Introducing our Healthcare Demand Model

Print Friendly
Share on LinkedIn0Tweet about this on Twitter0Share on Facebook0

Richard Evans / Scott Hinds

212.531.6101 / .6102 /

February 16, 2010

Introducing our Healthcare Demand Model

  • Using 8 explanatory variables that fall into five broad categories (demographics, macroeconomic conditions, pricing, health payor mix and healthcare-related capital investment), our model explains the majority of growth in total US healthcare spending across a nearly 40 year timeframe. Each descriptive economic variable is significant at the 99% level (p-value <0.01) or greater. The differences between our model’s predictions and actual values are appropriately small – falling within the reasonably narrow limits of +/- 3 percent in each year – and independent.
  • We expect 4.8% real growth in US healthcare spending over the period 2010-2019. Growth is back-loaded; we expect 4.0 percent real growth from 2010-2014, and 5.3% growth from 2015-2019.
  • Underlying our 2010-2019 growth expectation are several moving parts; three are key. First, a gradual return to full(er) employment levels explains roughly 15-20% of growth through 2015, Second, age effects are particularly important in the coming decade, as the proportion of the population aged 65 accelerates in 2011-2019 off of a base of very low growth from 2000-2007. Age mix accounts for nearly a quarter of growth across the ten year forecast period. Third, for structural reasons (remarkable price inelasticity) we expect healthcare pricing to continue to be inflationary – as has been the case for nearly all of the 48 years in our model’s history.
  • Our expectation of health spending acceleration to (even greater) GDP-plus levels contrasts with healthcare expectations and valuations generally, and with health insurers’ particularly. On a capitalization weighted basis, healthcare companies of >= $1B market cap trade at a 7 percent (2011 PE) to 10 percent (2010 PE) discount to the SP500; insurers trade at a 25 to 30 percent discount.
  • Consensus revenue expectations for insurers imply flat revenue in the face of both rising mid-term employment and accelerating health spending. We believe insurers are pricing at or ahead of trend, and so see consensus’ forecast as far too conservative. We continue to view insurers as the most under-valued sub-sector in healthcare.
  • To hear more about our views on healthcare, please join us for a brief conference call at 11 a.m. EST on Thursday, February the 18th. You can sign up at, or by contacting any of us at the e-mail addresses or phone numbers provided in the footer on each subsequent page of this call.

Modeling Aggregate Health Care Demand


This call introduces our model of total U.S. healthcare consumption over the mid- to long-term. This is a top-down model which describes the historical relationship between CPI-U deflated (real) health expenditures and a range of economic and demographic measures. In that total healthcare demand is very much influenced by the broader macroeconomic context; our modeling of these influences allows us to translate ‘standardized’ (e.g. CBO, BLS, consensus) macroeconomic assumptions into meaningful forecasts of total healthcare demand. In its current form, the model produces annual forecasts over a ten-year horizon.

The total demand model is the first in a suite of coordinated models that form the fundamental basis of our franchise. From here, we will add quarterly forecasts of total healthcare demand over a rolling 12 month horizon; and, a corresponding suite of bottom-up demand and earnings models for each healthcare sub-sector.

We plainly see the opportunity to simply add-up the revenue totals of our bottoms-up sub-sector models, but choose also to produce a top-down total demand forecast because of: (1) the historically tight relationships between healthcare consumption and the broader economy; (2) the importance of tracking total health cost pressures on government, industry, and consumers; and (3) our view that a rigorous top-down projection is an essential framework that informs our bottoms-up subsector models.

  1. Historical relevance. Using data over the past 40+ years, we have developed a statistically sound and analytically robust model that explains the vast majority of total healthcare spending as a function of various demographic, macroeconomic and sector-specific dynamics. The quality of the model plainly infers that the broader economic setting does much to determine total health spending; this in turn argues that a top-down model of total healthcare demand is essential to investing across healthcare on an informed basis.
  2. Economic relevance. On previous calls we have detailed the friction between health cost growth and the performance of the broader economy[1]. Given the apparent political failure of immediate health reform efforts, health costs are likely to reach a point at which fiscal, consumer, and/or industrial pressures compel yet another attempt at reform. In short, we see high odds of a health-cost-related economic crisis, and see our total demand model as a useful means of handicapping when such a crisis point might be reached, and what form(s) such a crisis might take.
  3. Better sub-sector forecasts. In a recent call we showed that consensus sub-sector forecasts presently fail to sum to a reasonable total healthcare view[2] – which creates substantial disconnects in valuations across several sub-sectors. To our minds, an aggregate top-down view helps us find and capitalize on these disparities, and in a similar fashion forces our own sub-sector forecasts to add to a meaningful total demand estimate.

