The Medical Innovation Premium: Reassessing Its Existence And Stability In A Pharmaceutical Portfolio

1

The Medical Innovation Premium: Reassessing Its Existence

And Stability In A Pharmaceutical Portfolio

Author: Floris P.H. Groeneveld

1

_{– s2041782}

Date: 13/06/2016

Supervisor: Jochen O. Mierau

2

Abstract

Recent studies show healthcare portfolios generate higher alpha returns than contemporary pricing models explain. One study introduced this phenomenon as a medical innovation premium for

government-induced profit risk. The current paper evaluates the existence and stability of pharmaceutical excess alpha returns using a regulatory and non-regulatory reform period from 1990

to 2015 with more asset pricing models, such as the four- and five-factor model. This paper finds

evidence of no excess alpha returns, in contrast with the results of previous studies, using the five-factor model. While the five-five-factor model might not yet be a widely accepted benchmark model, if

estimates are representative of the future, then investment managers can generate excess alpha returns by holding the pharmaceutical portfolio.

JEL-codes: G12, G17 and G18

Keywords: Medical innovation premium, pharmaceuticals, asset pricing and alpha returns

1

_{Floris P.H. Groeneveld: MSc. Finance student at the Faculty of Economics and Business,}

University of Groningen, the Netherlands. The author can be contacted via e-mail on

floris_groeneveld@hotmail.com. I am highly indebted to Kenneth French for including his

data and to Jochen O. Mierau for helpful comments.

2

_{Jochen O. Mierau: Associate Professor, Department of Economics, Econometrics and}

2

1. Introduction

Asset pricing models may not fully explain the financial returns of healthcare companies. Koijen, Philipson and Uhlig (2016) (hereafter cited as Koijen et al.) find that investing in healthcare companies results in four to six percent more excess alpha returns than asset pricing models by Sharpe’s CAPM (1964) or Fama and French’s (1992) three-factor model would explain. In an attempt to explain this anomaly, Koijen et al. find healthcare firms display 25 percent more government-risk-related words in annual reports than non-healthcare firms. Koijen et al. thus find suggestive evidence that the excess alpha returns of healthcare firms between 1961 and 2014 serve as compensation for government-induced profit risk, and they introduce this phenomenon as the medical innovation

premium.

Investment managers’ objective is to create excess alpha returns for their investors, as widely used approach to benchmark against other investment funds. For-profit companies in healthcare experience tension between providing returns to financial markets and funding innovations or broadening access to drugs, especially in lower-income countries. Koijen et al. conclude that healthcare research and development (R&D) spending can double in the absence of the medical innovation premium. In addition, the medical innovation premium is especially relevant today, as low R&D productivity has led to public concern about a drought in medical innovation in the near future (Pamolli et al., 2011). Therefore, an evaluation of the medical innovation premium is not only relevant and relatively unknown in the financial literature, but could also provide more background on the tension between financial and real healthcare markets.

3 Second, the medical innovation premium and factor betas are estimated using one-year and three-year rolling regression windows. In this way, the current paper provides further background on the stability of the medical innovation premium and associative factor betas. A descriptive analysis will highlight periods of higher volatility, which are suggestively linked to booms and busts in the market and Clinton’s healthcare reform.

Third, the current paper will simulate an investment manager who will invest €1,000 euro in the pharmaceutical portfolio and is interested in whether the investment creates value to the client. A theoretical prediction of returns will be made from 2008 until 2015 using data present at the time of investing. Theoretical returns are predicted by rolling regression estimates and using their respective historical factor returns. The theoretical return are compared to actual return and show a profitable investment strategy, also when correcting for risk using the four- and five-factor model.

2. Literature Review

2.1 Asset pricing theory

The capital asset pricing model (CAPM) was proposed by Sharpe (1964) and can be found in equation (1). However, CAPM has been critiqued (see e.g. Roll (1977)). Despite its limitations, the CAPM model has remained a workhorse in finance until today.

𝑟𝑖 (𝑡) − 𝑟𝑓 (𝑡) = ∝ + 𝛽𝑖(𝑟𝑚 (𝑡) − 𝑟𝑓 (𝑡)) + 𝑒𝑖 (𝑡) (1)

Two additional risk factors are included in Fama and French’s (1992) three-factor model (equation 2). The risk factors include market beta, small minus big (SMB) firms and high minus low book-to-market ratio companies (HML). SMB represents the return of a zero investment portfolio that is long on small cap firms and short on large cap firms. HML is a zero investment portfolio that is long on firms with a high book-to-market ratio and short on firms with a low book-to-market ratio, representing a value premium or growth discounted portfolio.

𝑟𝑖 (𝑡) − 𝑟𝑓 (𝑡) = ∝ + 𝛽𝑖(𝑟𝑚 (𝑡) − 𝑟𝑓(𝑡)) + 𝑠𝑖 𝑆𝑀𝐵 (𝑡) + ℎ𝑖 𝐻𝑀𝐿 (𝑡) + 𝑒𝑖 (𝑡) (2)

Jagadeesh and Titman (1993) conclude that firms with a high return in the current year have higher risk-adjusted returns the next year, while this momentum effect is absent in a subsequent period. Carhart’s (1997) four-factor model (equation 3) is included to deal with the previously mentioned short-term momentum factor.

4 Fama and French (2015) introduce the five-factor model (equation 4), in which two factors extend the three-factor model (1992). First, a robust minus weak profitability factor (RMW) has been added, which is the difference between a low and high profitability stock portfolio. Second, there is a conservative minus aggressive investment factor (CMA), which is the difference between a low and high investment stock portfolio. An evaluation of the medical innovation premium with this newly proposed model is a welcome addition to the literature, considering the medical innovation premium has been gaining wider acceptance and the five-factor model still requires more development. Hou, Xue and Zhang (2015) show a critical evaluation of the five-factor model and discuss its main limitations.

𝑟𝑖 (𝑡) − 𝑟𝑓 (𝑡) = ∝ + 𝛽𝑖(𝑟𝑚 (𝑡) − 𝑟𝑓 (𝑡)) + 𝑠𝑖 𝑆𝑀𝐵 (𝑡) + ℎ𝑖

𝐻𝑀𝐿

(𝑡) + 𝑟𝑖 𝑅𝑀𝑊 (𝑡)

+ 𝑐𝑖 𝐶𝑀𝐴 (𝑡) + 𝑒𝑖 (𝑡) (4) Lastly, share repurchases are popular among high free-cash flow pharmaceuticals. Firms that repurchase stock over-perform compared to firms that do not (Ikenberry, Lakonishok and Vermaelen (1995)). Large pharmaceuticals are known to generously repurchase stock and may over-perform as a result. Therefore, the presence of a medical innovation premium could also be related to a portfolio which has a large allocation for stock-repurchasing pharmaceuticals. Fama and French (2016) show the five-factor model works well in explaining high average returns of share repurchasing firms, contrary to most other factor models.

