Risk and return in the healthcare
industry and the structure of
healthcare funding
Nick Stoffelsma - s2010321
Special Research Project: Financial healthcare economics Study Program: MSc Finance
Supervisor: Dr. Jochen Mierau
Field Key Words: Health care economics, medical innovation, risk premiums Words: 9567
Abstract
1 Introduction
In recent years, healthcare and pharmaceutical firms are ever more prominent in the news. The healthcare industry has led all sectors in total returns to its shareholders (McKinsey & Company, 2016). In 2013, US pharmaceutical giant Pfizer achieved an “eye-watering” 42% profit margin. Four other pharmaceutical companies made a profit margin of 20% or more in the same year; Hoffmann-La Roche, AbbVie, GlaxoSmithKline (GSK) and Eli Lilly. The critique has grown on the healthcare sector and “Big Pharma”, stating they crossed the line between essential profits and profiteering from lifesaving treatments. It is argued that some drug treatments costs up to $100,000, while the manufacturing of these products is only a tiny fraction of this (Anderson, 2014). Critics point to monopolistic pricing and profits, stating that making excessive profits on essential lifesaving treatments is amoral (Scherer, 2004). The healthcare and pharmaceutical industry rationalize these by claiming their research and development (R&D) are enormous, while only a fraction of the products make a profitable launch. Furthermore, they state that treatments save huge amounts of money over the long term (i.e. preventing long term care). These are just a fraction of the arguments used in the heated debate surrounding healthcare and pharmaceutical profits. Examining this discussion from an academic perspective could prove an interesting subject for a thesis. As this is an academic thesis for finance, I will not discuss the issue of morality here or dive into the ethical discussion. Instead, I will attempt to shed a financial perspective on this matter, focussed on the suggested excess returns.
Fama and French already make notice of this in their 1997 paper. They test their 1992 asset pricing models for a number of different industries. They find that these models are rejected for healthcare and real estate industries, due to a positive and significant alpha (intercept). This would indicate that there is additional risk in those industries, not incorporated in or explained by their current asset pricing models. In recent work, Koijen et al (2016) go into more depth on this return anomaly in the US. They find that the returns of firms involved in medical R&D are substantially higher, around 4-6% on yearly basis, than standard asset pricing models would predict. They call this phenomenon the “medical innovation premium”. In their study they suggest that this premium is caused by the investor’s need to be compensated for non-standard risk. Koijen et al. explain this risk as risk associated with the possibility of government intervention.
This thesis attempts to further investigate this premium, as to date little is known about this topic. Findings could provide an interesting perspective in the discussion on healthcare returns. Furthermore, as this premium appears to be dependent on government intervention, and therefore have some relation to government policy, it is likely to vary around the world, as government policy is very diverse. This would suggest that firms in different regions would bear different amounts of government intervention risk, which could influence the investors demand for risk compensation, and thus the returns of firms. This implies that returns of healthcare firms around the world have different underlying risk. However, as profits of these firms are made worldwide, they could also be subject to several different governments and thus intervention risks. Hence, an interesting topic to study if one desires to discuss industry returns.
The question which I try to answer in this thesis, is whether a medical innovation premium exists in other markets, and whether there is a relationship with the share of publicly financed healthcare, a proposed measure of government intervention risk. This paper adds to the discussion on returns in the healthcare industry and provides an financial perspective on healthcare returns in European markets. It aims to provide and explanation for high industry returns.
This paper is organised in the following way; first an overview of the theoretical literature will be provided, followed by the methodology used and a description the data. An overview of the results will be provided in section 4, which will be succeeded by a discussion and conclusion section.
2 Literature
In this thesis, when the healthcare industry is discussed, this consists of healthcare equipment, services, pharmaceuticals and biotechnology firms. I will begin to discuss the industry, its profitability and R&D, continue with the medical innovation premium and its relation to healthcare and governments.
2.1 The healthcare industry
The health care industry is one of the world’s largest and fastest growing industries, accounting for around 17,5% of GDP in the US (EU average 10%) with an estimated average growth rate of 5,8% per year for the coming decade (CMS, 2014). Medical innovation is central to the growth of this sector and the improvement of health. The healthcare industry is characterized by extensive government intervention, intractable uncertainty, information a-symmetry, barriers to entry, externalities and the presence of a third-party agent (Phelps, 2016). One unique characteristic is the industry’s dependency on foreign government policy. Almost all of the profits are made worldwide, while the R&D is done locally. Thus domestic growth of these firms is stimulated by the ability to profit from innovations which are distributed worldwide. This implies that foreign government policy effects local healthcare R&D and spending (Egan and Philipson, 2013).
Angell claims these profits are excessive, and the result of excess prices, principally in the United States, the only advanced country that does not limit pharmaceutical price increases in some way. Most of the profits from all pharmaceutical giants are made in the US (Angell, 2004).
2.2 Abnormal returns and profitability
The pharmaceutical industry claims that these seemingly high returns are necessary to compensate for high risk in research and development. Research has shown that the healthcare industry has a higher cross-sectional variance in returns than other industries. Comanor (1986) however, shows that these variances are mainly high due to unusually high returns, rather than frequent sub-normal returns. Time-series analysis showed that returns in the healthcare industry are more stable than the average of the manufacturing industry. Furthermore, CAPM tests showed beta values lower than one, indicating no abnormal systematic market risk (Scherer, 1993).
However, research from the US office for technology assessment has shown that firms active in medical R&D tend to overstate true economic profitability when using conventional accounting practices (Herdman, 1993). Profit reports are prepared including current-year write-off of R&D expenses, however they do not properly account for growth rates of healthcare firms. R&D is the leading cause of growth in the healthcare industry. If the R&D outlays are capitalized and amortized at plausible rates, the overall rate of return in the 70’s and 80’s of the previous century was reduced to two or three percentage points more than other industries (Baber and Kang, 1996).
2.3 R&D and risk
In an analysis into the value of health and medical innovation by Murphy and Topel (2005), they come to the unexpected conclusion that “returns to basic research may be quite large, so that substantially greater expenditures may be worthwhile.” With this they imply that they would expect much larger investments in R&D than are actually observed, since these ‘investments’ would have a positive net present value. However, apparently investors require more annual return, which in turn cannot be invested in positive NPV R&D projects.