Model results, associated logic and caveats

Our log-linear regression model uses CPI-U deflated National Health Expenditures (NHE), as published by the Centers for Medicare and Medicaid Services, as the dependent variable. We utilize eight explanatory variables that fall into five broad categories: demographics, macroeconomic conditions, pricing, health payor mix and healthcare-related capital investment. We have specifically sought to eliminate sources of autocorrelation and collinearity, arriving at a construct that offers considerable explanatory power. In the current version of our total demand model, all descriptive economic variables are significant at the 99% level (p-value <0.01) or greater. And, the difference between our model’s predictions and actual values are random, falling within the reasonably narrow limits of +/- 3 percent in each year. Exhibit 1 illustrates how closely our model tracks with actual real NHE since 1970 (and how closely projections track with OACT projections through 2019)[3].

Demographics. Population growth and shifting age / sex mix are of course key contributors to changing total healthcare demand, and this relationship is reflected in our regressions. We estimate that demographics contribute approximately 1.4 percentage points of health consumption growth per year, two-thirds from population growth, one-third from shifting population mix.[4]

Note that the relationship between broad demographic variables and total demand is both glacial and predictable – and by extension that such a broad definition can do very little to help us find inaccurate forecasts and associated mis-valuations across shorter investment timeframes. However, when we consider demographics more narrowly, and particularly in conjunction with other relevant variables, we can find shorter-term signal value — for example modeling changes to income and/or health coverage across distinct age / sex groups can be essential to estimating near-term health spending during economic inflections.

Macroeconomic conditions (including inflation). Economic theory suggests that health consumption is strongly pro-cyclical, and our analysis supports this notion from several different perspectives. For example we find a strong negative relationship between health expenditures and unemployment, and of course recognize that this relationship likely reflects weak macroeconomic conditions as well as the important link between jobs and private health insurance. As with demographics, we can find added near-term signal value beyond these obvious bigger-picture relationships. For example, the effect of employment on demand appears to work in phases; as we’ve previously shown fear-of-job-loss actually appears to accelerate demand[5] rather sharply in an initial phase that precedes a more prolonged phase of lowered consumption that one might naturally expect in a slower economy with higher unemployment. This pattern of employment effect is supported by the regressions on which our total demand model is based.

Pricing. The most fundamental rule of economics also instructs that we should find a negative relationship between prices and demand. Again, this relationship is borne out in our regressions. While we are very cautious about attempting to make comparisons between our model and similar analyses conducted by CMS’ Office of the Actuary (OACT) for several reasons (more on that later in this call), not least which is that we using different measures of consumption, we note that we find very similar levels of aggregate price (in)elasticity of approximately -0.46.[6] In our own related work, we again find added near-term signal value at greater levels of resolution – beyond the simple notion of higher real system-wide prices having a negative effect on system-wide total unit consumption, we find that elasticity effects can vary greatly according to consumers’ levels of exposure to costs (e.g. out-of-pocket burdens, see immediately below), and to the nature of the product or service being consumed (e.g. discretionary v. mandatory).

Payor mix and consumer burden. Since the introduction of Medicare and Medicaid in 1966, the U.S. healthcare market has evolved considerably in terms of payor mix (a dynamic which has only accelerated in recent years as the aggregate burden of public payors via entitlement programs has swelled to 35% of NHE). At the same time – particularly when considered in the context of personal income – private expenditures for healthcare have also grown. We include in our model variables which capture the availability of health insurance, its source, and personal cost burden relative to income. Not surprisingly, the relationship between the availability and generosity of third-party insurance and total health spending is overwhelmingly positive, and the magnitude of effect is considerable. For example, our analysis of recent healthcare consumption from MEPS[7] data suggests that an insured adult (19-44) in good health consumes more than three times the healthcare of a comparable person who is uninsured. We found similarly large relative demand differences across age, income, and health status subpopulations.[8] Also, as would be expected, these relationships show very little time lag; accordingly we expect to be able to refine both our short- and long-term total demand forecasts to accommodate changing assumptions regarding personal income and the quality / availability of health insurance coverage.

Health capital investment. We have looked at a number of measures of healthcare-related capital investment (e.g. structures and equipment; research & development) to understand how investment today impacts consumption tomorrow. The results overwhelmingly support the hypothesis that incremental healthcare consumption does follow such investments. Obviously these relationships play out across long timeframes; our regressions are geared to the potentially longer (e.g. R&D) and shorter (e.g. hospital capacity) lags between spending and associated consumption.

As with any projections based upon historical regressions, the driving assumption is always that relationships are constant through time. The 40+ year horizon over which we’ve looked back gives us some confidence in the long-run relevance of the coefficients we’ve calculated, but of course there is always the risk that the system could face a shock that impacts a year, or series of years. These shocks are easy to deal with in hindsight and our model has been adjusted for outlier years with little difficulty, but the impact of such shocks on our projections is inherently difficult to predict.