2.2 Excess alpha returns and factor betas of healthcare firms

Tresl et al. (2014) find significant excess alpha returns for the period 1926-2009. They conclude that excess alpha returns have risen from 16 basis points per month from 1926-1985 to 22 basis points per month from 1985-2009. As a result, Tresl et al. advocate higher portfolio allocations for healthcare firms, while possible factors driving and sustaining excess alpha returns remain untouched. Koijen et al. (2015) document similar results and close the current gap by providing suggestive evidence of excess alpha returns to compensate investors for government-induced profit risk. Both papers use identical portfolios based on Standard Industry Classification (SIC) codes.

5 Previous studies have found an average market beta of healthcare firms approximately equal to one. Heterogeneity between firms and market betas can be explained by, among others, size (-) and the firm being pharmaceutical (-) or biotechnical (+). From the 1970s until the mid-1980s, estimates of market beta have been close to one using the CAPM framework (Grabowski & Vernon (1990), DHG (1991), Myers and Shyam-Sunder (1996)). Until 1993, greater heterogeneity in study samples could have led to more market beta variety in the studies performed (Myers and Shyam-Sunder (1996), Myers and Howe (1997), DHG (2003)). Using Fama and French’s three-factor model, Vernon, Golec and DiMasi find increasing SMB betas for pharmaceuticals of 0.66 and 0.99 ending in 1980 and 1990 respectively. Between 1982 and 2005, Golec and Vernon (2007) find a market beta for pharmaceuticals of 0.92, size beta of 0.8 and a low HML beta of 0.02.

Giaccotto, Golec and Vernon (2011) (hereafter cited as Giaccotto et al.) document a CAPM market beta of 0.69 for value-weighted pharmaceutical portfolios between 1950 and 2004. Giaccotto et al. suggest a change in the structure of the equity markets to explain a lower market beta than previously found. As the study sample of Giaccotto et al. consists of larger firms, an increase in initial public offerings by more small firms could have led to an increase in idiosyncratic risk and resulted in a lower relative riskiness of large firms in the 1980s and 1990s (Fink et al. 2005). Unfortunately, Giaccotto et al. do not present results on possible excess alpha returns; their study could have been a prime paper to evaluate a possible different hypothesis on drivers of excess alpha returns by healthcare firms. The sample of Giaccotto et al. consists primarily of larger high free-cash flow pharmaceuticals, which could have shed more light on whether the medical innovation premium is (partly) related to the share repurchasing premium. Lastly, Harrington (2009) found a 0.69 market beta for pharmaceuticals for the more recent 2001-2005 period using the CAPM model. From 2006-2008, the market beta of pharmaceuticals was lower, with a value of 0.61.

6

3.0 Methods

3.1 Study sample

The current paper uses Kenneth French’s industry portfolios to further evaluate the existence and stability of the medical innovation premium. Industry portfolios are value-weighted and based on the Standard Industry Classification (SIC) of firms, revised every year. All firms are listed on the NYSE, AMEX or NASDAQ stock exchange and have available return data on the Center for Research on Security Prices (CRSP). All portfolios are estimated from 1990-2015 and 1995-2015. The former period is used to provide representative estimates on contemporary risk profiles of portfolios and the latter is period used to analyse the medical innovation premium after Clinton’s healthcare reforms.

Descriptive statistics of pharmaceutical, medical equipment and health services portfolios can be found in table 1. The pharmaceutical and health service portfolio generate similar returns and are significantly positive. The return of the medical equipment portfolio was slightly lower compared to the pharmaceutical portfolio and significantly different from zero. The number of firms and the average value of firms were highest for the pharmaceutical portfolio, while health services was the smallest portfolio in both these dimensions, yet with high standard errors.

7

TABLE 1

DESCRIPTIVE STATISTICS

PORTFOLIOS1 _{(1) (2) (3)}

Variables Pharmaceuticals Medical equipment Health services

Annualized return 13.1% 12.8% 10.1%

Number of firms 278 (54) 164 (37) 93 (35)

Value of firm (in million USD) 3.271 (1.394) 1.238 (861) 962 (644) Total market value portfolio 956,835 (492,672) 178,320 (98,701) 72,064 (30,544) Book-to-market ratio 0.228 (0.088) 0.312 (0.095) 0.445 (0.091)

Market portfolio return 7.25%

SMB 2.13% HML 2.14% MOM 7.16% RMW 4.23% CMA 2.85% Risk-free rate 2.91%

1 _{Standard errors are in parentheses. Bold returns are significantly different from zero (p<0.05).}

3.2 Explanatory variables

A standard and accepted proxy for the market portfolio is the value-weighted portfolio of all stocks tradable on the NYSE, AMEX and NASDAQ stock exchanges. The risk-free rate is the one-month United States T-bills rate from Ibbotson Associates. For the three-factor model, the return on the small minus big (SMB) portfolio is a between and within weighted portfolio of small value, small neutral and small growth portfolios minus the respective big portfolios (small-neutral-growth). The return on the high minus low fundamentals (HML) portfolio is a between and within weighted portfolio of two small value and big value portfolios minus two small growth and big growth portfolios. Momentum (MOM) is the average of two big and small high return portfolios minus two big and small low return portfolios. The momentum portfolio is defined as the 30% highest and lowest returns of companies listed on the NYSE and determined on a monthly basis. For the five-factor models, the within and between averaged returns were marginally different due to the weighting of a greater number of factors. RMW factor is a portfolio with high operating profitability minus low operating profitability firms. CMA factor is a portfolio with low growth of total assets minus high growth of total assets. Correlations among explanatory variables are low. The correlation between HML and CMA factor is 0.67.

3.3 Methodology

8 pharmaceutical portfolio. In this section, the main outcome variable is potential significant alpha returns to provide a broader estimation of the previously documented medical innovation premium using more asset pricing models than before in identical portfolios.

Second, rolling regressions are estimated with a one- and three-year window and reported into figures of alpha and market, SMB, HML, MOM, RMW and CMA betas. Rolling regression window sizes vary widely in the literature. Therefore, it is important to include and compare multiple windows. The aim of this section is to describe the volatility of the rolling alphas and betas. The results are connected with booms and busts in the market and regulatory reforms.

Third, three-year rolling Carhart’s and five-factor regression model estimates are used to predict returns from 2008 to 2015. A variance analysis is provided for the difference between theoretical returns and actual returns. Estimates of alpha and betas are calculated using data from 2005 to 2008 and are multiplied by historical average returns on factor beta portfolios. This procedure is performed on a rolling basis every month. Some assumptions are required about factor portfolios’ returns. The current paper uses an average of factor beta portfolio returns between 1990 and 2008 as representative returns for the simulation. Market portfolio returns were 4.5 % and the SMB portfolio yielded 2.4 % on a yearly basis. The HML and MOM portfolios generate a return of 0.90 % and 11.6 % respectively. Current estimates are similar when annualizing Carhart’s (1997) estimates between 1963 and 1993, except for the HML portfolio. For the five-factor model, RMW yielded 4.6 % and CMA 3.6 % annually.