In recent work; “Financial health care economics” (2016), Raplh Koijen, Tomas Philipson and Harald Uhlig examined how financial returns of investing in medical R&D are related to growth of health care spending. They attempt to explain this “missing R&D puzzle”. Koijen et al. (2016) apply the CAPM and the Fama-French methodology, and find a positive and significant alpha (intercept) in the healthcare industry. This indicates that there is a risk premium for holding healthcare stocks which standard asset pricing models do not explain. This is in line with the paper of Fama and French (1997), in which they reject their Fama and French (1992) models for the healthcare and real estate industries, as they find positive and significant intercepts.
2.4 The medical innovation premium
Koijen et al. (2016) describe this phenomenon as the “medical innovation premium”, a significant risk-premium of 4-6% annually for equity returns of firms in the health care sector in the US for the last four decades. This premium, the authors find to be compensation for investors due to government-induced profit risk, or abnormal return patterns at threats of government intervention. The possibility of government interference results in investors demanding higher returns on firms involved with medical R&D.
In their paper they model that the size of the health care industry would be 3% of GDP larger than it currently is if there was no government intervention risk. Moreover, they show that R&D investment would double if the medical innovation premium would be absent. The authors conclude that the growth of the healthcare sector depends on medical R&D, which is a product of government intervention. However, government intervention induces greater uncertainty on government risk, which results in higher risk premium, which in turn discourages medical R&D. This feedback mechanism implies that growth intended from governmental programs is tied to the risk of policies surrounding those programs (Koijen et al., 2016).
2.5 Healthcare systems
Koijen et al. (2016) apply to U.S. firms. It is to date not known whether these premiums are similar in other markets. Furthermore, it is unclear whether accounting practices have a similar effect on the returns in other markets. However, since most of the largest healthcare firms are all active worldwide, it is reasonable to assume that these firms apply similar methods of accounting. Nonetheless, current research does not yet provide an industrywide explanation.
Six of the 10 largest healthcare firms are European, and the European healthcare market differs significantly from the US. In addition, the EU has developed healthcare and stock markets, and reliable data for the past 30 years. These characteristics make the EU a suitable candidate for research. The healthcare system in the US is quite different from Europe, as Europe has high heterogeneity within healthcare systems. The US healthcare spending per capita and healthcare expenditure as percentage of GDP are among the highest in the world. Healthcare spending in the US is estimated around 48% of the world’s total spending, but the US only amounts to ~24% of global GDP (Egan and Philipson, 2013). Furthermore, 45% of spending was publicly financed. (World bank, 2015)
In contrast, European markets have much lower healthcare expenditure per capita and as percentage of GDP. Moreover, in many European member States either government schemes or compulsory contributory health care financing schemes (Heron after referred to as “public”) dominate voluntary health care payment schemes and household out-of-pocket payments (referred to as “private”)(Eurostat, 2016). In Europe, between 70-80% of healthcare is funded through these public social security systems. Healthcare spending in the EU is composed of; curative care (~50%), Medical goods (~25%), long term care (~20%) and other (ancillary services, administration etc.)(Eurostat, 2016). Curative and long term care are usually non-profit or privately funded. For profit firms engaged in medical goods thus make up the bulk of the firms listed on exchanges.
2.6 Governments and Healthcare
additional compensation for investors in healthcare firms, hinting towards a larger innovation premium.
2.7 Summarizing the theory
To summarize the main findings from literature, the returns in the healthcare industry are found to be significantly higher than other industries. However, previous studies indicate that accounting practices are partially responsible for inflating returns. Furthermore, research suggests substantial risks in R&D for healthcare firms. Several studies have indicated that firms in the healthcare industry need to provide extra compensation for non-standard risk. There appears to be a significant “medical innovation premium” on firms in the healthcare industry, following government induced profit risk. Yet, the current findings apply to US firms. The height of the premium, if any, in the EU is to date not known. However, it is evident that in the EU the government has more potential to intervene, as a larger share of healthcare expenditure is financed publicly. More potential of governments to intervene would suggest a higher potential intervention risk, and therefore risk premium. However, it is yet unknown whether this relation holds. A higher medical innovation premium in the EU, due to this extra risk, would thus indicate that EU healthcare returns are higher than US healthcare returns.
Based upon the existing literature, I would expect that the returns of healthcare firms in the EU would be higher than in the US. This due to a larger medical innovation premium, stemming from more compensation due to government intervention risk, as EU governments have more potential to intervene.
To date I recognize two main gaps in theory; (1) to what extent, if any, medical innovation premium exists in the EU, and (2) whether the inferred relation between the medical innovation premium and the structure of healthcare funding holds. The first gap is essential in explaining industry returns, as the industry concerns many European firms. The second gap is a step towards explaining and predicting worldwide industry returns. In the remainder of this thesis I will try fill these gaps by testing the theoretical predictions empirically.
3 Methodology & data
3.1 Research objective & hypotheses
This research aims to examine the medical innovation premium in more detail and whether this premium is also evident in other markets. It attempts to identify what the size, if any, of this premium in European markets is, and whether this premium and is related to the structure of healthcare financing. From the theory the following hypothesizes are formulated;
H1: There is a medical innovation premium present in European markets
H2: The medical innovation premium is higher in regions with a larger share of publicly financed health care.
3.2 Methodology
The methodology in this paper start similar the one used in Koijen et al. (2016). The returns of firms in the healthcare industry are studied using the Capital asset pricing model (Sharpe, 1964) and the Fama and French three factor model (Fama and French, 1993) with momentum factor (Carhart, 1997). I am interested in the intercept α of the time-series regressions. A significant alpha measures a differential average return of the healthcare industry which cannot be explained by the asset pricing models. The capital asset pricing model
𝑅𝑝𝑡− 𝑟𝑓𝑡 = 𝛼 + 𝛽(𝑅𝑚𝑡− 𝑟𝑓𝑡) + 𝜀𝑡 (1)
Where Rpt is the return of a portfolio of assets at time t, rft is the risk free rate at time t, Rmt the return of the market portfolio at time t. This results in an intercept α, and slope β and error term ε. The four factor model
𝑅𝑝𝑡− 𝑟𝑓 = 𝛼 + 𝛽1∗ (𝑅𝑚𝑡− 𝑟𝑓) + 𝛽2∗ 𝐻𝑀𝐿𝑡+ 𝛽3∗ 𝑆𝑀𝐵𝑡+ 𝛽4∗ 𝑊𝑀𝐿𝑡+ 𝜀𝑡 (2)
poorly in the past year) have lower expected returns than the CAPM prediction. The HMLt, SMBt and WMLt factors are account for these regularities in asset markets. α being the intercept, and β1-4 slope coefficients to the factors, and εt being the residuals. Fama and French (2012) find that size, value and momentum all provide a better explanation of variation in returns in both US and EU markets, although this effect might be stronger in the US. Therefore, all four factors will be incorporate in the study.
