THE RELATIONSHIP BETWEEN CEO COMPENSATION PERFORMANCE IN DIABETES SECTOR

THE

RELATIONSHIP

BETWEEN

CEO

COMPENSATION

PERFORMANCE

IN

DIABET

ES

SECTOR

Finance Bachelor Thesis

University of Amsterdam

Amsterdam Business School

Statement of Originality

This document is written by student Liam Sanders who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents

Table of Contents

Though executive compensation and firm performance has been the topic of much research over the last decade, it is still unclear how this relationship holds when it comes to pharmaceutical firms producing diabetes products. To examine this relationship a sample of 6 firms performance and compensation metrics were taken over the years 2000-2019. We expected to find a positive relationship between compensation and performance. The results supported the expectation that higher levels of CEO compensation are associated with higher performance outcomes, with several performance variables (EPS, Revenue, Total Average Assets, ROA) being statistically significant predictors of CEO compensation.

1) Introduction

Diabetes mellitus, or simply diabetes, is an autoimmune disorder which attacks the pancreas. The pancreas is responsible for the production of insulin, a hormone that converts glucose into workable energy within our bodies (Mellitus, 2005.) Diabetes is a serious illness affecting more than 10.5% of the American population, with it being the 7th leading cause of death in America (National Diabetes Data Group,1995.) Prior to the discovery of insulin, diabetes was a death sentence. (Lakhtaki, 2013.) That was until the 11th January, 1922, when Frederick Banting discovered the use of insulin for treatment of diabetes. Following his

discovery, he altruistically sold his patent to the University of Toronto for one dollar, hoping to ensure everyone who had the disease could receive affordable treatment (Simoni, Hill & Vaughn, 2002.). Almost 100 years later, his discovery is now being capitalized by large pharmaceutical firms setting high prices to turn a profit. Today insulin can cost hundreds of dollars a month, despite manufacturing being a fraction of the cost. The Federal Reserve found 44% of Americans could not cover a 400$ emergency expense fee (The Fed - Dealing with Unexpected Expenses, 2019), thus it is no surprise that everyday Americans with diabetes struggle to procure their life-saving medicine (Hersch, 2016.)

When the repercussions of pricing choices can be measured in deaths (Rosenthal, 2019) many are appalled when they learn that the CEOs of the insulin manufacturers are making millions of dollars a year in total compensation. Salary along with bonuses, stock grants and options are powerful mechanisms to retain and attract a competent management force (Anderson and Bizjack, 2003.) These compensation packages are also used to establish long term relationships with an executive and the corporation, as well as align the interests of managers with the interests of the shareholders. Thus, in theory one may conclude that their compensation is allotted due to their competence and prestige which in turn drives higher performance, however empirical evidence is still inconclusive as to whether CEO compensation packages drives higher performance. (Conmyon et al, 1995.) Throughout this paper, we aim to investigate the relationship between CEO pay and firm financial performance for companies producing diabetes supplies.

Throughout this paper, six pharmaceutical firms were analyzed over a 20-year span. Data involving both the levels of compensation, as well as several performance metrics were tested to find whether a statistically significant relationship between firm performance and compensation exists. Expanding upon existing literature, our a priori expectations were that higher levels of compensation is granted in reward as an incentive to improve the value of the firm. This work can be added to pre-existing literature to strengthen the knowledge we have on the pay for performance relationship that exists within the pharmaceutical industry.

This paper begins with a literature review highlighting the most relevant information regarding the relation between performance and compensation, as well as some crucial insights regarding the market for diabetes supplies. Further, to answer the research question, data involving firm performance and compensation will be analyzed and the methodologies used will be described. Upon completion of the statistical analysis, the results will be outlined and discussed. Finally, the implications of the findings will be outlined, and we will discuss how these results can be added to existing literature to strengthen our understanding of the topic at hand.

2) Literature Review

To fully grasp the relation between executive pay and performance, one must first step back and investigate the characteristics surrounding the diabetes market. Insulin production is a perfect exemplification of an oligopoly (Gale & Yudkin, 2011.) There are three corporations producing 90% of the world’s insulin supply: Eli lily, Novo Nordisk and Sanofi. These companies can set their own pricing and by ‘teaming-up’ they are able to continuously raise the cost of medicine (Gotham, Barber & Hill, 2018.) Furthermore, despite the fact that the patent for insulin is widely available, these companies use clever legal loopholes where they are able to exploit their patents through a process known as evergreening (Hemphillm C.S & Sampat, 2012.) Here, firms can create bio-similars to extend their patents, further protecting their position in the market. Additionally, through clever patent loopholes, firms will also pay millions of dollars out in what is known as a “pay for delay scheme,” ensuring that only they can produce that particular strand of insulin, further protecting their stake in the diabetes markets (Fialkoff & Michael, 2013.)

