Why Insurers Work in a Recovery; Introducing our Model of Private Insurance Premiums and MLRs

Print Friendly
Share on LinkedIn0Tweet about this on Twitter0Share on Facebook0

Richard Evans



April 5, 2010

Why Insurers Work in a Recovery; Introducing our Model of Private Insurance Premiums and MLRs

  • Insurers are better levered to economic recovery than other healthcare subsectors. Health insurance demand is more elastic than healthcare demand, meaning premiums accelerate faster than healthcare product and service sales. And job gains boost enrollment, growing total premiums, and lowering medical loss ratios (MLR) by improving risk pools. Despite this, on the doorstep of a widely anticipated recovery, insurers trade at a 20+% discount (PE) to healthcare, and a 25+% discount (PE) to the SP500
  • Our MLR model correctly predicts the timing of 80 percent of MLR peaks and troughs since 1960; 93 percent if we include predictions that were one year early. Troughs occur when employment is growing and/or medical inflation is decelerating; peaks come when employment is falling and medical prices are accelerating. Our model suggests we are at or near peak MLR (we believe we are past an ’09 peak); further reason to favor health insurers, particularly at current discounts
  • Of note, the fact that 40+ years of MLR cyclicality is almost wholly explained by shifts in employment and/or medical inflation suggests that a classic explanation for MLR cyclicality − price competition among insurers − is incorrect
  • Before considering enrollment gains from 2014 reform provisions, we project 3.6 percent real premium growth from ’09 – ’14, and 3.4 percent real growth from ’15 to ’19. On top of this, we expect reform-associated enrollment gains will expand total private insurance premiums 5 to 10 percent by 2019
  • From an ’09 peak, we expect MLR’s to decline by an aggregate 1.3 percent by 2019 as employment gains improve the risk pool, and as insurers price in the incremental risks associated with pending underwriting restrictions

Summary and conclusions

Today’s call introduces the first of our sub-sector models: private health insurance. Our model consists of two distinct parts: premiums collected, and the medical loss ratio (MLR). We recently introduced our model of total US healthcare demand[1], which among other things showed that demand for healthcare is very sensitive to economic conditions. With our premiums model, we find that demand for insurance is even more sensitive – private insurance is dominated by employer sponsored insurance (ESI); as jobs are lost, so is eligibility for ESI. And, we find that demand for private insurance is far more elastic than demand for healthcare – insurance often is viewed as a discretionary purchase; healthcare often is not. It follows that demand for private insurance falls – and rises – with the economic cycle more dramatically than demand for healthcare, which implies that health insurers should tend to outperform other health sub-sectors during economic recovery – particularly if the recovery includes net employment gains, and particularly if insurers are priced at a discount to their healthcare peers in advance of such recovery.

With our MLR model, obviously we find that falling employment worsens the risk pool (MLRs rise) and rising employment improves the risk pool (MLRs fall) – thus further leveraging health insurers’ earnings power to economic (and employment) recovery, and further implying that insurers may be preferred relative to other healthcare sub-sectors in advance of, and during, recovery. Less obvious is that we can identify 80% of all peaks and troughs in the year that they occur, and 93% within one year prior, largely through two variables – employment, and medical inflation. MLR troughs tend to occur in periods characterized by medical price deceleration and/or improving employment; peaks are almost all in periods with accelerating medical pricing and deteriorating employment. To be clear, acceleration or deceleration in the rates of inflation or job loss/gain matter more than the rates themselves. Note what this implies about the limited intensity of competition among insurers – we can predict MLR peaks and troughs almost wholly as a function of unexpected changes (i.e., acceleration or deceleration) in costs (medical inflation) or the quality of the risk pool (employment gains / losses). This is to say, we find no evidence of the usual competitive determinants of service-industry pricing cycles.

We expect premium growth to accelerate as the economy enters recovery; before adding in enrollment gains due to health reform-related subsidies, we model 3.6 percent real premium growth from ’09 – ’14, and 3.4 percent real premium growth from ’15 – ’19. Note that these estimates are below our estimates of national health spending growth, as the employer-dominated private insurance market is less levered to population age effects than the national market as a whole, and as we anticipate continued cost-shifting to enrollees. Note also that unlike our back-loaded estimate of national health spending growth, private insurance premium growth is (very slightly) front-loaded, this reflects the employment and elasticity effects, explained immediately above, that lever premiums to economic recovery. Anticipated enrollment gains as reform-related subsidies unfold increase our estimate of total private insurance premiums by 5 – 10 percent[2].