To estimate these alphas, first portfolios of healthcare stocks will be to be composed. In this research simple returns will be used, as weighted portfolios of healthcare stocks are evaluated. This follows the line of reasoning used in Brooks (2014); as “the simple return on a portfolio of assets is a weighted average of the simple return on the individual assets”. This property does not hold for continuously compounded (log) returns, therefore simple returns will be used. Total return index (TRI) will be used to calculate the portfolio returns. The total return index is a price index which tracks price movements but also includes dividends, interest, rights offerings and other distributions realized over a given period of time. Simple returns are calculated as
𝑅𝑖𝑡 =
𝑝𝑖𝑡− 𝑝𝑖𝑡−1
𝑝𝑖𝑡−1 ∗ 100% (3)
Where Rit is the return of asset i at time t, pit the total return index of the asset i at time t and
pit-1 the total return index of asset i at time t-1. Value weighted portfolio returns are calculated as 𝑅𝑝𝑡 = ∑ 𝑤𝑖𝑅𝑖𝑡 𝑁 𝑖=1 , 𝑤𝑖= 𝑀𝑉𝑖𝑡 𝑀𝑉𝑇𝑡 (4)
Where Rpt is the return on the portfolio of assets at time t and Rit is the return of asset i at time t. wi is the value-weight for each asset, with MVit defined as the market value of asset i at time t and MVTt the total market value of all assets at time t.
explanatory value, and data is readily available, multiple regressions will be performed to determine the best indicator. The regression models are
𝛼𝑖 = z + 𝛽 ∗ 𝐻𝐶𝑖 (5) [ 𝛼𝑖1 ⋮ 𝛼𝑖7] = [ 𝑧1 ⋮ 𝑧7] + [ 𝛽1 ⋮ 𝛽7] ∗ [𝐻𝐶𝑖1… 𝐻𝐶𝑖7] (6)
Where αi is the alpha of country i, z1 the intercept, β the slope parameter and HCi1 a healthcare expenditure indicator.
3.3 Data
To estimate the premium, portfolios of healthcare stocks are to be regressed on the market portfolio and representative factors. Therefore, data on US and EU healthcare firms, market returns, risk free rates and factors are needed.
3.3.1 US data
The US industry portfolio and Fama -French US factors are gathered from Kenneth French’s website (which he in turn gathers from Center for Research in Security Prices (CRSP) and supplements by Datastream and Bloomberg). The US industry returns are classified in five portfolios: consumer goods, manufacturing, technology, healthcare and other. For this study, only the US healthcare portfolio and market returns are used. The Fama-French factors are different from the European ones. The risk free rate is assumed to be the one month US treasury bill rate, a relatively common assumption (see Fama and French 1993, 1997, 2012) , and is also collected from Kenneth French’s website. The risk free rate is assumed to be the same in the EU and the US regions, and not unique to the US, similar to Fama and French (2012). The returns acquired from Kenneth French’s website are value-weighted.
3.3.2 EU data
The returns on European healthcare firms are gathered from Thomson Reuters DataStream. Healthcare here includes the classifications; Healthcare equipment & services, Pharmaceuticals and Biotechnology, resulting in a total of 518 European firms. The total return index and market value metrics are used. The returns are value-weighted as described in the methods section. The most recent available Fama-French European factors range up until July 2015. From here on, a sample of 20 years, monthly, is selected, ranging from July 1995 to July 2015. Which provides a total sample of 241 (12*20+1) time series observations. The European data is combined to a European healthcare portfolio
3.3.3 Descriptive statistics
Table 1 provides the descriptive statistics for the US and EU return series. It is visible that the mean return of the EU portfolio is higher than the US portfolio, it also has a larger standard deviation, minimum and maximum values. Annualizing these monthly returns provides an average return of 13,6% for the US healthcare portfolio and 17,1% for the EU portfolio. This is in line with the expectation from the theory, that EU returns would be higher due to higher intervention risk.
Table 1
Descriptive statistics US and EU healthcare returns
The table provides an overview of the descriptive statistics for the four return series. All statistics are computed with monthly returns. Reported are the maximum and minimum one-month returns observed in the sample, the mean average excess return (monthly), the (monthly) standard deviation of each factor, excess kurtosis and skewness. The sample returns for the EU and US healthcare portfolios are from 1995.07 to 2015.07.
Portfolio Mean Median Maximum Minimum Standard deviation
Skewness Kurtosis Observa-tions EU healthcare 1.33 1.60 16.43 -13.31 4.29 -0.04 1.27 241 US healthcare 1.07 1.28 12.00 -12.26 4.18 -0.33 0.30 241
Figure 3-1 Histogram of returns
3.3.4 Regions
Separate healthcare portfolios per region are made. This is done for the five regions: Denmark, France, Germany, United Kingdom and Switzerland, composed of 321 firms (~60% of total) contributing to almost 90% of the total value of the European portfolio. These regions provide enough firms each country to enable statistical results, and their value share serves representative for the total sample. The regional portfolios are shown in table 2.
Table 2
Descriptive statistics regional returns
The table provides an overview of the descriptive statistics for the five selected regions. All statistics are computed with monthly returns. Reported are the maximum and minimum one-month returns observed in the sample, the mean average excess return (monthly), the (monthly) standard deviation of each factor, excess kurtosis and skewness. Firms denotes the number of firms in the regional sample, share MV denotes the average relative market value of each region to the total European portfolio over the sample period. The sample returns for regional portfolios are from 1995.07 to 2015.07, providing 241 time series observations.