Executive compensation is a governance mechanism designed to incentivize managers to act in the interest of its shareholders. By tying incentive plans, such a bonus or stock grants, to good performance, it is seen to motivate managers to act in the best interest of the company (Abowd & Kaplan, 1999.) Thus, by tying monetary compensation to performance one would expect a positive relationship between pay and performance (Brick, Palmon & Wald, 2006.) There are countless research papers that support this position. Hall and Lierblan (2000) found that the relationship between pay and performance was positive, particularly when it came to equity-based incentives, such as stock grants. Equity incentive programs help to align the interest of the shareholder with that of the CEO, since the stock compensation has more value if the stock appreciates, so the theory goes that CEO’s will not engage in unnecessary risk-taking behavior. This work, along with papers by Murphy (1999) and Core et al. (2003,) found that incentive programs confirms the value of compensation when it comes to aligning managerial interest with that of the shareholders.

However, many modern scholars refute this theory, most notably explained in the principle agent theory. Under the Anglo-Saxon model of corporate governance, the principle-agent theory stipulates that agents, or managers, are responsible for safeguarding the interests of principles, but do not always protect stockholder interests and will act in their own self-benefit(Garen, 1994.) This theory is particularly of interest when you consider that it is not stockholders who determine the appropriate level of compensation for CEO’s. This power is given to the non-executive directors, more commonly referred to as the Supervisory board (Pierer Duffhues, 2008.) Given that stockholders do not influence the pay, many argue that compensation is not granted fairly (Oxelheim and Clarkson, 2015)

There are numerous factors that could lead board members to giving excessive compensation plans that do not reflect positive performance. Firstly, CEO compensation is often granted in comparison to similar (industry specific) firms of comparable size (Bizjack, Lemmon and Naveen 2008). Though in theory this should be equitable, as it provides a ‘reasonable’ point of reference, it was also found that firms tend to select other ‘comparison’ companies which set its compensations level highest so they can over compensate their own CEO. (Faulkender and Yang, 2010.) Further, some argue that CEO compensation contains an element known as a social premium (Ang, Nagel and Yang, 2014.) Through frequent interaction between board members and

the CEO, such as from attending charity or industry events, golfing together or corporate meals, the supervisory board comes to ‘befriend’ or acquaint themselves better with the CEO. This paper found that increased social interaction and a closer proximity of board members tends to result in higher levels of compensation when controlling for external factors.

Though there is extensive research into whether or not higher levels of compensation drive higher performance, there remains a gap in knowledge in how this relationship holds for pharmaceutical firms producing either insulin or diabetes supplies. Throughout this paper, we aim to fill this gap by deconstructing CEO compensation into its individual components and see how performance drives compensation. Based on the evidence suggesting that compensation and performance are positive related we will run multiple regressions on several different

performance and compensation metrics to ensure the ‘whole’ relationship is captured. This hypothesis will be tested through ordinary least squares regressions, with a 5% significance value.

Hypothesis 1: There is a positive relation between executive compensation and firm performance

This hypothesis implies an alternative hypothesis that there is not a positive relationship between executive compensation and firm performance. Failing to accept the null hypothesis implies that there is either no relation between compensation and performance or that there is a negative relationship between compensation and performance.

3) Methods

To test hypothesis one, we examined the data utilizing an OLS (ordinary least squares) regression. Following the model prescribed by the regression equation, we examined the relationship between firm performance and CEO compensation. The methods and tests implemented will be elaborated on below.

Participants

To outline the methods implemented into the research, we shall begin by discussing the participants. To capture the effect of compensation onto performance, we opted for a core sample group of six corporations all producing diabetes supplies. These companies are Abbott Laboratories, Johnson and Johnson, Eli Lilly, Medtronic, Pfizer and Baxter. Apart from Medtronic, which is based in Ireland for tax purposes, these are all American firms. These companies were selected due to their geographical location and readily available stock/financial information, as they were all subject to publish 10-K statements, annual reports, and pro forma financial information. Only six of the largest publicly traded diabetic firms were chosen because they were most representative of the relationship we were aiming to examine. More firms were initially included, however were ultimately dropped due to their foreign origin and/or lack of publicly available data.

Design/materials

Through the Wharton Data Base, as well as through company posted annual proxy/10-K statements, data regarding CEO compensation was collected for the years 2000-2019. This time period was selected, as having a longitudinal study allowed us to best examine the relationship between performance and compensation and see how this relationship has evolved over time. The data collected included details on CEO salary, CEO bonus, total current compensation, which is combination of the two. In addition, we examined stock grants and options, as well as a variable combining all elements of compensation (named total comp.) In addition to the variables listed, several performance metrics were gathered and analyzed. To capture stock performance, annual log returns were implemented. Log returns were implemented due to its time additive properties, as well as its mathematical convenience. Further, other barometers such as return on assets, revenue, net income and EPS were also analyzed. These additional variables were analyzed to ensure that performance was not limited to a singular measure that could be distorted by macro-economic trends. In addition, characteristics such as gender were included as dummy variables, but because all CEOs were found to be men, this had no implication on the study. Further worth noting is the inclusion of a dummy variable, crisis, which took on a value of 1 if the associated year was during an economic recession and a 0 if it was not to control for the effect of systematic

shocks across the economy. Additionally, a final dummy variable USnot was included to control for the effect of being based in the US.

