1b – Regression 2 (SD_EBIT, including all outliers)

What immediately strikes in both graphs is that both regressions have a number of large outliers that affect the graphical presentation of the regression. In these graphs, with a scale such that all observations are covered in the graph, it is more difficult to see the positive slope of the regression line. But what you can see is that the outliers are more at the right side of the graph. Although there are also a few firms with a performance-based compensation of 0 with a high standard deviation, most of the firms with a high standard deviation value also have higher percentages of performance-based compensation.

(13)

13 For a better overview of the graphs, the scale is changed to exclude these outliers. Such that you can better see the positive slope of the regression line corresponding to the coefficients as described in the result tables (5a an 5b) above. The outlier value to change the scale of the new graphs is for both regressions determined by the mean value plus three times its own standard deviation. This gives the following results for the scales of the outlier values:

- Outlier value for ‘SD_Share Price’ = 4.27 + (3 * 6.45) = 23.62 (74 outliers) - Outlier value for ‘SD_EBIT’ = 236.4653 + (3 * 842.8995) = 2765 (19 outliers) Graph 2a – Regression 1 (SD_SharePrice, excluding all outliers)

(14)

14 In these renewed graphs 2a and 2b, as showed on the previous page, it’s easier to see the positive slope of the regression line. A possible effect of these outliers on the original regressions could be that they pulled up the regression line (it’s coefficient). To see if the excluding of the outliers has this effect of a decreasing slope of the regression line we will check below by dropping the outliers as calculated for the used scales of graphs 2a and 2b. The outcomes of these newly generated regressions compared to the original models are listed below in Table 5a an 5b.

Table 6a – excluding outliers

Regression 1 – SD_SharePrice

Original Excl. outliers RPBA 0.9398963 .9878199

(p-value) (0.001) (0.000)

R2 6.07% 11.68%

Table 6b – excluding outliers

Regression 2 – SD_EBIT

Original Excl. outliers

RPBA 397.386 351.652

(p-value) (0.000) (0.000)

R2 4.28% 10.17%

As described above you would expect lower RPBA coefficients after dropping the outliers because these outliers would have pulled up the regression line in the original model. What you see in reality for the first regression on Share Prices (Table 6a) is that this is not the case and that the coefficient even increased a little bit. For the second regression on profits (Table 6b) you do see a small drop in the RPBA coefficient. So after dropping of the outliers both coefficients doesn’t change a lot and also not in one single direction, so this small effect would probably not be a big issue to consider. So from these results I conclude that the distorting effect of the outliers is not a very important one. Also because the amount of other observations is so large that it doesn’t have a large impact.

But what you do see in the Tables 6a and 6b is a quite large improvement of the R-squared statistics in both models. This indicates that the models without the outliers explain the relationship between the both studied variables better without the outliers. This could indicate that the first R-squared values of the original regression could have been the reason for a a little distorted picture. But the fact remains that still now the R-squared values are still quite low so there are definitely many other factors which influence the volatility of firm performances other than the relative size of performance-based compensation only.

(15)

15

4. Discussion

In the above paragraphs of the data and methodology I already mentioned some shortcomings of the data and of the analysis of this thesis.

First important issue to mention is the time period which I am studying in this thesis. The years studied in this thesis were very instable because of the financial crisis. Since comparing with the years before 2006 is impossible because of the introduction of the new reporting system of the SEC fillings it’s hard to say what the influence of the economic crisis has been on the outcomes of this study. The added control variable dummies per year are of course an attempt to control for this factor as good as possible but this is of course not a full control. In this thesis in case of simplicity I didn’t use other control variables, which can be possible in more extensive studies. The yearly control variable used in this study is of course a very general one which could be replaced by more specific measures. You could think of a combined yearly/industry specific control variable or even better with more firm specific trends in for example profits. If you would combine these more specific controlling variables with the fluctuations in the ‘RPBA’ your results will probably be less diverse. A first option for the analysis of this thesis was also a panel data regression in blocks of 3 years with the profits of the firms. Because of the available data this was technically not possible in a bachelor thesis but for future studies this might be an opportunity to study.