Region Mean Median Maximum Minimum Standard deviation Skewness Kurtosis Firms Share MV (%) Denmark 2.28 2.49 23.20 -23.35 6.19 -0.15 1.33 20 6.7 France 1.53 2.20 21.14 -18.81 5.48 -0.20 0.84 90 12.2 Germany 1.65 1.67 45.67 -23.13 7.03 1.40 8.68 61 4.2 Switzerland 1.03 1.36 15.57 -12.71 5.13 0.11 0.11 29 39.2 United Kingdom 1.14 0.99 23.73 -17.56 5.94 0.18 1.44 121 33.3
Evident from table 2 is that some regions exhibit a higher mean returns than others, however they vary with standard deviations. Switzerland and the UK compose the largest value-share of total EU healthcare portfolio. Furthermore, Germany’s return appear to be leptokurtic, as it has a relatively high excess kurtosis.
Figure 3-2 depicts the histogram of the different regions. The graph also shows leptokurtic returns for Germany. In general, the returns appear relatively normal distributed, and as each regional series has 20 years totalling 241 time series observations, central limit theorem can be applied to assume normalized means. (Rice, 2006)
Figure 3-2 Histogram of regional returns
3.3.5 Healthcare
To evaluate whether the premium is different per region or dependent on healthcare expenditure, data is gathered from the World Bank (2016) over the last 20 years. An overview of the descriptive statistics is given in table 3. Total expenditure shows an increasing trend over time, which is in line with general consensus of growing healthcare expenditure over time. However, the financing of these healthcare costs, the share which is funded private or publicly remains relatively constant over time.
Table 3
Descriptive statistics healthcare expenditure
The table provides an overview of the healthcare indicators per region. Health expenditure per capita is denoted in US$ and in international dollars converted using 2011 purchasing power parity (PPP) rates. Total healthcare spending is given as a percentage of Gross Domestic Product(GDP). Public and private share of healthcare funding (as % of GDP) add up to total healthcare (% GDP) by definition. Public expenditure is also provided as percentage of total health expenditure and as percentage of total government expenditure. All values are the arithmetic means. The samples are from 1995 to 2014, resulting in 20 observations per region. Health expenditure in absolute terms is showing an upward trend over time in all regions.
Country Health expenditure per capita (US$) per capita, PPP total (% of GDP) private (% of GDP) public (% of GDP) public (% of total health) public (% of government) Denmark 4540 3358 9.7 1.5 8.2 84.2 14.8 France 3612 3241 10.6 2.3 8.3 78.4 15.3 Germany 3735 3484 10.5 2.3 8.2 77.9 17.6 Switzerland 5970 4281 10.6 4.2 6.3 59.7 18.4 UK 2759 2525 8.1 1.5 6.6 81.6 15.1 EU 2531 2481 9.2 2.1 7.1 77.1 14.8 US 6485 6485 15.0 8.2 6.8 45.4 18.2
4 Results
4.1 Medical innovation premium
The CAPM and Fama–French regression models are estimated using the data described in the data section. The models are repeated for reader convenience.
𝑅𝑝𝑡− 𝑟𝑓𝑡 = 𝛼 + 𝛽(𝑅𝑚𝑡− 𝑟𝑓𝑡) + 𝜀𝑡 (1)
𝑅𝑝𝑡− 𝑟𝑓 = 𝛼 + 𝛽1∗ (𝑅𝑚𝑡− 𝑟𝑓) + 𝛾 ∗ 𝑆𝑀𝐵𝑡+ 𝜆 ∗ 𝐻𝑀𝐿𝑡+ 𝜑 ∗ 𝑊𝑀𝐿𝑡+ 𝜀𝑡 (2)
Table 4
Regression results
The table provides the results of CAPM and Fama-French regressions on healthcare returns, and the values reported by Koijen et al (2016). The first two coloums are the CAPM coefficients, column 4-8 show the Fama-French regression coefficients. The first row shows the alphas as found by Koijen et al. Their study did not provide other statistics than alphas, and the other coefficients are personal estimates. Two portfolios are reviewed,the EU and US healthcare portfolios as discussed in the data section. All statistics are computed with monthly returns. The EU results are estimated using european market returns and factors, while for the US results north american market returns and factors are applied. β 's are the market portfolio coefficients, γ, λ, φ, the SMB HML and WML coefficients respectively. The R2 provides a measure of fit for the model, where the value below the R2 is the adjusted R2. The sample from Koijen et al. is from 1961.01-2013.07, the sample from the EU/US healthcare portfolios is from 1995.07-2015.07. All results are monthly parameters, and HAC corrected using the Newey-west method.
Portfolio CAPM Fama-French α (t-Stat) β (t-Stat) R2 (Adjust.) α (t-Stat) β (t-Stat) γ (smb) (t-Stat) λ (hml) (t-Stat) φ (wml) (t-Stat) R2 (adjust.) *Koijen et al 0.220** 0.849 0.59 0.398** 0.852 -0.251 -0.310 - 0.64 (1.76) (18.37) 0.59 (3.29) (20.81) (-4.66) (-3.06) - 0.64 EU 1.055 0.126 0.02 1.100 0.117 -0.485 -0.168 0.035 0.10 (4.24) (2.21) 0.01 (4.16) (2.22) (-4.29) (-1.65) (0.50) 0.08 US 0.469 0.619 0.44 0.455 0.660 -0.247 -0.065 0.045 0.47 (2.63) (10.98) 0.44 (2.48) (9.97) (-2.72) (-0.52) (0.62) 0.46
* Koijen et al only provide alphas statistics. Other values are personal estimates over the (same) sample period 1961.01 to 2013.07
** These are the alphas reported by Koijen, personal estimates amount to CAPM alpha of 0.224 (1.74) and F-F alpha of 0.408 (3.50)
The first row shows the monthly results Koijen et al. (2016) find. As they do not provide other estimates than alphas, the other coefficients are personal estimates. The alphas found in the estimation over the same period as Koijen are very close to the ones reported. Their results in both the CAPM and Fama-French models a highly significant and positive beta. The other Fama-French coefficients are also significant at a 1% level. Furthermore they find two positive alphas, of which the three factor one is significant at a 1% level. They therefore conclude that there is a premium for risk not explained by the model, which they address as the medical innovation premium. (Koijen et al., 2016)
The alphas of both portfolios in both models are positive, and except for the alpha in the US Fama-French model(significant at 5%), significant at a 1% level. These significant intercepts imply there is a part of the return not explained by these models. This is an indication of a premium for a risk not incorporated by these asset pricing models. This thus provides evidence for the medical innovation premium, as described by Koijen et al. (2016), in both the US and the EU for the period 1995-2015. Additionally, the premium appears to have increased in recent years.