Measurements of Variables

Compensation: To begin the analysis, we began by deconstructing the elements of CEO compensation into their individual components. Firstly, we began with the variable total

compensation. Total Compensation was an accumulation of all other metrics of compensation, namely it combined salary, bonus and stock options/grants. While analyzing stock

options/grants, it was important to recognize that these values are stated in terms of the worth according to their FSR 123R reports. Further worth noting is that Total Current Comp is a combination of both salary and bonus. Salary was a straightforward variable and one that was easily found either through the CompuStat data base or individual company reports. Bonus was also easily collected, but it is important to note that in addition to just a flat-out bonus,

compensation in the form of other compensation was added to the variable bonus to simplify analysis. In addition to these variables, most of the regressions were run in logarithmic form meaning that changes in the dependent variable were caused by a 1% change in our independent variables.

Firm Performance:

To measure firm performance, many different variables were used and analyzed throughout the regressions. One of the first variables taken into consideration was revenue. Revenue showed us the income earned through the firm’s operations and was a powerful proxy to illustrate how much sales/value the corporation was producing without considering the costs of those operations. Another key figure was the return on assets, this metric served to illustrate how profitable a company was given its total assets. Therefore, to calculate this value, we needed additional variables: total average asset and the company’s net income. Total average assets are determined by the average amount of assets taken from the company’s balance sheet at the end of the year. Once we found the net income at the bottom of the company’s income statement and then divided this value by the average total assets to give us the ROA. A higher return on assets implies higher efficiency. Firms with higher ROA are better able to raise money in security markets because they offer attractive prospects for high returns (Bodie, Kane & Marcus, 2018) Net income was also utilized as its own separate variable as a measure to find out how profitable the firm was over the

fiscal year. Further, yearly return was denoted by taking the logarithmic change in value of a firm’s stock over the basis of a fiscal year. By regressing this value, we were able to see how a stock’s performance leads to changes in how the compensation package CEOs were awarded. Finally, the variable EPS was also determined. EPS is a measure of earnings per share, which serves as a measure of profitability as it divides the company’s profits over the number of shares outstanding.

Procedure:

The actual tests that were performed will be outlined in this section. To examine the relationship between compensation and performance, we utilized numerous different regressions to get a clearer understanding on how each individual component of compensation is influenced by the performance measures. We began by regressing one component of compensation onto a singular outcome variable (performance.) Then moving forth we systematically added other performance metrics into the regression equation to see how the addition of another variable impacted results and to see how much explanatory power it added (seen through the adjusted R-squared.) Following regressing all performance elements onto a singular compensation component, we continued to do the same operation onto the other compensation metrics, thus giving a more complete view on how all individual components of compensation are influenced by performance. Further, to examine the relationship between compensation and performance a correlation matrix was produced to see how two variables coincide/influence one another.

In addition to this, many regressions were ran with simply one dependent and one independent variable. The reason this was done was to avoid any potential problems with multicollinearity or correlations amongst variables that would invalidate the results of our analysis. Further, some regressions included multiple control variables, along with the

performance variable of interest. This was done to compare the outcomes of different DV’s when regressed on our IV, while controlling for the effects of other variables in order to more fully understand our model/relationships.

Company Data

Outlined below in table one, one can see the descriptive statistics regarding both our independent and dependent variables. Worthy of noting is the means of the salary, bonus, stock

options, current compensation, and total compensation. All these variables were reduced by a multiplier of 1000. The ‘average’ salary for all observations is equal to 1,514,412$ annually. Further, this means that the average total annual compensation a CEO received over this period was16,801,780$. In addition to these stated facts, it is of particular interesting to look at the min/max of our performance metrics. When examining yearly returns for example we see that our lowest return was -47.95% while the highest was 65.59%, this means that the difference between our most extreme values was 113.5%. We find a similar story when examining EPS, with the lowest value being -5.12 and the highest being 9.01(a 14.03 difference.) Like the findings above, we also can note that net income had a large amount of variability. Because of these huge discrepancies, the fiscal year served as control variable in further analysis. One may also take note of the fact that for the dummy variables USNOT and crisis the means represent the portion of the data set that received a value of one. This means that 83% of the firms were based in the US and only 10% of our historical data was during an economic recession.