4.0 Results

4.1 CAPM’s, Fama-French’s and Carhart’s estimates, 1990-2015

9 Second, market portfolio betas remain relatively constant in all models employed. A higher number of model factors relates to a higher estimate of the market beta, except for medical equipment, for which market beta is only marginally different in all models. For the pharmaceutical portfolio, the market beta is stable at around 0.70, and according to the five-factor model is higher at 0.81. The health services portfolio shows a similar pattern as well, though differences across the models employed were larger compared to the pharmaceutical and medical equipment portfolios.

All portfolios have significant exposure to the SMB and/or HML portfolios. The pharmaceutical portfolio exhibits negative coefficient estimates of SMB and HML in all models. Therefore, the pharmaceutical portfolio is tilted towards large cap discounted firms and low fundamentals of growth firms, while its coefficient size is high for SMB and low for HML. Coefficient sizes of SMB and HML become conversely low and high in the five-factor model, yet remain negative. The medical equipment portfolio has significant positive exposure to the SMB portfolio, yet it is economically low in size. The medical equipment portfolio does not feature a significantly different from zero HML beta. Therefore, this portfolio is only tilted towards small cap premium stocks. The health services portfolio has a relatively large positive and statistically significant HML beta in all models except for the five-factor model. Therefore, the health services portfolio is tilted to value premium stocks, while the five-factor model only suggests positive exposure to the small cap premium.

According to Carhart’s model, no significant exposure to the momentum portfolio is documented. The pharmaceutical portfolio has significant positive exposure to the RMW and CMA portfolios and is therefore tilted towards robust profitability and conservative investing firms. The medical equipment portfolio is related to robust profitability significant at the ten percent level, while the health services profitability beta is significant at one percent and economically large. Therefore, it appears the health services portfolio moves one-in-one with the robust minus weak profitability portfolio.

R2 _{for the pharmaceutical portfolio is between 0.40 and 0.50 for CAPM and the five-factor}

model, respectively. R2 _{for the medical equipment portfolio is lower, between 0.27 and 0.54 for the}

CAPM and five-factor model, respectively. R2 _{for health services is between 0.27 and 0.43 for the CAPM}

and five-factor model, respectively. As a result, the five-factor model explains approximately one half of proportional changes in returns and performs better than other models according to R2_{. In the next}

10

TABLE 2

Least Square estimates (1990-2015) 1 CAPM model

Variables Pharmaceuticals Medical equipment Health services

Alpha 5.25** 3.92 1.51

Market portfolio 0.68*** 0.82*** 0.78***

Fama-French’s three-factor model

Variables Pharmaceuticals Medical equipment Health services

Alpha 6.32*** 3.87 -0.52

Market portfolio 0.70*** 0.80*** 0.83***

SMB -0.37*** 0.12* 0.25

HML -0.25** -0.01 0.55***

Carhart’s four-factor model

Variables Pharmaceuticals Medical equipment Health services

Alpha 5.50*** 3.32 -1.54

Market portfolio 0.73*** 0.82*** 0.87***

SMB -0.38*** 0.11*** 0.24

HML -0.22** 0.01 0.59***

MOM 0.08 0.05 0.10

Fama-French’s five-factor model

Variables Pharmaceuticals Medical equipment Health services

Alpha 3.26 2.16 -6.35* Market portfolio 0.81*** 0.84*** 0.98*** SMB -0.31*** 0.22*** 0.62*** HML -0.50*** -0.14 0.21 RMW 0.27** 0.23* 0.98*** CMA 0.57*** 0.15 0.17

1_{Alpha estimate is annualized. * Denotes significance at 10%, ** denotes significance at 5%, *** denotes significance at 1%. A value weighted portfolio of}

11

4.2 CAPM’s, Fama-French’s and Carhart’s estimates, 1995-2015

For the pharmaceutical portfolio, excess alpha returns are higher in all models employed for 1995-2015 estimates compared to 1990-2015 estimates. The three-factor model shows an almost seven percent excess alpha return, which is economically significant. The five-factor model suggests an almost five percent excess alpha return, yet is not significant at five percent. For the medical equipment portfolio, only CAPM’s alpha is different compared to 1990-2015, yet the alpha is only significant at the ten percent level. For the health services portfolio, the negative excess alpha return of ten percent significance from 1990-2015 becomes not different from zero in the period 1995-2015.

For all portfolios, models with more factors show higher market betas; difference between all models is only marginal (0.12) for the pharmaceutical portfolio and is higher (0.22) for the health services portfolio. For 1995-2015, market betas are lower compared to estimates from 1990-2015. In general, SMB beta is relatively constant when comparing among size and statistical significance. The SMB beta of the medical equipment portfolio is different for the five-factor model in both periods. In general, HML betas are similar according to different asset pricing models, except for the pharmaceutical portfolio estimates when comparing the three- and five-factor model. Health services portfolio had the largest book-to-market ratio compared to the other portfolios. Therefore, health services is expected to have the largest HML beta. Health service portfolio does not have a significant HML beta from 1990 to 2015, indicating no exposure to the value premium or growth discount. From 1995-2015, health service portfolio does have a significant exposure to the value premium using the five-factor model, yet its size is relatively low compared to other estimations of the period.

For the period 1995-2015, no significant momentum is documented in any of the portfolios which is consistent with results of period 1990-2015. Pharmaceutical’s exposure to the RMW beta is not significantly different from zero in 1995-2015, while RMW beta was significant during 1990 to 2015. Pharmaceutical’s exposure to investment factor remains identical. Compared to 1990-2015, the RMW beta of the medical equipment portfolio does not deviate from zero in 1995-2015. The health services profitability beta remains constant and is economically high, an almost one-in-one relation to the profitability portfolio.