The premium in the EU is higher than in the US, which is in line with the theoretical expectation. However, one important side note must be made: the models in the US have much greater explanatory power (measured by R2) than the European models. In the US, the models appear to explain around 50% (R2 = 0.5) of the variation in asset returns, while in the EU this is only around 10% (R2 = 0.1). Therefore, inherently the alphas are higher, as the intercept measures the part not explained by model parameters, which thus must be higher in the EU. Hence, the asset pricing models do a relatively poor job at explaining healthcare returns in the EU. However, no better asset pricings models are currently at hand.
Annualizing these “premiums” (multiply by 12, see (Koijen et al., 2016)) results in annual premiums of 5-6% on the US healthcare portfolio and 12-13% on EU healthcare portfolio. This seems like a high premium in the EU, however, this is could in part be explained by the fact that the asset pricing models provide poor explanation of variation in European healthcare returns. Hence, the critical assumption here is, that with the identified premium is not caused by exposure to other risk factors. This assumption this could be justified, as the asset pricing models explain reasonable amount of the variation in the returns of the EU (Fama and French, 2011).
4.2 Regional premiums
Table 5
Regional regression results
The table provides the results of CAPM and Fama-French regressions on healthcare returns of different european regions. The first two coloums are the CAPM coefficients, the last five show the Fama-French regression coefficients. Five regions are reviewed, which are discussed in the data section. All statistics are computed with monthly returns, and European market returns and factors are applied. β 's are the market portfolio coefficients, γ, λ, φ, the SMB, HML and WML coefficients respectively. The R2 provides a measure of fit for the model, where the value below the R2 is the adjusted R2. The regional healthcare portfolios are from 1995.07-2015.07. All results are monthly parameters, and HAC corrected using the Newey-west method.
Portfolio CAPM Fama-French α (t-Stat) β (t-Stat) R2 (adjusted) α (t-Stat) β (t-Stat) γ (smb) (t-Stat) λ (hml) (t-Stat) φ (wml) (t-Stat) R2 (adjusted) Denmark 1.871 0.346 0.08 1.939 0.377 -0.092 -0.293 0.029 0.10 (4.68) (4.78) 0.08 (4.82) (5.15) (-0.45) (-1.47) (0.38) 0.08 France 1.186 0.238 0.05 1.231 0.221 -0.367 -0.081 0.003 0.07 (3.70) (3.85) 0.05 (3.79) (3.21) (-1.92) (-0.56) (0.04) 0.06 Germany 1.175 0.468 0.12 1.630 0.525 0.039 -0.945 -0.129 0.23 (2.34) (3.84) 0.11 (3.15) (5.48) (0.10) (-3.64) (-1.12) 0.22 Switzerland 0.810 0.081 0.01 0.845 0.056 -0.619 -0.103 -0.033 0.08 (2.72) (1.18) 0.00 (2.64) (0.82) (-4.23) (-0.76) (-0.41) 0.07 UK 0.962 -0.017 0.00 1.028 -0.029 -0.549 -0.197 0.029 0.05 (3.17) (-0.19) 0.00 (3.32) (-0.37) (-3.08) (-1.31) (-0.27) 0.03
It is evident form the table that for the regions Denmark, France and Germany, both models have significant betas and provide reasonable fit, with the Fama-French providing better explanatory value. However, for the regions Switzerland and the UK, no significant betas are found, and only the SMB coefficient appears to be significant at a 1% level. Therefore, the estimations of alpha are not very meaningful, as the CAPM model does not provide fit and thus all is explained by the intercept, while the Fama - French provides some fit. The CAPM premium in these regions will not provide a fair measure of premium. The part not explained by the model amounts to 9-22% annually. Similar to section 4.1, a critical assumption is made that the identified premiums are not caused by exposure to other risk factors.
4.3 Regional healthcare
The medical innovation premium is hypothesized to be higher in countries with a larger share of publicly financed health care. To test this, the premiums obtained per region are regressed on the different healthcare indicators. The models is repeated for reader convenience. [ 𝛼𝑖1 ⋮ 𝛼𝑖7] = [ 𝑧1 ⋮ 𝑧7] + [ 𝛽1 ⋮ 𝛽7] ∗ [𝐻𝐶𝑖1… 𝐻𝐶𝑖7] (6)
I am interested in models which estimate significant slopes (β’s) and insignificant intercepts (z’s). Furthermore, the model must have explanatory value, measured by the sum of squared residuals (R2). Table 6 shows the results of the regressions.
Table 6
Healthcare regression results
The table shows the result of the premiums found in previous models regressed on the different healthcare indicators. Z is the intercept of the regression, Beta the slope and the R2 provides a measure of fit for the models. The value in brackets is the t-statistic of the coefficient. The value below the R2 provides the adjusted R2. The premiums used are over the period of 1995-2015, on the average health care indicators over the same period.
Health expenditure indicator CAPM premium Fama- french premium
z (t-statistic) β (t-statistic) R2 (adjusted) z (t-statistic) β (t-statistic) R2 (adjusted) Public (% of GDP) -1.62 0.37 0.50 -1.87 0.44 0.58 (-1.34) (2.25) 0.40 (-1.49) (2.61) 0.49
Public (% of government expenditure) 3.76 -0.16 0.41 3.72 -0.14 0.24 (2.59) (-1.86) 0.29 (2.00) (-1.26) 0.09
Public (% of total health expenditure) -0.68 0.02 0.64 -0.63 0.03 0.67 (-1.15) (3.01) 0.57 (-0.97) (3.15) 0.60
Thus, the results suggest that the public health expenditure, as a % of total health expenditure and as a % of GDP, provide an explanation of the healthcare premium found in both European and US markets. However, as the R2 of the public expenditure as a % of total health expenditure is larger, this would imply a more representative measure. Furthermore, this indicator is relatively stable over time, as the public healthcare as a % GDP shows an increasing trend over time. Table A.6 in the appendix gives an overview of the alphas regressed on all healthcare indicators. It is evident from the table that only public healthcare expenditure indicators provide explanatory power. The findings are in line with the hypothesis and thus provide empirical evidence for this relation.