Table 1: Descriptive Statistics

Linear regression

lnsalary Coef. St.Err. t-value p-value [95% Conf Interval] Sig

Actualrevenue 0.000 0.000 5.38 0.000 0.000 0.000 ***

Constant 7.171 0.030 239.01 0.000 7.111 7.230 ***

Mean dependent var 7.302 SD dependent var 0.210

R-squared 0.202 Number of obs 116.000

F-test 28.925 Prob > F 0.000

Akaike crit. (AIC) -58.374 Bayesian crit. (BIC) -55.621

*** p<0.01, ** p<0.05, * p<0.1

Descriptive Statistics

Variable Obs Mean Std.Dev. Min Max

lnsalary 116 7.302 .21 6.635 7.728

lnbonus 108 7.71 1.253 2.969 10.15

lncurrentc~p 116 8.282 .604 6.904 10.207

lntotalcomp 116 9.636 .441 8.484 10.578

lnstock 116 8.811 .632 6.717 10.235

Actualreve~e 116 3.02e+08 2.17e+08 3.58e+07 1.11e+09

ActualNetI~e 116 9.35e+07 4.33e+08 -2.07e+07 4.68e+09

TotalAvgAs~s 116 68499.11 95950.49 2423.2 914000

EPS 116 2.509 1.728 -5.12 9.01

YearlyReturn 116 .062 .196 -.479 .656

4) Results

In table 2, we see the output of the natural log of salary when regressed against the fiscal year. Here we see a P value of .002 which is significant at both the five and ten percent level. This implies that there is a relationship between the actual year and salary, which would lead to the conclusion that salary tends to increase over time, particularly given that the coefficient for fiscal year is indeed positive.

Taking this a step further, we then added an additional variable of revenue. In table 2, you will find that both the fiscal year and the revenue are significant. This implies that with a greater amount of revenue, the CEO is likely to receive a higher base pay. What is important to note, however, is that revenue and fiscal year likely have a positive relation as it is (given a correlation of .3493.) Since we know that salary and fiscal year have a positive relationship, there is a chance that revenue increases coinciding with salary increases are merely a consequence of the passage of time. Thus, to ensure that we captured the full extent of the effect, more variables were added in. Worthy of noting is the fact that each individual variable was added one by one, if the variable did not increase the adjusted R-squared or have a significant relation it will be added to the appendix but will not be discussed here.

Further, lnsalary was regressed onto EPS in table shown below. Here we can see that the P value is indeed significant signifying that the relationship was statistically significant. Similar to the situation above, EPS is also likely influenced by the fiscal year. For this reason, EPS and Fiscal year were correlated to find a value of .454 which may indicate that the salary increased along with EPS due to the power of time and not necessarily because of the actual increase in EPS.

Table 2: Regression (1) Results lnsalary Regression (1) (2) (3) (4) Fiscal year 0.000 0.005 0.006 (4.29)*** (2.83)*** (2.83)*** Actual revenue 0.000 0.003 (4.20)*** (3.06)*** EPS 0.012 0.828 (2.57)*** (.022)*** USnot 0.002 (3.23) R-squared 0.1388 0.2552 0.0546 0.3202 Adjusted R-squared 0.1312 0.2420 0.0463 0.2957 *p<0.1, ** p < 0.05, *** p < 0.01 N = 116 ()*** = T-value

Combining all the elements of performance, we then regressed lnsalary onto all the perfomance metrics and the table below (table 3) showcases the outputs from this regression. There are several things worth elaborating on in this table. Firstly, it is important to note that the adjusted R-squared was the highest given the inclusion of all the variables. This showcases the fact that by including all the performance metrics, we are able to get the most explanatory power. Thus, we can see that 36% of the variance we see can be explained by our model (based on simple R-squared.) Further, it is worthy to note which variables had a statistically significant relationship with lnsalary. These variables include USNOT, this served as a dummy variable that isolated the effect of whether a company was based in America or not. Since 0 indicated that it was centered in a foreign country, having a positive coefficient for this variable implies that salaries were higher in the US versus in a foreign country. Though this is useful for the sake of my research, it is important to note that only one firm was headquartered outside of the US, so no

concrete conclusions should be made based upon this result. Further, one can note that the variable Fiscal year was also significant with a P value of .026, making it statistically significant at 5%. Further, we can see a similar story play out with total average assets this also intuitively makes sense given that as the size of a firm increases, naturally so does the salary of the chief operating officer. What is intriguing to note is the fact that the inclusion of other variables can make other variables less significant. This is the case with EPS while in isolation it was

statistically significant, when you regress it along with other predictor variables it no longer is. This is likely due to multicollinearity or correlation amongst variables, particularly EPS and ROA are likely to measure the ‘same’ thing (as seen by their correlation.)

Table 3:

Moving on from salary, the next component of compensation explored was bonus. As in our last model, we first transformed the variable bonus into its logarithmic form lnbonus. This means that all interpretations of the results correspond to a percentage change in bonus and not actual monetary (incremental) changes in the bonus. To begin the analysis the first regression ran was lnbonus and fiscal year. The outputs of this regression are listed below in table 6. Here one can note that the P value is significant (.000) at both the five percent and one percent level. Once again this can lead one to assume that the relation between time and bonus is positive,

particularly given the positive coefficient of Fiscal Year.