12

TABLE 3

Least Square estimates (1995-2015)2 _{CAPM model}

Variables Pharmaceuticals Medical equipment Health services

Alpha 6.30** 4.60* 1.53

Market portfolio 0.62*** 0.75*** 0.68**

Fama-French’s three-factor model

Variables Pharmaceuticals Medical equipment Health services

Alpha 6.92*** 3.97 -1.16

Market portfolio 0.66*** 0.75*** 0.75***

SMB -0.34*** 0.17** 0.29

HML -0.13 0.15 0.76***

Carhart’s four-factor model

Variables Pharmaceuticals Medical equipment Health services

Alpha 6.03*** 3.43 -2.05

Market portfolio 0.70*** 0.77*** 0.79***

SMB -0.36*** 0.16** 0.27

HML -0.10 0.17 0.80***

MOM 0.09 0.05 0.09

Fama-French’s five-factor model

Variables Pharmaceuticals Medical equipment Health services

Alpha 4.68* 3.62 -6.01 Market portfolio 0.74*** 0.76*** 0.90*** SMB -0.37*** 0.17 0.59*** HML -0.34*** 0.07 0.43** RMW 0.09 0.00 0.79*** CMA 0.58*** 0.13 0.11

2 _{Alpha estimate is annualized. * Denotes significance at 10%, ** denotes significance at 5%, *** Denotes significance at 1%. A value weighted portfolio of}

13

4.3 Rolling regressions, one- and three-year windows

The previous section suggests that different asset pricing models result in different estimates and statistical significances, yet it is possible the differences could be of only marginal economic importance. The current and subsequent sections will evaluate possible differences in rolling regression windowed estimations and the time-variability of alphas and betas, respectively. From now on, this paper will focus primarily on the pharmaceutical portfolio, although the results for the medical equipment and health services portfolios can be obtained by contacting the author.

The one-year rolling excess alpha returns of CAPM are significantly and sizeable different compared to excess alpha returns using the three-factor model (p=0.01). The difference with Carhart’s model is also significant (p=0.02). Five-factor rolling alpha is not statistically different from all other models employed. Market beta from the five-factor model is statistically and economically different than CAPM (p<0.01) and three-factor market beta (p=0.03), though not different than Carhart’s beta (p=0.10). The momentum factor is significant at ten percent, yet is not meaningful in terms of size.

TABLE 4

Rolling estimates

pharmaceuticals1

One-year rolling

estimates CAPM F-F three-factor Carhart F-F five-factor

Alpha 3.45*** 6.08*** 6.01*** 4.89*** Beta market 1.20*** 0.74*** 0.76*** 0.83*** SMB -0.42*** -0.34*** -0.49*** HML -0.47*** -0.38*** -0.72*** MOM 0.05* RWA -0.31*** CMA 0.46***

1_{Alpha estimate is annualized. * Denotes significant deviation from zero at 10%, ** denotes}

significance at 5%, *** denotes significance at 1%.

14

TABLE 5

Rolling estimates

pharmaceuticals1

Three-year rolling

estimates CAPM F-F three-factor Carhart F-F five-factor

Alpha 2.89*** 5.78*** 5.47*** 5.10*** Beta market 0.71*** 0.73*** 0.75*** 0.77*** SMB -0.40*** -0.38*** -0.45*** HML -0.38*** -0.33*** -0.57*** MOM 0.02* RMW -0.19*** CMA 0.47***

1_{Alpha estimate is annualized. * Denotes significant deviation from zero at 10%, **}

denotes significance at 5%, *** denotes significance at 1%.

4.4 Time-variability of rolling regression factor betas for the

pharmaceutical portfolio

To investigate the time-variability of previous estimates, rolling alphas and betas are evaluated in this section. First, rolling alphas can be found in figures 1 and 2 between 1990m1 and 2015m12. The vertical axis shows annualized alpha returns. The horizontal axis reflects the start of the window. In figure 1, 1990m1 shows the one-year rolling alpha estimated using the period 1990m1 until 1990m12.

15 Figure 1: Rolling alphas of the pharmaceutical portfolio with one-year rolling window

Rolling three-year windowed alphas show lower values and less volatility than one-year windowed alphas. In general, the alphas are positive. Alphas marginally increase from 1990m1 until 1995m10, decrease until 2005m1 and rise thereafter, except for CAPM’s alpha. One major similarity between the CAPM three-year and CAPM one-year rolling alphas is that they show underperformance between 1990m1 and 1992m10. The CAPM alphas also suggest an underperformance of the portfolio during the early 2000s until 2005, which is a prolonged period of the technology boom and bust. Furthermore, three-year alphas remain relatively similar across different models; only the five-factor model shows some deviation around the technology boom period of 1997-2001. According to all asset pricing models, the financial crisis of 2008-2009 exhibites positive three-year rolling alpha returns, while the average Sharpe ratio indicate few additional return compensation in volatile periods. In conclusion, all three-year rolling alphas except CAPM alpha suggest that the pharmaceutical portfolio performs reasonably well during Clinton’s healthcare reform, the technology boom and bust and the financial crisis. -4 -2 0 2 4 6 8 1 9 9 0 m 1 1 9 9 0 m 1 2 1 9 9 1 m 1 1 1 9 9 2 m 1 0 1993m9 19 9 4 m 8 1 9 9 5 m 7 1 9 9 6 m 6 1 9 9 7 m 5 1 9 9 8 m 4 1 9 9 9 m 3 2 0 0 0 m 2 2 0 0 1 m 1 2 0 0 1 m 1 2 2 0 0 2 m 1 1 2 0 0 3 m 1 0 2 0 0 4 m 9 2 0 0 5 m 8 2 0 0 6 m 7 2 0 0 7 m 6 2 0 0 8 m 5 2 0 0 9 m 4 2 0 1 0 m 3 2 0 1 1 m 2 2 0 1 2 m 1 2 0 1 2 m 1 2 2 0 1 3 m 1 1 2 0 1 4 m 1 0

Rolling alpha (one-year window)

16 Figure 2: Rolling alphas of the pharmaceutical portfolio with three-year rolling window

The rolling betas of the market portfolio can be found in figures 3 and 4. One-year rolling market betas are more volatile than three-year market betas. One-year variability is high and wanders between almost three and minus two. Market beta estimates are economically significant, as beta is a multiplier of market portfolio returns; the average return is substantial at 7.3 percent annually. The pharmaceutical portfolio’s relation to the market was high in 1993m9, 2004m1 and 2012m1 and low or negative in 1992m10, 1990m3-2000m2 2007m6 and 2012m1. The pharmaceutical portfolio is not related to the market portfolio in 1992m10 across all models and was negatively related during the 2000s, which is economically significant as the market portfolio declines during that period. As a result, alphas can be underestimated due to a negative exposure to an overall declining market in the 2000s. Rolling market betas are more similar in the multiple asset pricing models but show relatively high volatility. -1.5 -1 -0.5 0 0.5 1 1.5 2 1 9 9 0 m 1 1 9 9 0 m 1 1 1 9 9 1 m 9 1 9 9 2 m 7 1 9 9 3 m 5 1 9 9 4 m 3 1 9 9 5 m 1 1 9 9 5 m 1 1 1 9 9 6 m 9 1 9 9 7 m 7 1 9 9 8 m 5 1 9 9 9 m 3 2 0 0 0 m 1 2 0 0 0 m 1 1 2 0 0 1 m 9 2 0 0 2 m 7 2 0 0 3 m 5 2 0 0 4 m 3 2 0 0 5 m 1 2 0 0 5 m 1 1 2 0 0 6 m 9 2 0 0 7 m 7 2 0 0 8 m 5 2 0 0 9 m 3 2 0 1 0 m 1 2 0 1 0 m 1 1 2 0 1 1 m 9 2 0 1 2 m 7

Rolling Alpha (three-year window)

17 Figure 3: Rolling market betas of the pharmaceutical portfolio with one-year rolling window

From 1990m1, the three-year window market beta has a clear downwards trend from about 1.1 to 0.1, while the slope of decline increases up to 1998m8. From 1998m8 until the end of sample, it increases back to a one-in-one relation to the market portfolio post-technology crash. Similar market betas are reported among all models, except for CAPM during the first and post-technology boom period. Asset pricing models feature more discrepencies among three-year rolling betas; CAPM’s market beta especially deviates in some periods (1990m1-1991m9 and 2002m7-2005m1).