Another shortcoming of the analyses which I already shortly mentioned in the previous chapter of the results (3) is the low R2 value of both regressions. This indicates that the relationship between the studied data is only for a very small part explained by the model. So only a very small percentage of the volatility of the firms is explained by the models. The process of excluding the outliers shows that this increases the R2 values as well: from 6% to 11.7% for regression 1 and from 4.3% to 10.2% in the second regression (see Table 6a and 6b). A significant increase which could of course also be an explanation for the initially low values of R2 of both models. That the outliers had such a big impact on the regressions that this gave a higher image than the explanatory relationship actually was. But this of course doesn’t explain fully for the low R2 value, the most important reason is that there are probably many more factors that influence the firm volatility which are not considered in this thesis. This is for example the influence of the state of the economy as described above. But also many other factors influence the firm profit. Very basic factors that are highly firm specific

(16)

16 and also change yearly such as the degree of competition, the strength of demand, available substitutes, a successful advertising campaign, the degree of costs etc. All these factors and many more have an important influence on the profits of the firms that is not affected by the compensation structure. Including all these other potential impacting variables in a model was for this bachelor thesis not possible but for a renewed study this could be taken into account.

All these possible factors that could have an influence on the profit will be difficult to account for in a general model for the broad range of firms as studied in this thesis but a possible solution could be to split the study into different models for different industries and only focus on a specific industry. This could exclude some of the above mentioned factors that influence the volatility of the firms and create a model with a higher explanatory value. Last thing to mention is the observation dropping in the merging process of the first regression on the standard deviations of share prices (‘SD_SharePrice’). As already described in the methodology paragraph (2.4.1), more than 3.000 observations are dropped in the merging process which can be seen as some sort of selection. As discussed in the methodology part this could have several reasons. One of them is the different approach of both databases of which firms they cover in their databases. The Datastream database only consists of firms that are still in the S&P 1500 at this time while the ExecuComp databse covers data on all companies that were in the S&P 1500 for that specific year. This creates missing values if you are merging both databases. The first reason for the missing observations is in this case is when a firm goes bankrupt. And precisely these firms are interesting to study because they probably got highly volatile performance measures. The next group of companies that drops out for the above described reason are the firms are at the smaller firms which are at the downside of the S&P 1500 index. If they drop out of the index at some point this also creates missing values and drops in the merging process.

(17)

17

5. Conclusion

This thesis analyzes what the influence of the relative size of the performance-based part of the compensation of CEO’s is on the volatility of firms with the volatility defined first as the standard deviation of share prices and in the second regression as the standard deviation of the profits of the firm. Since this had already been studied in the financial sector this research focusses on the non-financial firms. Just like the hypothesis expected we find a significant positive influence of the relative performance-based compensation on the volatility of the firms for both regressions. The coefficient of the relative performance-based compensation (RPBA) are in both models significant but what we do see also in both regressions is a very low R2. This low R2 is not very strange if you consider all other possible influences on the firm performances. Even though these both models predict only a very small part of the volatility the calculated F-statistic and the significant coefficients confirm the H1 hypothesis

that the relative size of the performance-based compensation has positive effect on the volatility of firms in the non-financial sector.

In addition must be said that the influence is only a very small one. So you could ask if all the commotion about the excessive risk-taking effects of performance-based compensation is necessary. All the legal restriction politicians are willing to introduce may seem a bit overdone if you consider all other possible influences that have way larger effects on the firm volatility and could also be legally restricted.

(18)

18

6. References

Aggarwal, R. and Samwick, A., 1999. The Other Side of the Trade-Off: The Impact of Risk on Executive Compensation. Journal of Political Economy, 107(1), 65–105

Belghitar, Y. and Clark, A., 2012. The Effect of CEO Risk Appetite on Firm Volatility: An Empirical Analysis of Financial Firms. International Journal of the Economics

of Business, Vol.19(2), 195–211

Chen, C., Steiner, T. and Whyte, A., 2005. Does stock option-based executive compensation induce risk-taking? An analysis of the banking industry. Journal of Banking &

Finance, 30(3), 915–945

Chen, Y. and Ma, Y., 2011. Revisiting the risk-taking effect of executive stock options on firm performance. Journal of Business Research, 64(6), 640–648

Coughlan, A. and Schmidt, R., 1985. Executive compensation, management turnover, and firm performance. Journal of Accounting and Economics, Vol.7(1), 43–66

Efing, M., Hau, H., Kampkötter, P. and Steinbrecher, J., 2014. Incentive Pay and Bank Risk Taking: Evidence from Austrian, German, and Swiss Banks. Journal of

International Economics

Gehrig, T., Lütje, T. and Menkhoff, L., 2009. Bonus Payments and Fund Managers’ Behavior: Transatlantic Evidence. CESifo Economic Studies, 55(3-4), 569-594

Gregg, P., Jewell, S. and Tonks, I., 2012. Executive pay and performance: Did bankers’ bonuses cause the crisis? International Review of Finance, 12(1), 89–122.