Next, the variables revenue, net income and total average assets were added but the resulting regression was both insignificant for all these variables. In addition, the adjusted R-squared decreased meaning that it actually hurt the explanatory power of the model. Thus, we will move onto the variables that were found to have a significant relation. Through regressing each individual element with lnbonus one variable that was found to have a statistically

significant relationship was with EPS (P value = .000.) Logically, this makes sense as in many of the corporate reports detailing how bonus was set was through actual EPS. Consequently, it follows that with a higher EPS CEO’s will receive higher bonuses to compensate their performance. Further, one can note that lnbonus also had a significant relation with revenue (P=.047) as well as total average assets (P=.048.)

Table 4: Regression (1) Results lnbonus

Regression (1) (2) (3) Fiscal year 0.000 .0000 (6.15)*** (4.32) EPS 0.000 .025 (4.30)*** (2.28)*** Crisis 0.012 (2.57)*** USnot .194 (-1.31)*** R-squared 0.2630 0.1462 0.3488 Adjusted R-squared 0.2561 0.1406 0.3032 *p<0.1, ** p < 0.05, *** p < 0.01 N = 116 ()*** = T-value

Moving forth from regressing each individual performance outcome, we combined all the elements that contributed to a higher R-square. The reason we did not include all variables is by including IV that lower the adjusted R-square we are essentially adding ‘white noise’ to our model and errs further analysis. Posted below in table 8 is the output which maximized our adjusted R-squared. There are several things worthy of noting from this table. First, we see a few variables that stand out as statistically significant. First of which, is the variable Crisis, this dummy variable indicates that during the 2007/2008 recession, bonuses were one of the elements of compensation that was affected (to a statistically significant extent.) Further, we see that the variable EPS also retains its statistically significant P value (.025.) Finally, one can see that the

Fiscal year has a statistically significant impact on the bonuses (.000.) Worth noting here is the fact that ROA was not included given the inclusion of EPS, including both measures would likely lead to some issues as they essentially measure the same thing.

After running regressions on both salary and bonuses, we combined the two elements into a single variable denoted currentcomp. Current compensation was then transformed into its logarithmic form lncurrentcomp. It was from here that we began regressing our performance metrics onto this variable. Beginning with fiscal year and revenue we can see that both metrics were statistically significant at the five percent level (P=. 000 & .002 respectively.) This comes as no surprise as in both cases of lnsalary and lnbonus both these values were significant, so it comes as no shock that a combination of the two variables also yields significant results.

Table 5: Regression (1) Results lncurrentcomp

Regression (1) (2) (3) Fiscal year 0.000 .0000 (6.37)*** (4.82) EPS 0.000 .031 (4.62)*** (2.18)*** Crisis 0.673 (0.42)*** USnot .031 (2.19)*** R-squared 0.2625 0.1576 0.3267 Adjusted R-squared 0.2561 0.1502 0.3024 *p<0.1, ** p < 0.05, *** p < 0.01 N = 116

Following a similar methodology, performance variables were individually added to see what model most improved our adjusted R-squared. Posted below in the table showcases the model in which the adjusted R-squared was maximized. Immediately visible is the variables which have statistically significant results, namely, fiscal year (P=.002) and EPS (P=.024.) Once again this comes as no shock, as the same were true for both bonus and salary individually. What is further worth exploring is the values of USnot and Total Average assets. We can see that USnot has a P value of .031 which is significant at the 5 percent. This would imply that

American firms (within this sample) pay more in both bonuses and salary. Further, total average assets have a P value of .07 which is also very close to being significant at the 5% level. This follows the same line of reasoning for both salary and bonus; as the size of the firm increases the salary and bonuses are likely to increase as well, thus the combination of the two likely will as well.

Subsequently after running analysis on salary, bonus, and current compensation our analysis then delved into CEO equity options. Instead of treating stock options and grants as two different variables, they were combined into a single unit, stockoptions, to simplify analysis. Moreover, the variable stockoptions was then transformed into its logarithmic form to lnstockoptions before regressions were run. Similar to the scenarios above, we began our analysis by regressing lnstockoptions onto fiscal year and revenue and found that the results were statistically significant. This once again follows logically that over time compensation increases, as well as the fact that as firms grow so does compensation. (Riahi-Belkaoui & Pavlik, 1993)

Table 6: Regression (1) Results lnstock Regression (1) (2) (3) (4) Fiscal year 0.000 (-3.64)*** Actual revenue 0.0071 (-1.82)*** EPS 0.000 (-4.24)*** USnot 0.007 (2.76) R-squared 0.1042 0.0284 0.1361 0..0625 Adjusted R-squared 0.0963 0.0198 0.1285 0.0543 *p<0.1, ** p < 0.05, *** p < 0.01 N = 115 ()*** = T-value

Following the same procedure implemented on the prior variables, more predictive variables were added to maximize our adjusted R-squared. Here all performance metrics were included as they all increased the adjusted R-squared. Posted below is the output of the

regression. Here we see a similar tale as the variables above. One should note that the significant variables in this regression were EPS (.013) as well as USnot (.000.) Interpreting these results, one may conclude that American firms tend to pay out more in stock options or grants than foreign corporations (in this sample.) Further, one could propose that with increased Earnings per share CEO’s are granted more stock options or grants. However, given that the coefficient is negative this hurts the argument that positive performance drives higher levels of compensation for CEOs.