Figure 4: Rolling market betas of the pharmaceutical portfolio with three-year rolling window

-3 -2 -1 0 1 2 3 4 1 9 9 0 m 1 1 9 9 0 m 1 2 1 9 9 1 m 1 1 1 9 9 2 m 1 0 1 9 9 3 m 9 1 9 9 4 m 8 1 9 9 5 m 7 1 9 9 6 m 6 1 9 9 7 m 5 1 9 9 8 m 4 1999m3 20 0 0 m 2 2 0 0 1 m 1 2 0 0 1 m 1 2 2 0 0 2 m 1 1 2 0 0 3 m 1 0 2 0 0 4 m 9 2 0 0 5 m 8 2 0 0 6 m 7 2007m6 20 0 8 m 5 2 0 0 9 m 4 2 0 1 0 m 3 2 0 1 1 m 2 2 0 1 2 m 1 2 0 1 2 m 1 2 2 0 1 3 m 1 1 2 0 1 4 m 1 0

Rolling market beta (one-year window)

CAPM Beta FF3 Beta Carhart beta FF5 Beta

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1 9 9 0 m 1 1 9 9 0 m 1 1 1 9 9 1 m 9 1 9 9 2 m 7 1 9 9 3 m 5 1994m3 19 9 5 m 1 1 9 9 5 m 1 1 1 9 9 6 m 9 1 9 9 7 m 7 1 9 9 8 m 5 1 9 9 9 m 3 2 0 0 0 m 1 2 0 0 0 m 1 1 2 0 0 1 m 9 2 0 0 2 m 7 2 0 0 3 m 5 2 0 0 4 m 3 2 0 0 5 m 1 2 0 0 5 m 1 1 2 0 0 6 m 9 2 0 0 7 m 7 2 0 0 8 m 5 2 0 0 9 m 3 2 0 1 0 m 1 2 0 1 0 m 1 1 2 0 1 1 m 9 2 0 1 2 m 7

Rolling market beta (three-year window)

18

4.5 Five-factor rolling regression of SMB, HML, RMW and CMA betas

for the pharmaceutical portfolio

To further investigate the stability of the factors SMB, HML, RMW and CMA, this paper provides figures with three-year rolling factor betas and 95% confidence intervals for the five-factor model. Three-year rolling factors are chosen for more representative estimates, especially considering the subsequent section on portfolio return simulations. The five-factor model is a natural choice in order to better evaluate the performance of this newly proposed model.

From table 5, the three-year rolling SMB beta estimate is significantly negative. Figure 5 shows that the period 2000m11 - 2005m1 is driving the statistical significance of the rolling three-year SMB beta. A negative SMB beta suggests the pharmaceutical portfolio is exposed to the large cap discount. The three-year rolling SMB estimates are relatively stable at approximately minus one half, and 95% CI is slightly wider between 2005m5 and 2008m5. The three-year rolling SMB is closer to zero over the past ten years, meaning that fewer large cap discounted firms are allocated in the portfolio or that there is greater allocations for small cap premium firms. Yet still, the SMB portfolio yields 2.13% annually, which is relatively low. Therefore, the figure suggests that three-year rolling SMB returns could be of limited economic significance compared to alphas and market betas in explaining the portfolio’s returns.

Figure 5: Rolling SMB beta pharmaceutical portfolio with three-year rolling window

The three-year rolling HML beta is significantly negative due to the period from the start until 1990m11 and after 2010m11. As a result, three-year rolling estimates indicate the pharmaceutical portfolio is tilted towards growth stocks during these periods. In all other periods, HML beta is not significantly different from zero. HML beta features more volatility than SMB beta, from minus one to

-2 -1.5 -1 -0.5 0 0.5 1 1 9 9 0 m 1 1 9 9 0 m 1 1 1 9 9 1 m 9 1 9 9 2 m 7 1 9 9 3 m 5 1 9 9 4 m 3 1 9 9 5 m 1 1 9 9 5 m 1 1 1 9 9 6 m 9 1 9 9 7 m 7 1 9 9 8 m 5 1 9 9 9 m 3 2 0 0 0 m 1 2 0 0 0 m 1 1 2 0 0 1 m 9 2 0 0 2 m 7 2 0 0 3 m 5 2 0 0 4 m 3 2 0 0 5 m 1 2 0 0 5 m 1 1 2 0 0 6 m 9 2 0 0 7 m 7 2 0 0 8 m 5 2 0 0 9 m 3 2 0 1 0 m 1 2 0 1 0 m 1 1 2 0 1 1 m 9 2 0 1 2 m 7

SMB Beta (three-year window)

19 about zero from 2000m11 to 2010m11. Results could have been significant, as HML beta is a multiplier of HML portfolio returns. However, HML returns are relatively low, with 2.1% annually (same as the SMB portfolio).

Figure 6: Rolling HML beta pharmaceutical portfolio with three-year rolling window

The three-year rolling RMW beta is significantly negative due to the short period after 2011m1. In all other periods, RMW rolling beta is not significantly different from zero. As a result, the pharmaceutical portfolio is tilted towards robust profitably firms after 2011m1. The three-year rolling RMW beta figure shows relatively low volatility and the RMW portfolio yields one of the highest factor portfolio returns at 4.2% annually. Therefore, a change from negative one to smaller negative beta in more recent times is still a significant finding.