We believe that 2009 was an MLR peak, though our model argues the peak occurs in 2010. We stand by the 2009 prediction – our model does not use flu-related regressors (flu effects probably were much larger in 2009 than they will be in 2010); and, our model’s use of annual changes in employment may cause it to under-appreciate the MLR effect for years in which very rapid job losses were isolated to certain periods of the year – as occurred in 2009. Consistent with our prediction on MLR trend, the model projects falling MLR’s beyond 2010 (falling an aggregate 1.3 percent from 2009 to 2019), largely as a result of expected gains in employment in the near-term, and as an accommodation of greater underwriting risks as reforms unfold in the mid- to longer-term.

Importantly, our inputs assume that the percent of employees’ total compensation packages dedicated to health insurance grows more slowly than in the past – and more slowly than the excess of health inflation over wage inflation. This assumption is conservative, as it has the effect of lowering premium growth and raising MLR’s. Our projections are very sensitive to this assumption; accordingly we see room for upside to both premiums and gross profits.

As is the case with our aggregate health consumption model, our private insurance premiums and MLR models contrast with estimates and associated valuations for the health insurance sub-sector. We re-emphasize that health spending growth generally, and private insurers’ premium growth and gross margin particularly, tie in large part to GDP growth and employment levels; and, that a general expectation of improving employment conditions should benefit health insurers more than either their healthcare peers or the broader SP500 group of companies. With insurers trading at substantial discounts to both healthcare and the SP500 – in advance of an economic recovery – we continue to find the sub-sector meaningfully undervalued.

Health premiums model results and associated logic

Our log-linear regression model of private insurance premiums uses the CPI-U deflated private insurance expenditures portion of National Health Expenditures (NHE), as published by the Centers for Medicare and Medicaid Services, as the dependent variable. We utilize six explanatory variables plus identifiers for several outlier periods; these variables fall into five broad categories: demographics, macroeconomic conditions, pricing, health payor mix/consumer health cost burden and healthcare-related capital investment. As usual we have specifically sought to eliminate sources of autocorrelation and collinearity. In the current version of our private insurance premiums model, all descriptive economic variables are significant above the 96% 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 +/- 5 percent in each year[3]. Exhibit 1 illustrates how closely our model tracks with actual real NHE since 1970 (and how closely projections track with OACT projections through 2019). Note also that the shape of the real health insurance premium curve is largely similar to total health consumption (Exhibit 2), with long-run, smooth GDP-plus growth.

Demographics. Population growth and shifting age / sex mix are of course key drivers of total insurance premiums, which is borne out in our regressions. Of interest, even though the private insurance is predominantly driven by employer sponsored insurance, and employees are predominantly younger than 65 years of age, the over-65 population still weighs heavily in our estimates of premium growth — 59% of the over-65 population were covered by some form of private insurance in 2008 v. 67% of under-65’s[4]. Our demographic descriptor is a composite of population growth and shifting population age, weighted to those age groups that most heavily influence change in spending on private insurance. We recognize that the percent of retirees with private health insurance from a prior employer is slowly but steadily declining, and so realize that historic age-weightings may tend to over-estimate premium growth in our longer-term forecasts.

Macroeconomic conditions. Clearly we would expect health insurance premiums to be strongly pro-cyclical, since premiums are directly linked to employment through employer sponsored insurance (ESI) plans; and, since premiums are a discretionary purchase made in relation to income, which of course falls when jobs are lost. I.e., the decision of whether to purchase health insurance is much more income and employment-sensitive (and price-sensitive) than the decision of whether to purchase health care. While an unemployed person may elect to discontinue paying insurance premiums, she may still consume a significant amount of healthcare if she has an (unplanned) medical emergency. Consistent with this we find a very strong negative relationship between health insurance premiums and unemployment, and estimate that the magnitude of impact of unemployment is three to four times as large for health premiums as for aggregate health consumption. Thus our model of total private insurance premiums is more sensitive to macro-economic conditions than our model of total health care consumption; and, the macroeconomic descriptors we rely on in the premiums model are heavily weighted to employment.

Pricing. Consistent with the preceding arguments, we find a negative relationship between premium prices and the decision to enroll in insurance; and, we find that the decision to purchase health insurance is significantly more price-sensitive than the decision to purchase health care. Personal illness may compel health spending even when the prices of health care are rising, i.e., the choice to consume often is not made freely. There need not be any such undesired consumption of insurance – if the price rises too quickly, enrollees can drop coverage. Note that this should remain true despite the passage of the health reform package, since penalties for not carrying insurance are very small relative to the costs of coverage; and, the absence of restriction on pre-existing conditions provide for enrollment even after an illness has begun[5]. Specifically, we find that a one percent increase in insurance premium prices is associated with an approximate two percent decrease in total premiums. We would note that this is not precisely a calculation of own-price elasticity, since CPI-M is not a measure of health insurance prices (Bureau of Labor Statistics has produced a health insurance-specific CPI since only 2005). Our conjecture is that over the past 40 years, the cost of health insurance has likely risen significantly faster than CPI-M. If this is true, then price elasticity is lower than the -2.0 we reference here.