Table 7: Regression (1) Results lnstock Regression (1) (2) (3) (4) Fiscal year 0.000 0.0096 (-3.64)*** (-1.62)*** Actual revenue 0.0071 0.147 (-1.82)*** (1.46)*** EPS 0.000 0.002 (-4.24)*** (-3.18)*** USnot 0.002 (3.23) .4859 (-.69) Crisis R-squared 0.1042 0.0284 0.136 0.2660 Adjusted R-squared 0.0963 0.0198 0.0463 0.2327 *p<0.1, ** p < 0.05, *** p < 0.01 N = 116 ()*** = T-value

Finally, after conducting regressions on salary, bonus, current compensation and stock options a final variable total compensation encompassed all of these elements and represented the total monetary sum CEOs were paid annually that year. Beginning by regressing lntotalcomp against fiscal year, we immediately note a statistically significant relationship amongst the variables (P=.002.) In addition we see that EPS is significant with a P value of .071 (at 10% level.) Further, revenue was significant with a value of .000.

Table 8: Regression (1) Results lnstock Regression (1) (2) (3) Fiscal year 0.002 (3.2)*** Actual revenue 0.0000 (5.03)*** EPS 0.071 (1.82)*** R-squared 0.0823 0.1816 0.0283 Adjusted R-squared 0.0743 0.1744 0.019 *p<0.1, ** p < 0.05, *** p < 0.01 N = 116 ()*** = T-value Discussion Limitations:

One of the main limitations faced in this study is the lack of firms. Though the research

conducted does expand on the pre-existing knowledge surrounding performance and executive compensation, the relatively low numbers of firms in this study limits external validity. To establish better external validity more firms would have to be included to make conclusive remarks regarding the relationships between CEO compensation plans and firm performance.

Outliers were not present in our analysis as outliers were tested for and none were found to be present (at the 1.96 times Standard Error interval.) However, what is important to note about this study is that there were systematic shocks that affected the greater economy and hurt individual firm performance. For example, the 2007/2008 housing crisis which brought on a recession.

(Goodman and Mance, 2011.) Notably, that despite ruling out any outliers, the wages of those years were not drastically affected. So despite the fact that none of the CEOs were vastly overpaid that year (in comparison to the other years) their relatively stable salaries, bonuses and stock options hurt the ‘relationship’ between performance and compensation as a whole as it reflected the fact that compensation was awarded in ‘relatively’ large amounts despite the lack of positive firm performance. For this reason, it may be possible that the lack of statistically significant relationships can be partially attributed to the performance of these years. Consequently, we treated the fiscal year (for crisis) as a dummy variable, to control for the effect of the systematic shocks. Another potential solution to this might have been to implement local linear regression or the least absolute errors method, as opposed to using a global one like OLS. However, this was not done as even though it would have been more robust to outliers in specific regions as it fits a line locally rather than on the constant (Fan, 1993.) The reason it was not implemented was it may have potentially biased the results as an extreme value would either be removed or weighted less intensely, thereby artificially skewing our results. Thus OLS, despite its restrictions, was the method best fit to be implemented.

Another limitation encountered throughout this study was the large number of variables present.One might state that the larger the number of variables present in a study, the better able you are to make predictions regarding the model, but this is not always the case.. This tends to be true when you have a large number of independent variables, where adding more variables can actually have a detrimental effect to the predictive power of the model. OLS is particularly vulnerable to this, because adding more independent variables is only beneficial if those variables are good predictors of the output variable. Thus, one could contend that within this data set, adding in extra variables became redundant and hurt the algorithms ability to forecast accurate predictions. To minimize the ‘noise’ present in the regression, each variable was individually regressed onto the DV to see whether or not it added predictive power. Further, when creating a regression including multiple IV’s, each variable was added one by one to see how it impacted the adjusted R-squared. In the event it decreased the overall predictive power of the model, the IV would be excluded from the model. For future research that wishes to build upon this study, it would be recommended utilizing lasso linear regression which could automatically eliminate the effects of certain IVs.

Finally, we should consider the potential dependence amongst variables, particularly independent variables which are significantly related/correlated to one another. By including variables that correlate with one another in the same regression, you face potential issues with multicollinearity. An important implication this would have on our results is that individual coefficients would be predicted incorrectly by the model. The reason being is that if 2 different IV are essentially measuring the same thing then including both of them within the regression will ‘minimize’ the effect of a variable onto the DV, because part of the ‘effect’ would be attributed to the second variable it is correlated with. This was a concern of this study as it included a plethora of ‘performance’ variables that could be argued measure the same thing (for example EPS and ROA.) To solve for this a correlation matrix was produced to highlight the variables that tend to covary with one another. Further, separate regressions were ran independently of other variables to see what the effect of that IV was in isolation.