Figure 7: Rolling RMW beta pharmaceutical portfolio with three-year rolling window

-3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1 9 9 0 m 1 1 9 9 0 m 1 1 1 9 9 1 m 9 1 9 9 2 m 7 1 9 9 3 m 5 1 9 9 4 m 3 1 9 9 5 m 1 1 9 9 5 m 1 1 1 9 9 6 m 9 1 9 9 7 m 7 1 9 9 8 m 5 1 9 9 9 m 3 2 0 0 0 m 1 2 0 0 0 m 1 1 2 0 0 1 m 9 2 0 0 2 m 7 2 0 0 3 m 5 2 0 0 4 m 3 2 0 0 5 m 1 2 0 0 5 m 1 1 2 0 0 6 m 9 2 0 0 7 m 7 2 0 0 8 m 5 2 0 0 9 m 3 2 0 1 0 m 1 2 0 1 0 m 1 1 2 0 1 1 m 9 2012m7

HML Beta (three-year window)

HML Beta Upper HML Lower HML

-3 -2 -1 0 1 2 3 1 9 9 0 m 1 1 9 9 0 m 1 1 1 9 9 1 m 9 1 9 9 2 m 7 1 9 9 3 m 5 1 9 9 4 m 3 1 9 9 5 m 1 1 9 9 5 m 1 1 1 9 9 6 m 9 1 9 9 7 m 7 1 9 9 8 m 5 1 9 9 9 m 3 2 0 0 0 m 1 2 0 0 0 m 1 1 2 0 0 1 m 9 2 0 0 2 m 7 2 0 0 3 m 5 2 0 0 4 m 3 2005m1 2 0 0 5 m 1 1 2 0 0 6 m 9 2 0 0 7 m 7 2 0 0 8 m 5 2 0 0 9 m 3 2 0 1 0 m 1 2 0 1 0 m 1 1 2 0 1 1 m 9 2 0 1 2 m 7

RMW Beta (three-year window)

20 The three-year rolling CMA beta is significantly positive due to the short period after 2010m11. In all other periods, CMA rolling beta is not significantly different from zero. As a result, the pharmaceutical portfolio is tilted towards conservative investment firms after 2010m11. The three-year rolling CMA beta features relatively high volatility compared to other non-market betas, yet the CMA portfolio yields 2.9% on average. The 95 % CI is narrower in more recent times. Therefore, a change from negative CMA beta in the first three years of the sample to a beta equal to one in more recent times is a significant finding. The 95 % CI is relatively more narrow compared to the first decade of sample period.

Figure 8: Rolling RMW beta pharmaceutical portfolio with three-year rolling window

4.6 Out-of-sample forecast

In the previous sections, this paper shows Carhart’s estimates and evaluated the betas of Fama and French’s five-factor model for the total sample period 1990m1-2015m12. Both models exhibit explanatory power for approximately half of the variation in returns. The current paper will now implement both models to forecast returns after 2008, using three-year rolling regressions obtained from 2005.

Carhart’s simulation can be found in figure 7. Carhart’s model simulation performs well until 2012m5 with an economically small difference in 2009m3. After 2012m5, the pharmaceutical portfolio performs better than Carhart’s model’s prediction. The performance of Fama-French’s five-factor model, shown in figure 10, is similar to Carhart’s model. However, Carhart’s model predicts a €2,134 portfolio value and the five-factor model predicts a €1,853 portfolio value at the end of 2015. The actual portfolio value is €2,982, so the pharmaceutical portfolio performs better than both asset pricing models can explain, even when including three-year rolling alphas. Between 2008 and 2015,

-3 -2 -1 0 1 2 3 1 9 9 0 m 1 1 9 9 0 m 1 1 1 9 9 1 m 9 1 9 9 2 m 7 1 9 9 3 m 5 1 9 9 4 m 3 1 9 9 5 m 1 1 9 9 5 m 1 1 1 9 9 6 m 9 1 9 9 7 m 7 1 9 9 8 m 5 1 9 9 9 m 3 2 0 0 0 m 1 2 0 0 0 m 1 1 2 0 0 1 m 9 2 0 0 2 m 7 2 0 0 3 m 5 2 0 0 4 m 3 2 0 0 5 m 1 2 0 0 5 m 1 1 2006m9 20 0 7 m 7 2 0 0 8 m 5 2 0 0 9 m 3 2 0 1 0 m 1 2 0 1 0 m 1 1 2 0 1 1 m 9 2 0 1 2 m 7

CMA Beta (three-year window)

21 Carhart’s three-year rolling regressions predict 57.2 % of actual returns, while the five-factor model predicts only 43.0 % of actual returns.

Figure 9: Theoretical vs actual portfolio returns using Carhart’s three-year rolling window estimates updated on a monthly basis

Figure 10: Theoretical vs actual portfolio returns using Fama-French’s five-factor one-year rolling window estimates updated on a monthly basis

5. Discussion

This paper documents similar excess alpha returns in the pharmaceutical portfolio between 1990-2015 using the CAPM and three-factor models. Koijen et al. find 4-6 % excess alpha return using the CAPM and three-factor models and data from 1961 to 2012. Using these same models, this paper

0 500 1000 1500 2000 2500 3000 3500 2 0 0 8 m 1 2 0 0 8 m 5 2 0 0 8 m 9 2 0 0 9 m 1 2 0 0 9 m 5 2 0 0 9 m 9 2 0 1 0 m 1 2 0 1 0 m 5 2 0 1 0 m 9 2 0 1 1 m 1 2 0 1 1 m 5 2011m9 201 2 m 1 2 0 1 2 m 5 2 0 1 2 m 9 2 0 1 3 m 1 2 0 1 3 m 5 2 0 1 3 m 9 2 0 1 4 m 1 2 0 1 4 m 5 2 0 1 4 m 9 2 0 1 5 m 1 2 0 1 5 m 5 2 0 1 5 m 9

Carhart's simulation portfolio value (€)

Theoretical value Actual value

0 500 1000 1500 2000 2500 3000 3500 2 0 0 8 m 1 2 0 0 8 m 5 2 0 0 8 m 9 2 0 0 9 m 1 2 0 0 9 m 5 2 0 0 9 m 9 2 0 1 0 m 1 2 0 1 0 m 5 2 0 1 0 m 9 2 0 1 1 m 1 2 0 1 1 m 5 2011m9 201 2 m 1 2 0 1 2 m 5 2 0 1 2 m 9 2 0 1 3 m 1 2 0 1 3 m 5 2 0 1 3 m 9 2 0 1 4 m 1 2 0 1 4 m 5 2 0 1 4 m 9 2 0 1 5 m 1 2 0 1 5 m 5 2 0 1 5 m 9

Fama-French's five-factor simulation portfolio

value (€)

22 finds higher excess alpha returns from 1995 to 2015, ranging from 6-7 %, which is likely caused by the better performance of the portfolio in the absence of Clinton’s healthcare reforms between 1992 and 1993. Tresl et al. (2014) find excess alpha returns of 3.8 % using Carhart’s four-factor model and data from 1985 to 2014. This paper employs Carhart’s model as well and finds excess alpha returns of 5.5 to 6.0 %. Unfortunately, Tresl et al.’s (2014) study is not available through the University of Groningen library portals, so no further reason can be provided on this return discrepancy, apart from the relatively small difference in the period used.