4.4 Summary results
To summarize, a statistically significant premium has been identified on healthcare stocks. This indicates that there is a premium which conventional asset pricing models do not explain. However, the models provide limited fit, and thus it must be assumed that these returns are not due to other risk factors. Nonetheless, significant premiums have been found in several European regions. Furthermore, evidence has been put forward that these premiums are related to the structure of healthcare financing. Public healthcare expenditure as a percentage of total healthcare expenditure yields the most explanatory power over the premiums. Therefore providing indication that this serves as a proxy of government interference. This is in line with the theory, which predicted that governments with more intervention potential would cause more intervention risk and thus premium.
4.5 Robustness tests
4.5.1 Stationarity and normality
The data is tested for stationarity and normality. As in the CAPM and Fama-french regressions have a sample size of more than 150 observations, central limit theorem would
suggest that the mean is distributed normally (Rice, 2006). The time-series data are returns and thus all will be expected to be stationary. The series are tested for unit root using the Augmented Dickey-Fuller test, and all series showed stationarity (Dickey and Fuller, 1979). Furthermore, the expected value of the residuals is equal to zero as there is an intercept present in the regression. As the data shows stationary and normality, the results are robust towards stationarity and normality issues.
4.5.2 Heteroscedasticity and Autocorrelation
The residuals of the regression are tested for heteroscedasticity using White’s heteroscedasticity test of squared residuals (White, 1980). The null hypothesis of homoscedasticity is rejected for the returns of different regions. Therefore, some measure of heteroscedasticity correction needs to be applied. Furthermore, the residuals are tested for autocorrelation using the Durbin Watson test-statistic and Breusch-Godfrey serial correlation tests for 2 and 12 lags, to find possible one month “momentum” and 12 month “calendar effect” autocorrelation (Breusch and Pagan, 1980). There was little indication of 12 lag autocorrelation, but several occurrences of the a rejection of the H0, no autocorrelation, in one lag. As there is both indication of autocorrelation and heteroscedasticity, Newey-West Heteroscedasticity and Autocorrelation Consistent (HAC) standard errors are applied (Newey and West, 1986). Applying these corrections ensures that the results are robust for heteroscedasticity or autocorrelation in the data.
4.5.3 Data
As the models provide relatively poor fit, they are tested on other data sources, to validate the appropriateness of the data. Recently, exchange traded funds focussed on healthcare regions have become available. These ETFs seek to track the investment results of an index composed of equities in the healthcare sector. They are value-weighted return trackers, similar to the portfolio’s I have constructed from EU and US data. These ETFs will be used to benchmark the data gathered on the returns.
For the US the “iShares U.S. Healthcare ETF” is available, and for the EU the “iShares
STOXX Europe 600 Health Care UCITS ETF” is available. Additionally, the global ETF: “iShares Global Healthcare ETF” is supplied. The benchmarks ETF have data from 2001
Figure 4-2 Descriptive statistics return series
Descriptive statistics of the samples can be found in table A.1 in the appendix, histograms of in Figure 4-2 and regression results in table A.2. Although the European ETF is not identical to the one constructed in this thesis, a premium is evident as positive and significant in all ETFs. The premium is lower on the ETF, as the model has somewhat better fit (measured by adjusted R2), and thus less is explained by the intercept . However, the composition of the ETF is largely the same as the portfolio, therefore suggesting the same result. See appendix table A.3 for more details. Thus, the results appear to be robust for different datasets.
As most of the healthcare industry’s profit is made in the US, north American factors are applied to European healthcare returns, in an attempt to find a better explanation for the returns. However, the estimators and fit these models provided was worse than when applying European factors, when compared by R2.
4.5.4 Sample period
The regression results in section 4.3, although significant, are based on relatively few data points. Furthermore, it is possible that this relationship holds only for this full time period. To test whether the results are representative, the sample is split in periods of 5 and 10 years. This increases the number of data points, and could strengthen the relationship found. However, this also has a negative effect on the estimated parameters in the models. Splitting the sample reduces significance of α and β estimates, and therefore the precision of the estimates. The results are depicted in Figure 4-3 and Figure 4-4, and the regression results shown in Table A.4 and A.5 in the appendix.
Figure 4-3 Regional alphas over 10 year periods vs publicly financed health expenditure (as % of total health expenditure)
Figure 4-3 show the premium versus the share of publicly financed healthcare, with the sample split in two 10 year periods, from 1995 to 2005 and from 2005 to 2015. Figure 4-4 shows the sample split in four 5 year periods. Although the models and coefficients are not significant in several estimates of the 10 year periods, the results indicate a similar relationship. In the 5 year periods, many estimates are insignificant, however the graph does not indicate a large deviation from the relationship found. The premiums are increasing with the share of publicly financed healthcare. The reason for insignificant estimates is probably a small sample of data and high standard deviations in these samples. Nonetheless, the results appear to be robust for different time periods.
4.5.5 Other
To test whether the relation between the components might be non-linear, the regressions have been estimated with squared explanatory variables. The results yield no significant results coefficients on squared explanatory variables, indicating linearity (except WML). The sample includes dead firms, and thus it is unlikely that a survivor bias is present in the results. Furthermore, industry returns are studied and are thus value-weighted. The regressions are re-estimated without insignificant independent variables, but this did not influence the findings in a substantial way.
5 Discussion
This thesis intended to study the returns in the healthcare industry in more detail. Returns in this industry appear excessive, however studies have revealed that these returns might not be as high as they appear. Furthermore, recent research has shown that this industry has inherently more risk than other industries. So far, this has been studied for the US, however, a large share of this industry is located in Europe. Therefore these US premiums do not provide a consistent explanation for the industry.
5.1 Limitations of research 5.1.1 Limited model fit
The main limitation of this study is the assumption that the premiums are not attributable to other risk factors. Even tough asset pricing models work reasonably well in Europe and therefore this assumption can be justified, a low model fit suggests clouded results. Compared to the US, where model fit from 0.4 to 0.6 are achieved, the model fits achieve a mere 0.1 in the EU. It is therefore possible that the healthcare industry is exposed to other risks, not incorporated in these models, which are not attributable to government intervention risk. All market regularities not explained by the asset models are incorporated in the alpha. If this assumption is found not to hold, it is not possible to separate the part of the premium is attributable to the potential of government intervention, and which is attributable to other risk factors not explained by the models. Then the alphas are not a fair indicator of government intervention risk.