Results

Through regressing numerous elements of CEO compensation onto several performance metrics, this study aimed to investigate the relationship between executive compensation and firm performance.Throughout our analysis, there is evidence that higher levels of compensation derive higher firm performance, and vice versa.. Though the bulk of our analysis came back statistically insignificant, there were several metrics that were significantly related to CEO compensation that will be elaborated on here.

One of the first metrics that stands out as being significantly related to our compensation variables was earnings per share. EPS was found to have a significant relationship (at 5% level) with lnsalary, lnbonus, lnstockoptions, lncurrentcomp, and lntotalcomp (at the 10% level.) This was in line with Murphy (1985) who found that ‘executive compensation is strongly positively related to corporate performance as measured by shareholder return.’ Earnings per share is one of the many tools utilized to measure the profitability of a corporation, it is calculated by dividing net income over the average number of shares. Since we find a relationship between several compensation variables and EPS one could make the claim that a higher EPS drives higher levels of compensation. This was the strongest evidence found in our analysis.

Additionally, worth discussing is the variable USNOT. This dummy variable controlled for the effect that being in the US has on CEO compensation. Through our analysis we found that it had a statistically significant relationship with lnsalary, lnstock, lncurrentcomp, lntotalcomp. Taken at face value, this would entail that whether or not a pharmaceutical firm is based in the US would have an effect on the compensation of a CEO. In our case, being based in the US led to higher compensation overall, however no conclusion should be drawn from this, as our data lacked a sufficient number of foreign funds to establish sufficient external validity. Had external validity been met, this discovery would be in line with the past work of Core, Guay and Thomas (2009) who found that American CEOs are paid more than their European or Japanese counterparts. This leads some to believe that American CEO’s are overpaid in comparison to foreign corporations.

The fiscal year also had significant relationships with lnsalary, lncurrentcomp, lntotalcomp, and lnbonus. This finding supports the notion that salaries, bonuses as well as stock options/grants tend to increase over time (Cleementey & Cooley, 2009.) Also, research conducted by Frydman and Saks (2006) further shows that compensation has increased 6.8% annually from 1980 to 2003 for top executives. Further illuminating this reality was the work of Bebchuck and Grinstein (2005) who found that fortune 500 CEOs compensation packages jumped 146% in a span of 10 years. Cleementey & Cooley also found that median compensation has still increased even with years associated with poor performance. Consideration of these papers indicates that compensation increases as a function of time regardless of performance. Further worth examining is how compensation has evolved and changed over time. The work of Hall and Murphy (2003) illuminated how option grants became a more common means of compensation due to a revision in the tax code which allowed companies to deduct compensation awarded in excess of one million dollars, given that the compensation was performance based. For this reason, one can note that option/stock grants tended to increase over time regardless of firm performance, which is likely one of the reasons that most performance metrics did not have a significant relation with this variable. Additionally, given that the largest contributor to total compensation was typically stock options/grants we also found that not many performance metrics had a significant relation.

Even though many of our variables had no significant relationship, it is not to say that there is no relationship at all. The data collected and analyzed goes hand in hand with the work of Core, Guay & Thomas (2009) who could not make conclusive remarks as to whether CEO compensation

levels or growth are optimal as compensation rates have increased at the same rate regardless of circumstances surrounding the company. However, these authors contend that CEO compensation levels have grown because of two elements. The first element is investors force CEOs to bear greater equity risk (higher risk demands higher individual return for CEO.) Secondly, an increase in the size and complexity of a firm drives higher compensation levels. Additional research supports the notion that the larger the firm, the higher the level of compensation (Zhou, 2000.) Thus, one can see that compensation is granted in larger amounts largely due to the size of the firm. Following, it is logical to see that as firm size increases, so does the level of compensation. This was confirmed in our results by regressing totalaverageassets onto our compensation metrics. Performing this operation, we were able to see that average assets had a significant relationship with lnsalary, lnbonus, lncurrentcomp, lntotalcomp. Here we can note that with higher levels of average assets, we in turn see larger amounts of compensation confirming this notion within our own study.

Sales or revenue was also a figure worthy of examination. Within our analysis we found that revenue was one of the strongest predictors of compensation. The compensation metrics lnsalary, lnbonus, lncurrentcomp, and lntotalcomp all had statistically significant relations at the 5% level. Further, lnstock was significant at the 10% level (P=.071.) This finding is congruent with the work of Canyon, M.J (2014) who found that the ‘growth in compensation is positively and significantly related to a growth in firm sales.’

Conclusion: Despite the surplus of research done into executive compensation and its relationship with firm performance, there was a gap in knowledge as to how this relationship persists when it comes to pharmaceutical firms. Throughout our analysis, we aimed to expand our understanding of how CEO compensation drives higher performance for firms producing insulin. What was found throughout our thorough analysis of 6 core companies was that there does appear to be a positive relationship between firm performance and CEO compensation. This was confirmed through a multitude of OLS regressions, which found significant relationships between several compensation elements and ‘most’ performance variables. As denoted above, one can see that EPS, revenue, ROA, USnot and crisis all served as powerful predictors of

compensation levels. From this work, we conclude that there is some evidence to support the pay for performance hypothesis. Though our work was internally valid, I would recommend further research into the topic to see whether our findings are applicable to the pharmaceutical industry as a whole or if my findings only apply to those working within the diabetic sector.