The current paper uses the five-factor model by Fama and French (2015). In this new framework, no excess alpha returns or medical innovation premium is found using data from 1990 to 2015. Between 1995 to 2015, a sizable excess alpha return of 4.7% is found, yet it does not reach statistical significance. The current paper estimated the five-factor model from 1963 until 2015 to keep comparability with the results of Koijen et al. From 1963, the five-factor model indicates a 4.9 % excess return, which is not significant (p=0.051). According to the R2_{, the five-factor model seems to explain}

returns of pharmaceuticals better, but the five-factor model features higher alpha and market beta volatility in rolling regressions compared to other models. The five-factor model does not perform better than Carhart’s model in the portfolio simulation of 2008 to 2016. However, this approach provides a snapshot and is not meant to favor the use of either one model or the other. The coefficient of CMA suggests the pharmaceutical portfolio is exposed to a conservative investment firms portfolio, which is in contrast to expectations of the high-profitability and high-investment nature of pharmaceuticals.

Two reasons may be behind this discrepancy, and they are not mutually exlcusive. First, the investments portfolio is defined as high minus low total asset growth firms, which may not serve as a good investment proxy in general or for pharmaceutical R&D specifically. Second, Higgins and Rodriguez (2006) find that large pharmaceuticals outsource or complement in-house R&D by engaging in mergers and acquisitions (M&A). Despite this, portfolio’s exposure to CMA would most likely not be affected by this, as using M&A to drive R&D happens predominantly between firms within the portfolio.

23 the value premium. This is just an illustration how factors may interplay with one another as the current paper also documents a high correlation between HML and CMA. Yet, Fama and French (2016) argue the five-factor model explains the high average returns of low market betas, share repurchases and low stock return volatility which is highly relevant for pharmaceuticals. Therefore, the current paper is cautious in drawing hard conclusions from the five-factor model. In conclusion, the five-factor model is the prime candidate model to proxy high profitability and R&D investments of pharmaceuticals but needs more understanding in the future.

This paper’s estimates are affected by healthcare reforms. From 1992m1, Clinton’s campaign launched healthcare reform proposals in the American national presidential election. In addition to price reforms, the proposals included more access to and consequently demand for drugs, but the market value of pharmaceuticals declined sharply. Between 1992m1 and 1993m10, Ellison and Mullin (2001) identify a cumulative 52.3% negative CAPM alpha return in a portfolio consisting of exclusively large pharmaceuticals. Ellison and Mullin are able to establish a causal relationship between these returns and Clinton’s reform agenda. The current paper finds excess alpha returns to be higher ex-post healthcare reform (1995-2015), but the impact of reform on the full sample period (1990-2015) is relatively limited given the findings by Ellison and Mullin. One reason behind this discrepancy could be that healthcare reforms influence large and high book-to-market pharmaceuticals more than smaller low book-to-market pharmaceuticals. Therefore, a part of the negative reform impact could be though the SMB or HML portfolios. However, SMB and HML portfolio returns are relatively low and constant, so the current paper is not able to provide strong evidence of this explanation. Despite this, the current paper does not aim to establish a causal relationship between factor betas and Clinton’s reform policies.

24 or CMA, which have minor added explanatory power but remain useful in determining the investment style of portfolios.

One- and three-year rolling alphas demonstrate short periods of excess alpha returns. The current approach uses averages of rolling regression estimates and thereby determines statistical significance more easily. The rolling regression section is intended to provide an evaluation of the time-variability of average windowed alphas and betas, yet is not meant to determine the presence of excess alpha returns or even a better causal relationship with the government-induced profit risk of the medical innovation premium.

The main contribution of the current paper is evidence that the pharmaceutical portfolio generated excess alpha returns between 1990 to 2015 according to widely accepted asset pricing models. The size of excess alpha returns of pharmaceuticals are impacted by Clinton’s healthcare reform period. Therefore, current paper’s results also indicate to the existence of a medical innovation premium for government induced profit risk or a similar phenomenon. The five-factor model does not document excess alpha returns of pharmaceuticals between 1990 and 2015. Between 1963 and 2015, estimation of the five-factor model results in an excess return of 4.9 %, yet not statistical significant (p=0.051). As a result, the paper by Koijen et al. would have concluded similar findings as the current paper when estimating the five-factor model.

6. Conclusion

25 interesting leads for new research in providing more background on the drivers of financial market returns by pharmaceuticals.

The five-factor model exhibits more volatility in rolling regressions estimates. After 2008, a simulation of theoretical returns resulted in marginally better prediction by Carhart’s model, compared to the five-factor model. It remains uncertain why a less volatile Carhart model would imply better predictions, apart from possible exposure to the momentum portfolio from 2013 to 2015. Certainly, employing three-year rolling windows does not incorporate short-term over-performances well, which was especially apparent after 2013.

The current findings confirm that investment managers would have generated excess alpha returns by holding the pharmaceutical portfolio between 1990 and 2015. For now, investment managers could significantly create excess returns through allocations for the pharmaceutical portfolio. It is not likely that excess alpha returns will be different or absent in the near future, despite whether excess alpha returns are a true asset pricing anomaly or serve as compensation for industry-specific risks. According to Sharpe ratios, the pharmaceutical portfolio does not compensate generously in more volatile periods, yet rolling alpha indicate outperformance of the pharmaceutical portfolio during financial crisis. For the future, investment managers aiming to create excess alpha returns are encouraged to be more cautious in allocations for pharmaceuticals once or if the five-factor model becomes a more widely accepted benchmark model. New innovative approaches must further document the medical innovation premium, such as possible interactions with large pharmaceuticals’ share-repurchasing premium or a premium of volatile cash-flows associated with patents. A wider replication of Koijen et al. could also contribute to the literature, for example addressing whether more government-risk related words in annual reports implies a higher medical innovation premium.

26

References:

- Carhart, M. M. (1997). On persistence in mutual fund performance. The Journal of Finance, 52(1), 57-82.

- DiMasi, J. A., Hansen, R. W., Grabowski, H. G., & Lasagna, L. (1991). Cost of innovation in the pharmaceutical industry. Journal of Health Economics, 10(2), 107-142.

- DiMasi, J. A., Hansen, R. W., & Grabowski, H. G. (2003). The price of innovation: new estimates of drug development costs. Journal of Health Economics, 22(2), 151-185. - Ellison, S. F., & Mullin, W. P. (2001). Gradual Incorporation of Information: Pharmaceutical

Stocks and the Evolution of President Clinton’s Health Care Reform*. Journal of Law and

Economics, 44(1), 89-129.

- Fama, E. F., & French, K. R. (1992). The cross‐section of expected stock returns. The Journal

of Finance, 47(2), 427-465.

- Fama, E. F., & French, K. R. (2015). A five-factor asset pricing model. Journal of Financial

Economics, 116(1), 1-22.