Jensen, M. And Murphy, K., 1990. Performance Pay and Top-Management Incentives. Journal of Political Economy, 98(2), 225–264

(19)

19

7. Appendix

7.1 Dropped financial firms

An overview of the dropped sectors with financial firms. These are dropped from the data from the ‘Compustat ExecuComp’ database on Executive Compensation (2.1):

inddesc

(Industry Group Description)

spindex

(Industry Group)

Diversified Banks 4010

Regional Banks 4010

Thrifts & Mortgage Finance 4010

Diversified Capital Markets 4020

Multi-Sector Holdings 4020

Consumer Finance – Discontinued effective 05/01/2 4020

Consumer Finance 4020

Other Diversified Financial Services 4020

Asset Management & Custody Banks 4020

Investment Banking & Brokerage 4020

Specialized Finance 4020

Insurance Brokers 4030

Reinsurance 4030

Life & Health Insurance 4030

Multi-line Insurance 4030

Property & Casualty Insurance 4030

Diversified REITs 4040

Diversified Real Estate Activities 4040

Health Care REITs 4040

Hotel & Resort REITs 4040

Industrial REITs 4040

Mortgage REITs 4040

Office REITs 4040

Real Estate Development 4040

Real Estate Investment Trusts – Discontinued effe 4040

Real Estate Services 4040

Residential REITs 4040

Retail REITs 4040

Specialized REITs 4040

(20)

20

7.2 List of the industry dummies

An overview of the used industry dummy variables with which dummy number corresponds to which industry.

Industry 'spindex' Dummy

Energy 1010 DIndustry1

Materials 1510 DIndustry2

Capital Goods 2010 DIndustry3

Commercial & Professional Services 2020 DIndustry4

Transportation 2030 DIndustry5

Automobiles & Components 2510 DIndustry6

Consumer Durables & Apparel 2520 DIndustry7

Consumer Services 2530 DIndustry8

Media 2540 DIndustry9

Retailing 2550 DIndustry10

Food & Staples Retailing 3010 DIndustry11

Food, Beverage & Tobacco 3020 DIndustry12

Household & Personal Products 3030 DIndustry13

Health Care Equipment & Services 3510 DIndustry14

Pharmaceuticals, Biotechnology & Life Sciences 3520 DIndustry15

Software & Services 4510 DIndustry16

Technology Hardware & Equipment 4520 DIndustry17

Semiconductors & Semiconductors Equipment 4530 DIndustry18

Telecomunications Services 5010 DIndustry19

Utilities 5510 DIndustry20

7.3 STATA Do-File (SD_SharePrice)

1. keep if pceo == "CEO"

2. drop if spindex == 4010, 4020, 4030, 4040, 0 3. drop if year <= 2005

4. egen id = group(coname)

5. drop co_per_rol salary bonus noneq_incent stock_awards_fv othann tdc1 tdc2 option_awards_blk_value ltip allothtot gvkey

6. encode coname, generate(coname1) 7. drop coname

8. rename coname1 coname

9. encode pceo, generate(pceo1) 10. drop pceo

11. rename pceo1 pceo

12. order coname year stock_awards option_awards total_sec pceo spindex inddesc id ticker, first

(21)

21 13. sort spindex 14. replace spindex = 1010 in 11503 15. replace spindex = 1010 in 11504 16. replace spindex = 1010 in 11505 17. replace spindex = 1010 in 11506 18. replace spindex = 1010 in 11507 19. replace spindex = 1010 in 11508 20. replace spindex = 1010 in 11509 21. replace spindex = 1010 in 11510

22. summarize coname year stock_awards option_awards total_sec pceo spindex 23. egen TickerYearCode = concat(ticker year)

24. merge m:m TickerYearCode using "C:\Users\6286844\Dropbox\Studie\Bachelor Thesis - Joris Boelens\STATA\Datastream.dta

25. keep if _merge==3 26. drop if total_sec >= .