Citations:

Abowd, J. M., & Kaplan, D. S. (1999). Executive compensation: six questions that need answering. Journal of Economic Perspectives, 13(4), 145-168.

Anderson, R. C., & Bizjak, J. M. (2003). An empirical examination of the role of the CEO and the compensation committee in structuring executive pay. Journal of Banking & Finance, 27(7), 1323-1348.

Ang, J. S., Nagel, G. L., & Yang, J. (2014). The effect of social pressures on CEO compensation. Available at SSRN 1107280.

Board of Governors of the Federal Reserve System. 2019. The Fed - Dealing With Unexpected Expenses. [online] Available at:

<https://www.federalreserve.gov/publications/2019-economic-well-being-of-us-households-in-2018-dealing-with-unexpected-expenses.htm> [Accessed 26 June 2020].

Brick, I. E., Palmon, O., & Wald, J. K. (2006). CEO compensation, director compensation, and firm performance: Evidence of cronyism?. Journal of Corporate Finance, 12(3), 403-423. Clementi, G. L., & Cooley, T. F. (2009). Executive compensation: facts (No. w15426). National Bureau of Economic Research.

Conyon, M. J. (2014). Executive compensation and board governance in US firms. The Economic Journal, 124(574), F60-F89.

Conyon, M., Gregg, P., & Machin, S. (1995). Taking care of business: Executive compensation in the United Kingdom. The Economic Journal, 105(430), 704-714.

CORE, J., GUAY, W., & THOMAS, R. (2009). Is U.S. CEO Compensation Broken? In Chew D. & Gillan S. (Eds.), U.S. Corporate Governance (pp. 144-158). NEW YORK: Columbia University Press. doi:10.7312/chew14856.10

Duffhues, P., & Kabir, R. (2008). Is the pay–performance relationship always positive?: Evidence from the Netherlands. Journal of multinational financial management, 18(1), 45-60. Fan, J. (1993). Local linear regression smoothers and their minimax efficiencies. The annals of Statistics, 196-216

Fialkoff, M. L. (2013). Pay-for-Delay Settlements in the Wake of Actavis. Mich. Telecomm. & Tech. L. Rev., 20, 523.

Gale, E. A., & Yudkin, J. S. (2011). Commentary: Politics of affordable insulin. Bmj, 343, d5675.

Garen, J. E. (1994). Executive compensation and principal-agent theory. Journal of political economy, 102(6), 1175-1199.

Gotham, D., Barber, M. J., & Hill, A. (2018). Production costs and potential prices for biosimilars of human insulin and insulin analogues. BMJ global health, 3(5).

Hall, B., and K. Murphy (2003): “The Trouble with Stock Options,” Journal of Economic Perspectives, 17(3), 49–70.

Hemphill, C. S., & Sampat, B. N. (2012). Evergreening, patent challenges, and effective market life in pharmaceuticals. Journal of health economics, 31(2), 327-339.

Hirsch, I. B. (2016). Insulin in America: a right or a privilege?.

Lakhtakia, R. (2013). The history of diabetes mellitus. Sultan Qaboos University Medical Journal, 13(3), 368.

Mellitus, D. (2005). Diagnosis and classification of diabetes mellitus. Diabetes care, 28(S37), S5-S10.

Murphy, K. J. (1985). Corporate performance and managerial remuneration: An empirical analysis. Journal of accounting and economics, 7(1-3), 11-42.

Murphy, K. J. (1999). Executive compensation. Handbook of labor economics, 3, 2485-2563. National Diabetes Data Group (US), National Institute of Diabetes, Digestive, & Kidney Diseases (US). (1995). Diabetes in America (No. 95). National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases.

Oxelheim, L., & Clarkson, K. (2015). Cronyism and the determinants of chairman compensation. Journal of business ethics, 131(1), 69-87.

Riahi‐Belkaoui, A., & Pavlik, E. (1993). Effects of ownership structure, firm performance, size and diversification strategy on CEO compensation: A path analysis. Managerial Finance. Rosenthal, E. (2019). When high prices mean needless death. JAMA internal medicine, 179(1), 114-115.

Simoni, R. D., Hill, R. L., & Vaughan, M. (2002). The discovery of insulin: the work of Frederick Banting and Charles Best. Journal of Biological Chemistry, 277(26), e15-e15.

Zhou, X. (2000). CEO Pay, Firm Size, and Corporate Performance: Evidence from Canada. The Canadian Journal of Economics / Revue Canadienne D'Economique, 33(1), 213-251. Retrieved June 20, 2020, from www.jstor.org/stable/2667376