- Fama, E. F., & French, K. R. (2016). Dissecting anomalies with a five-factor model. Review of

Financial Studies, 29(1), 69-103.

- Fink, J., Grullon, G., Fink, K., & Weston, J. (2005). IPO vintage and the rise of idiosyncratic risk. 7th Annual Texas Finance Festival Paper.Available at SSRN:

http://ssrn.com/abstract=661321 or http://dx.doi.org/10.2139/ssrn.661321

- Giaccotto, C., Golec, J., & Vernon, J. M. (2011). New estimates of the cost of capital for pharmaceutical firms. Journal of Corporate Finance, 17(3), 526-540.

- Golec, J., & Vernon, J. M. (2007). New estimates of pharmaceutical research and development spending by US-based firms from 1984 to 2003. Managerial and Decision

Economics, 28(4), 481-483.

- Grabowski, H. G., & Vernon, J. M. (1987). Pioneers, Imitators, and Generics--A Simulation Model of Schumpeterian Competition. The Quarterly Journal of Economics, 491-526. - Grabowski, H. G., & Vernon, J. M. (1990). A new look at the returns and risks to

pharmaceutical R&D. Management Science, 36(7), 804-821.

- Harrington, S. E. (2009). Cost of capital for pharmaceutical, biotechnology, and medical device firms. Biotechnology, and Medical Device Firms. Oxford University Press

- Hassan, M., Patro, D. K., Tuckman, H., & Wang, X. (2007). Do mergers and acquisitions create shareholder wealth in the pharmaceutical industry?. International Journal of

Pharmaceutical and Healthcare Marketing, 1(1), 58-78.

27

- Hou, K., Xue, C., & Zhang, L. (2014). A comparison of new factor models (No. w20682). Unpublished working paper. National Bureau of Economic Research.

- Ikenberry, D., Lakonishok, J., & Vermaelen, T. (1995). Market underreaction to open market share repurchases. Journal of Financial Economics, 39(2), 181-208.

- Kaushik, A., Saubert, L. K., & Saubert, R. W. (2014). Performance and persistence of performance of healthcare mutual funds. Financial Services Review, 23(1), 77.

- Koijen, R.S.J., Philipson, T.J., & Uhlig, H. (2016). Financial Health Economics. Econometrica, 84(1), 195-242

- Myers SC, Shyum-Sunder L. Measuring pharmaceutical industry risk and the cost-of-capital. In: Helms RB, editor. Competitivestrategies in the pharmaceutical industry. Washington (DC): AEI Press, 1996: 208-237 Pharmaceutical Industry. American Enterprise Institute, Washington (DC).

- Myers, S. C., & Howe, C. D. (1997). A Life-cycle Financial Model of Pharmaceutical R & D. Program on the Pharmaceutical Industry, Sloan School of Management, Massachusetts Institute of Technology.

- Nijman, T., Swinkels, L., & Verbeek, M. (2004). Do countries or industries explain momentum in Europe?. Journal of Empirical Finance, 11(4), 461-481.

- Pammolli, F., Magazzini, L., & Riccaboni, M. (2011). The productivity crisis in pharmaceutical R&D. Nature reviews Drug discovery, 10(6), 428-438

- Roll, R. (1977). A critique of the asset pricing theory's tests Part I: On past and potential testability of the theory. Journal of Financial Economics, 4(2), 129-176.

- Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk. The Journal of Finance, 19(3), 425-442.

- Tresl, J., Karels, G., & Payne, B.C.(2014). Health Care Investing: Is a Higher Dose of Health Care Good for the Portfolio?. Journal of Investing, 23(1), 53-66.

28

Appendices

Appendix A: SIC codes per portfolio

1. Pharmaceutical portfolio

2830-2830 Drugs 2834-2834 Pharmaceutical preparations 2831-2831 Biological products 2835-2835 In vivo diagnostics

2833-2833 Medicinal chemicals 2836-2836 Biological products excl. diagnostics

2. Medical Equipment portfolio

3693-3693 X-ray and electromedical app 3840-3849 Surgical & medical instruments 3850-3851 Ophthalmic goods

3. Healthcare services portfolio

8000-8099 Health services

Appendix B: Figures of factor returns

-20 -15 -10 -5 0 5 10 15 1990m01 1990m12 1991m11 1992m10 1993m09 1994m08 1995m07 1996m06 1997m05 1998m04 1999m03 2000m02 2001m01 2001m12 2002m11 2003m10 2004m09 2005m08 2006m07 2007m06 2008m05 2009m04 2010m03 2011m02 2012m01 2012m12 2013m11 2014m10 2015m09

29 -20 -15 -10 -5 0 5 10 15 20 25 1990m01 1990m12 1991m11 1992m10 1993m09 1994m08 1995m07 1996m06 1997m05 1998m04 1999m03 2000m02 2001m01 2001m12 2002m11 2003m10 2004m09 2005m08 2006m07 2007m06 2008m05 2009m04 2010m03 2011m02 2012m01 2012m12 2013m11 2014m10 2015m09

Returns of SMB portfolio (%)

-15 -10 -5 0 5 10 15 20 1990m01 1990m12 1991m11 1992m10 1993m09 1994m08 1995m07 1996m06 1997m05 1998m04 1999m03 2000m02 2001m01 2001m12 2002m11 2003m10 2004m09 2005m08 2006m07 2007m06 2008m05 2009m04 2010m03 2011m02 2012m01 2012m12 2013m11 2014m10 2015m09

Returns of HML portfolio (%)

-40 -30 -20 -10 0 10 20 30 1990m02 1991m01 1991m12 1992m11 1993m10 1994m09 1995m08 1996m07 1997m06 1998m05 1999m04 2000m03 2001m02 2002m01 2002m12 2003m11 2004m10 2005m09 2006m08 2007m07 2008m06 2009m05 2010m04 2011m03 2012m02 2013m01 2013m12 2014m11 2015m10

30 -20 -15 -10 -5 0 5 10 15 1990m01 1990m12 1991m11 1992m10 1993m09 1994m08 1995m07 1996m06 1997m05 1998m04 1999m03 2000m02 2001m01 2001m12 2002m11 2003m10 2004m09 2005m08 2006m07 2007m06 2008m05 2009m04 2010m03 2011m02 2012m01 2012m12 2013m11 2014m10 2015m09

Returns of RMW portfolio (%)

-8 -6 -4 -2 0 2 4 6 8 10 12 1990m02 1991m01 1991m12 1992m11 1993m10 1994m09 1995m08 1996m07 1997m06 1998m05 1999m04 2000m03 2001m02 2002m01 2002m12 2003m11 2004m10 2005m09 2006m08 2007m07 2008m06 2009m05 2010m04 2011m03 2012m02 2013m01 2013m12 2014m11 2015m10