5.1.2 Regional medical innovation premiums
Another limitation of the results is limited number of data on regional premiums. The sample is split in five regions, of which only three regions provide good model fit for the alpha. The two regions which did not provide good model fit, did provide alphas which were in line with the expectation, however their betas did not provide explanatory power. Using these data points is thus questionable. Furthermore, the additional region used, the EU, is largely composed of the other regions, and therefore does not provide (completely) unique explanatory power.
In an attempt to overcome this issue, different sample periods are applied in section 4.5.4. By splitting the sample in 5 and 10 year periods, more data becomes available. However, the significance of the model parameters estimated over these periods is greatly reduced. A similar relation is found, and thus reduces the need for extra regional premiums. However, this does not solve the issue of the low model fit. The strength of the relationship found is therefore debatable.
5.2 Implications of results
and the medical innovation premium. This supports the evidence put forward by Koijen, as this relation is unlikely to hold if government intervention is not the leading explanation of the medical innovation premium. Given that these findings hold, the results have some interesting implications.
This research thus implies that when studying returns in the healthcare industry, one has to account for the additional risk the healthcare industries bears as a result of government intervention. Furthermore, these risks and thus returns vary per region, depending on how much potential a government has to intervene. Seemingly excessive returns in the healthcare industry are thus a result of accounting practices and risk of government intervention. Returns in the healthcare industry might therefore appear excessive, while in fact they are justified by additional risks, as this study shows.
5.3 Suggestions for future research
As model fit has an influence on healthcare returns, and is relatively limited in this study, future studies might first want to identify methods to improve fit. Therefore, the first suggestion would be to healthcare study risk loadings in the EU. Secondly, this research paves the way for future research in government intervention risk. Government intervention risk was in this study measured by government intervention potential through the share of publicly funded healthcare. However, other possibly more exact proxies are available. As the premium on the healthcare industry is significant, an easier and more accurate measure can be supportive in the discussion surrounding healthcare returns.
6 Conclusion
7 References
Anderson, R., 2014. Pharmaceutical industry gets high on fat profits. BBC News.
Angell, M., 2004. Excess in the pharmaceutical industry. Can. Med. Assoc. J. 171, 1451–1453. doi:10.1503/cmaj.1041594
Baber, W.R., Kang, S.-H., 1996. Estimates of economic rates of return for the U.S. pharmaceutical industry, 1976–1987. J. Account. Public Policy 15, 327–346. doi:10.1016/S0278-4254(96)00002-6
Breusch, T.S., Pagan, A.R., 1980. The Lagrange Multiplier Test and its Applications to Model Specification in Econometrics. Rev. Econ. Stud. 47, 239–253. doi:10.2307/2297111 Brooks, C., 2014. Introductory Econometrics for Finance. Cambridge University Press. Carhart, M.M., 1997. On Persistence in Mutual Fund Performance. J. Finance 52, 57–82.
doi:10.1111/j.1540-6261.1997.tb03808.x
Comanor, W.S., 1986. The Political Economy of the Pharmaceutical Industry. J. Econ. Lit. 24, 1178–1217.
Dickey, D.A., Fuller, W.A., 1979. Distribution of the Estimators for Autoregressive Time Series with a Unit Root. J. Am. Stat. Assoc. 74, 427–431. doi:10.1080/01621459.1979.10482531
DiMasi, J.A., Hansen, R.W., Grabowski, H.G., 2003. The price of innovation: new estimates of drug development costs. J. Health Econ. 22, 151–185. doi:10.1016/S0167-6296(02)00126-1
Egan, M., Philipson, T.J., 2013. International Health Economics (Working Paper No. 19280). National Bureau of Economic Research.
Eurostat, 2016. Healthcare expenditure statistics - Statistics Explained [WWW Document].
URL
http://ec.europa.eu/eurostat/statistics-explained/index.php/Healthcare_expenditure_statistics (accessed 1.6.17).
Fama, E.F., French, K.R., 2012. Size, value, and momentum in international stock returns. J. Financ. Econ. 105, 457–472. doi:10.1016/j.jfineco.2012.05.011
Fama, E.F., French, K.R., 2011. Size, Value, and Momentum in International Stock Returns (SSRN Scholarly Paper No. ID 1720139). Social Science Research Network, Rochester, NY.
Fama, E.F., French, K.R., 1997. Industry costs of equity. J. Financ. Econ. 43, 153–193. doi:10.1016/S0304-405X(96)00896-3
Fama, E.F., French, K.R., 1993. Common risk factors in the returns on stocks and bonds. J. Financ. Econ. 33, 3–56. doi:10.1016/0304-405X(93)90023-5
Herdman, R., 1993. Pharmaceutical R&D: Costs, Risks, and Rewards. U.S. Office for technology assessment.
Koijen, R.S.J., Philipson, T.J., Uhlig, H., 2016. Financial Health Economics. Econometrica 84, 195–242. doi:10.3982/ECTA11182
Medicare, C. for, Baltimore, M.S. 7500 S.B., Usa, M., 2014. Centre for medicaid and medicare services (CMS) [WWW Document]. URL https://www.cms.gov/Research-
Statistics-Data-and-Systems/Statistics-Trends-and-Reports/NationalHealthExpendData/ (accessed 12.11.16).
Murphy, K.M., Topel, R.H., 2005. The Value of Health and Longevity (Working Paper No. 11405). National Bureau of Economic Research.
Newey, W.K., West, K.D., 1986. A Simple, Positive Semi-Definite, Heteroskedasticity and AutocorrelationConsistent Covariance Matrix (Working Paper No. 55). National Bureau of Economic Research.
Phelps, C.E., 2016. Health Economics. Routledge.
Rice, J.A., 2006. Mathematical Statistics and Data Analysis. Cengage Learning.
Scherer, F.M., 2004. The pharmaceutical industry-Prices and progress. N. Engl. J. Med. 351, 927–932.
Scherer, F.M., 1996. Industry structure, strategy, and public policy. Prentice Hall.
Scherer, F.M., 1993. Pricing, Profits, and Technological Progress in the Pharmaceutical Industry. J. Econ. Perspect. 7, 97–115.
Sharpe, W.F., 1964. Capital Asset Prices: A Theory of Market Equilibrium Under Conditions of Risk*. J. Finance 19, 425–442. doi:10.1111/j.1540-6261.1964.tb02865.x
Smith, A., Nicholson, J.S., 1887. An Inquiry Into the Nature and Causes of the Wealth of Nations... T. Nelson and Sons.