Projections and caveats

For 2010 – 2019, our baseline aggregate health consumption model calls for annual growth in real health care expenditures of 4.8 percent – precisely in line with recently issued OACT projections. This compares with a 5.7 percent growth rate over the model’s 48 year history and a 3.4 percent rate over the past five years (Exhibit 2). Thus in very simple terms we project an acceleration of health spending from present economically suppressed levels, to a level of growth that is more in-line with longer-term historic rates.

Our 4.8 percent real growth expectation for 2010 – 2019 is back-loaded; we expect 4.0 percent real health spending growth in 2010 – 2014, and 5.3 percent from 2015 – 2019 (Exhibits 2, 3). Generally speaking, these numbers represent a fairly constant rate of inflation in per-unit healthcare prices, compounded over the period by increasing per-capita unit consumption as income and employment levels gradually normalize, and as an increasing proportion of the population becomes Medicare eligible.

Let’s take each of these moving parts in turn, beginning with the inflation assumption. Healthcare pricing has been inflationary relative to prices in the broader economy in almost every year of our model’s 48 year history (Exhibit 4). Based on a regression of the historical relationship between general consumer price inflation and medical cost inflation, we assume that healthcare prices grow 2.4% faster than CPI-U. We’ve written fairly extensively on why healthcare pricing is inflationary[9]; for the moment we’ll confine our arguments to the notion that price inflation is a predictable feature of an economic system having substantial inelasticity – if prices rise and unit consumption remains strong, producers have every reason to raise price further. Aggregate healthcare demand is remarkably inelastic; estimates of elasticity range between roughly 0.17 and 0.5. Recognize that this aggregate number changes as incomes rise or fall, and/or as people fall in and out of health insurance coverage. However, in that the majority of consumers are covered, even during sharp downturns, aggregate demand is effectively always inelastic – i.e. always permissive of real price inflation. As such we feel very comfortable with the assumption that healthcare inflation continues.

As economic growth returns and unemployment falls, our model predicts an associated acceleration in health spending. The effect is somewhat gradual; CBO projects that unemployment will remain above 9 percent until at least 2012, and we rely on these projections. CBO projections also assume employment reaches steady state levels around 2015, at which point the accelerating effect of jobs recovery on our estimate of healthcare demand is fully played out. Very roughly, the employment effect accounts for about 15% – 20% of the estimated growth in real healthcare demand from now until 2015.

Age mix, and in particular Medicare eligibility plays a more significant role. The percentage of the population of age 65 or greater held steady from roughly 2000 to 2007, began growing again in 2007, and accelerates sharply from 2011 to the end of the forecast period (Exhibit 5). The near-overlap of this no-growth to rapid-growth pattern in age mix with a comparably shaped employment recovery does much to accelerate our estimates of total health spending. As we model population age effects on historic demand, not surprisingly the transition from less than 65 years of age to greater than or equal to 65 years of age has considerable explanatory power, as it serves as a proxy for both biologic and economic shifts in demand. Biologically, the percent of the population above 65 serves as a proxy for average age of the broader population, which of course serves as a predictor of per-capita need for care. Economically, the transition from less than 65 to greater than or equal to 65 marks a transition from lack of insurance to Medicare coverage for some portion of the population – and this effect on demand arguably is greater than normal during periods of economic weakness. All in, shifting age mix of the population is responsible for roughly a quarter of real demand growth during the forecast period.

Summary and conclusions

To summarize, the 2010 – 2019 total demand picture is one of continued real growth in healthcare pricing, with unit demand driven by a gradual return to more normal levels of employment, as well as the forecast era’s unique pattern of very low age-mix effect followed quickly by a burst of substantial age-mix effect. The risks to our forecasts lay predominantly with our assumptions on real pricing and employment; there is very little risk in the age-mix effect assumption, as both the population age-mix and the effect of age on consumption are very well characterized. As one would suspect given the predominant role of healthcare price inflation in total health spending growth, our model is very sensitive to the assumed health cost inflation rate – an assumption of 1 percent greater or lesser health cost inflation raises or lowers the annual cost growth assumption by roughly 72 bp. Obviously the effect compounds over time, and can produce substantial differences in real estimated demand across the 10-year forecast timeframe. The model is similarly sensitive to assumptions regarding employment levels; for each additional 1 percent employment we assume, real health spending increases by roughly 70 bp. This illustrates the importance of the timing of an economic recovery to the real health spending picture – though admittedly there is not the same compounding effect as we find for inflation.