27. gen perf_based_awards = stock_awards + option_awards 28. gen RPBA = perf_based_awards / total_sec

29. generate Dindustry1 = 0

30. replace Dindustry1 = 1 if spindex == 1010

(22)

22

69. generate D2006 = 0 70. replace D2006 = 1 if year == 2006 71. generate D2007 = 0 72. replace D2007 = 1 if year == 2007 73. generate D2008 = 0 74. replace D2008 = 1 if year == 2008 75. generate D2009 = 0 76. replace D2009 = 1 if year == 2009 77. generate D2010 = 0 78. replace D2010 = 1 if year == 2010 79. generate D2011 = 0 80. replace D2011 = 1 if year == 2011 81. generate D2012 = 0 82. replace D2012 = 1 if year == 2012 83. generate D2013 = 0 84. replace D2013 = 1 if year == 2013 85. generate D2014 = 0 86. replace D2014 = 1 if year == 2014

87. reg SD_SharePrice RPBA Dindustry1 Dindustry2 Dindustry3 Dindustry4 Dindustry5 Dindustry6 Dindustry7 Dindustry8 Dindustry9 Dindustry10 Dindustry11 Dindustry12 Dindustry13 Dindustry14 Dindustry15 Dindustry16 Dindustry17 Dindustry18 Dindustry19 Dindustry20 D2006 D2007 D2008 D2009 D2010 D2011 D2012 D2013 D2014

(23)

23

7.4 STATA Do-File (SD_EBIT)

1. keep if pceo == "CEO" 2. drop if spindex == 4010 3. drop if spindex == 4020 4. drop if spindex == 4030 5. drop if spindex == 4040 6. drop if spindex == 0 7. drop if year <= 2005 8. egen id = group(coname)

9. drop co_per_rol salary bonus noneq_incent stock_awards_fv othann tdc1 tdc2 option_awards_blk_value ltip allothtot gvkey

10. encode coname, generate(coname1) 11. drop coname

12. rename coname1 coname 13. encode pceo, generate(pceo1) 14. drop pceo

15. rename pceo1 pceo

16. order coname year stock_awards option_awards total_sec pceo spindex inddesc id ticker, first 17. sort spindex 18. replace spindex = 1010 in 11503 19. replace spindex = 1010 in 11504 20. replace spindex = 1010 in 11505 21. replace spindex = 1010 in 11506 22. replace spindex = 1010 in 11507 23. replace spindex = 1010 in 11508 24. replace spindex = 1010 in 11509 25. replace spindex = 1010 in 11510

26. summarize coname year stock_awards option_awards total_sec pceo spindex 27. egen TickerYearCode = concat(tic year)

28. merge m:m TickerYearCode using "C:\Users\6286844\Dropbox\Studie\Bachelor Thesis - Joris Boelens\STATA\Compustat.dta

29. keep if _merge ==3 30. drop if total_sec >= .

31. gen perf_based_awards = stock_awards + option_awards 32. gen RPBA = perf_based_awards / total_sec

33. bysort id: egen Mean_RPBA = mean(RPBA) 34. bysort id: egen SD_EBIT = sd(ebit)

35. by id: gen firstobs = (_n==1) 36. keep if firstobs

37. drop if SD_EBIT >= . 38. drop if ebit >= .

(24)

24

79. generate D2006 = 0 80. replace D2006 = 1 if year == 2006 81. generate D2007 = 0 82. replace D2007 = 1 if year == 2007 83. generate D2008 = 0 84. replace D2008 = 1 if year == 2008 85. generate D2009 = 0

(25)

25 86. replace D2009 = 1 if year == 2009 87. generate D2010 = 0 88. replace D2010 = 1 if year == 2010 89. generate D2011 = 0 90. replace D2011 = 1 if year == 2011 91. generate D2012 = 0 92. replace D2012 = 1 if year == 2012 93. generate D2013 = 0 94. replace D2013 = 1 if year == 2013 95. generate D2014 = 0 96. replace D2014 = 1 if year == 2014

97. reg SD_EBIT Mean_RPBA Dindustry1 Dindustry2 Dindustry3 Dindustry4 Dindustry5 Dindustry6 Dindustry7 Dindustry8 Dindustry9 Dindustry10 Dindustry11 Dindustry12 Dindustry13 Dindustry14 Dindustry15 Dindustry16 Dindustry17 Dindustry18 Dindustry19 Dindustry20 D2006 D2007 D2008 D2009 D2010 D2011 D2012 D2013 D2014