The next act in healthcare private equity | McKinsey & Company [WWW Document], n.d. URL http://www.mckinsey.com/industries/private-equity-and-principal-investors/our-insights/the-next-act-in-healthcare-private-equity (accessed 12.11.16). Troy, T.D., 2015. How the Government as a Payer Shapes the Health Care Marketplace.
American Health Policy Institute.
White, H., 1980. A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica 48, 817–838. doi:10.2307/1912934
8 Appendix
Table A.1
Descriptive statistics US and EU returns vs ETF benchmarks
The table provides an overview of the descriptive statistics for the five return series. All statistics are computed with monthly returns. Reported are the maximum and minimum one-month returns observed in the sample, the mean average excess return (monthly), the (monthly) standard deviation of each factor, excess kurtosis and skewness. The sample returns are for the period 2002:07 to 2015:07
Portfolio Mean Median Maximum Minimum Standard deviation
Skewness Kurtosis Observa-tions EU healthcare 1.18 1.60 12.01 -9.32 3.66 -0.54 0.69 157 EU Healthcare ETF 0.76 0.96 10.22 -10.68 3.53 -0.22 0.62 157 US healthcare 0.94 1.23 9.00 -10.96 3.62 -0.36 0.32 157 US healthcare ETF 0.92 1.20 9.40 -12.57 3.62 -0.56 1.16 157 Global ETF 0.81 0.82 8.29 -12.31 3.47 -0.53 1.12 157 Table A.2
Regression results benchmarks
The table provides the results of CAPM and Fama-French regressions on healthcare returns. The first two coloums are the CAPM coefficients, the last five show the Fama-French regression coefficients. Five portfolios are reviewed, EU and US healthcare portfolios as discussed in the data section and EU US and global market-tracking ETFs. All statistics are computed with monthly returns. The EU results are estimated using european market returns and factors, the US results are esitmated using north american market returns and factors, while in for the global results global market returns and factors are applied. β 's are the market portfolio coefficients, γ, λ, φ, the SMB HML and WML coefficients respectively. The sample is from 2002.07-2015.07. All results are monthly parameters, and HAC corrected using the Newey-west method. Portfolio CAPM Fama-French
Table A.3
The table depicts the weight of the firms present in the portfolio and the ETF
Firm Country Holding
ETF Portfolio NOVARTIS AG Switzerland 19.6% 16.0% ROCHE HOLDING PAR AG Switzerland 15.2% 13.9%
SANOFI SA France 9.6% 8.4%
GLAXOSMITHKLINE PLC UK 9.6% 7.9%
ASTRAZENECA PLC UK 7.2% 5.8%
NOVO NORDISK CLASS B Denmark 6.8% 5.7%
SHIRE PLC UK 5.5% 4.6%
FRESENIUS SE AND CO KGAA Germany 3.2% 2.6% ESSILOR INTERNATIONAL SA France 2.5% 2.0% ACTELION LTD Switzerland 2.5% 1.4% FRESENIUS MEDICAL CARE AG Germany 1.8% 2.1%
MERCK KGAA Germany 1.4% 1.1%
SMITH & NEPHEW PLC UK 1.4% 1.0%
GENMAB Denmark 1.1% 0.9%
LONZA GROUP AG Switzerland 1.0% 0.8%
UCB SA Belgium 0.9% 1.1%
COLOPLAST B Denmark 0.8% 1.1%
NOVOZYMES B Denmark 0.8% 0.8%
Table A.4
Regional regression results, 10 year periods
The table provides the results of CAPM and Fama-French regressions on healthcare returns of different european regions, for two 10 year time periods, 1995 to 2015. The first two coloums are the CAPM coefficients, the last five show the Fama-French regression coefficients. Five regions are reviewed, which are discussed in the data section. All statistics are computed with monthly returns, and European market returns and factors are applied. β 's are the market portfolio coefficients, γ, λ, φ, the SMB HML and WML coefficients respectively. The regional healthcare portfolios are from 1995-2015. All results are monthly parameters, and HAC corrected using the Newey-west method.
Table A.5
Regional regression results, 5 year periods
The table provides the results of CAPM and Fama-French regressions on healthcare returns of different european regions, for four 5 year time periods, 1995 to 2015. The first two coloums are the CAPM coefficients, the last five show the Fama-French regression coefficients. Five regions are reviewed, which are discussed in the data section. All statistics are computed with monthly returns, and European market returns and factors are applied. β 's are the market portfolio coefficients, γ, λ, φ, the SMB HML and WML coefficients respectively. The regional healthcare portfolios are from 1995-2015. All results are monthly parameters, and HAC corrected using the Newey-west method.
Table A.6
Healthcare regression results, all HC indicators
The table shows the result of the premiums found in previous models regressed on the different healthcare indicators. Z is the intercept of the regression, Beta the slope and the R2 provides a measure of fit for the models. The value in brackets is the t-statistic of the coefficient. The value below the R2 provides the adjusted R2. The premiums used are over the period of 1995-2015, on the average health care indicators over the same period.
Health expenditure indicator CAPM premium Fama- french premium
z (t-statistic) β (t-statistic) R2 (adjusted) z (t-statistic) β (t-statistic) R2 (adjusted)
Per capita US$ 1.54 0.00 0.15 1.95 0.00 0.18 (2.95) (-0.94) -0.02 (3.38) (-1.04) 0.01 per capita PPP 1.74 0.00 0.33 2.14 0.00 0.34 (3.88) (-1.57) 0.20 (4.26) (-1.59) 0.20 Total (% of GDP) 2.20 -0.11 0.30 2.65 -0.12 0.30 (2.79) (-1.45) 0.16 (2.98) (-1.45) 0.15 Private (% of GDP) 1.50 -0.13 0.55 1.87 -0.15 0.58 (7.24) (-2.49) 0.46 (8.23) (-2.61) 0.49 Public (% of GDP) -1.62 0.37 0.50 -1.87 0.44 0.58 (-1.34) (2.25) 0.40 (-1.49) (2.61) 0.49
Public (% of government expenditure) 3.76 -0.16 0.41 3.72 -0.14 0.24
(2.59) (-1.86) 0.29 (2.00) (-1.26) 0.09
Public (% of total health expenditure) -0.68 0.02 0.64 -0.63 0.03 0.67