Our top-down model contrasts with estimates and associated valuations for healthcare as a broad industry, and for the health insurance sub-sector in particular. In that health spending growth ties in large part to GDP growth and employment levels, one might reasonably argue that a general expectation of improving economic conditions would benefit both the healthcare industry and the broader SP500 group of companies to roughly comparable degrees. And, in that we expect both real healthcare inflation and age-mix effects to drive health spending growth more rapidly than growth in the broader economy – and thus perhaps as fast or faster than in the SP500 – one might further argue that healthcare revenue expectations and valuations should broadly reflect the likelihood of GDP-plus top-line growth. Plainly they don’t; capitalization weighted valuations of the greater than $1B market cap companies that we include in our coverage universe are 7 percent (2011 PE) to 10 percent (2010 PE) below the SP500.

The discount on health insurers is much more substantial; here we find valuations that are 25 to 30 percent below those in the broader SP500, and revenue expectations that are well below both our top-down estimates, and the bottom-up sum of consensus expectations for the producers that represent insurers’ claims costs. We note that large commercial insurers, whose books of business are built largely around the employer sponsored insurance (ESI) market, obviously are beneficially levered to the improving employment levels that form the early foundations of a return to faster health spending growth. And, we emphasize that the cost-plus nature of the present health insurance model benefits from real health pricing growth, as long as the real rates of price growth are reasonably stable. Our sense is that insurers are pricing at or ahead of trend; accordingly we expect insurers to see the combined benefits of both improving employment and continued real growth in healthcare prices – all of which argues for a valuation at least on par with the broader healthcare industry.

Comparisons to OACT projections

As we’ve referenced already, OACT performs, and annually publishes, projections based on a model very similar to what we’ve constructed and described here. One could reasonably ask why, if there are OACT produced projections that extend out a decade or more available, are we going through the same exercise? We view the OACT process as theoretically and methodologically sound, but believe that having our own model offers investors at least two advantages. First, OACT’s forecasting assumptions are largely constrained to current law, whereas we obviously are able to make assumptions that are free of this constraint – a freedom that is particularly important as we are likely to be entering an era of many separate but meaningful changes to relevant laws. Second, OACT’s projections focus much more on the ten-year picture and changes thereto than on changes in demand across shorter timeframes. Plainly we care a great deal about the ten-year picture, but our jobs depend on finding incorrect forecasts and associated securities mis-valuations across much shorter timeframes – anywhere from three months to 2 or 3 years. Accordingly, our model is built with an emphasis on explanatory variables that are both predictable in the ten-year sense and ‘reactive’ in the ‘investable’ timeframe sense. Referring back to our characterization of explanatory variables this might be more clear – recall that we frame demographics in both long-term-descriptive (e.g. total population growth and age-sex mix) and more ‘reactive’ ways (changes to income and insurance by age-sex grouping), and frame macroeconomics (long term relationship between income and spending, short-term relationship between changes to employment levels and immediate healthcare demand) similarly.

What happens next

In the coming weeks and months we will roll-out sub-sector specific models, beginning with pharmaceuticals in March. These subsector models sit in a unique position, as they are both the parts that sum to total health expenditures, and the sum of parts represented by individual companies. They are, in effect, both the bottom-up components of the model described in this call; and they are the top-down frameworks which guide company-level modeling.

  1. See, for example, our October 6, 2009 call “The New Abnormal: How Health Costs Derail Our Return to Historic Notions of Fiscal Balance”
  2. February 1, 2010, “Mopping Up Residual Reform Risks; Why Consensus Expectations at the Sub-Sector Level Don’t

    Add Up”

  3. Our dataset extends from 1960, however our estimates of total demand begin somewhat later, as certain of our explanatory variables are lagged.
  4. This is close to the 1.3 percent per year health consumption growth projected by the CMS Office of the Actuary (OACT): Truffer, C. et al “Health Spending Projections Through 2019: The Recession’s Impact Continues.” Health Affairs 29, No 3 (2010).
  5. See for example our call from January 12, 2010, “GMs Too High for Pharma, Too Low for HMOs…”
  6. For more information, and an excellent explanation of discrepancies between macro- and micro-level price elasticities please see
  7. Medical Expenditure Panel Survey (
  8. By explicitly controlling for self-assessed health status we have tried to minimize the self-selection bias whereby sicker people, who consume more healthcare, choose insurance while the healthier forego it.
  9. See Evans “Health and Capital” at; see also Sector & Sovereigns’ “The Political Economics and Investment Relevance of American Health Reform” Aug. 18, 2009; and also “The New Abnormal: How Health Costs Derail Our Return to Historic Notions of Fiscal Balance”, Oct. 6, 2009, both at:
Print Friendly