Payor mix and consumer burden. Employers plainly dominate the private health insurance market ; since 1960, employers’ contributions to group health insurance premiums as a percentage of total (non-government mandated) compensation has grown more than six-fold – and since 1970 the annual growth rate of employer sponsored insurance has been nearly triple the growth rate of real wages (Exhibit 3). As more employers have offered insurance, (unsurprisingly) total premiums have likewise grown. Since employers offering insurance typically do not give employees the opportunity to opt-out and receive cash in lieu of benefits, and since employers are required to offer benefits to all employees, it is reasonable to assume that the ESI market has brought enrolled workers into private insurance who otherwise would not elect coverage. More crudely, it seems employers spend more on health insurance than employees would be willing to spend themselves. Over the past 40 years, this has had the effect of significantly increasing the premiums paid to private insurers, as well as the size and quality of the private insurance market risk pool. As the burden of payment for personal health consumption shifts increasingly to the individual in the form of out-of-pocket expenses, health premium growth declines, though at a less than dollar-for-dollar rate. Our descriptors for payor mix / consumer burden are heavily influenced by consumers’ out-of-pocket costs.

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 clearly support the hypothesis that such spending does increase real health premiums over longer time cycles, likely as a result of newer technologies replacing older standards of care at substantial premia. This relationship is very much as found in our model of total healthcare demand.

Outlier periods. Several eras emerge as outliers during which our regressors are relatively less descriptive of premium growth — for example, the emergence of HMO’s, President Clinton’s attempt at health reform, and the transition from more tightly controlled HMO’s to more loosely controlled network models such as preferred provider organizations (PPOs). Obviously given our beliefs that healthcare economics (even after the enactment of recent legislation) are on an unsustainable path and that additional large policy interventions ultimately are forthcoming, we fully anticipate additional disruptions to the premium (and other) model(s). Nevertheless our regressions across the 40+ years of modern health economic history help us find and understand the relationships that remain consistent across such disruptions; and, periods of disruption serve as useful models of premium behavior during policy shifts. For the purposes of our long-term model, we use identifiers of outlier periods as ‘dummy’ regressors in order to true the model across these policy shocks.

Medical loss ratio (MLR) model results and associated logic

The second (unique) component of our private health insurance model is the medical loss ratio (MLR)[6]; we use the year-on-year change in our imputed aggregate MLR as the dependent variable. We use six explanatory variables plus identifiers for several outlier periods; explanatory variables fall into four broad categories: macroeconomic conditions, pricing, health payor mix/consumer health cost burden and measures of MLR history. We again emphasize our efforts to eliminate sources of autocorrelation and collinearity. In the current version of our MLR model, five of the six economic variables are significant above 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 +/-130bp of the actual change in medical loss ratio in each year. Exhibit 4 illustrates how closely our model tracks with actual real NHE loss ratio values since 1970 (and how our projections depart from OACT’s projections through 2019). Note that the aggregate MLR has shown considerable cyclicality over the entire period, though in recent years that cyclical trend has attenuated as the MLR average has moved lower. We account for MLR cyclicality in the regression by introducing regressors that tend to mean-revert the MLR prediction toward a rolling mean.

Macroeconomic conditions. Logic calls for a strong positive correlation between changes in employment and changes in the MLR, which we find in our regressions. We’ve shown elsewhere that job loss has a double-barreled (negative) impact on medical claims – laid off workers self-select COBRA continuation of coverage, and are as a group less healthy and consume more healthcare than those who do not elect coverage; and even workers who do not elect COBRA coverage rationally anticipate their loss of insurance and over consume in the months leading up to their termination[7]. Thus unemployment not only lowers total premiums (see arguments in the previous section) by reducing the number of insured, but also weakens the risk pool by removing the healthiest (i.e., least likely to consume care) individuals, and accelerates spending by those still employed. To our minds, capturing the strength and timing of the relationship between insurer profitability and employment – and in particular the rate of change in employment — is critical to forecasting MLR. We find that a 1 percent increase in the unemployment rate, all else equal, translates into a 44bp increase in the medical loss ratio.

Pricing. As underlying health prices increase, so should the medical loss ratio, particularly if health prices are accelerating, and this is borne out in our analysis. We find a significant relationship between medical inflation and loss ratio changes, though the relationship is less than one-to-one. Insurers obviously price their coverage ahead, anticipating changes in benefit expenses, which accounts for the relatively modest 20bp increase in loss ratio that we find associated with a 1 percent increase in medical prices. Presumably unexpected price increases should have more of an effect, and we find evidence that this is true. The rate of change in medical inflation (i.e., crudely, the second derivative of CPI-M) is positively correlated with MLR. These apparent surprises (as approximated by year on year change in inflation) are very significant – the model suggests that a relatively modest 10% increase in the inflation rate (e.g., inflation moving from 5 percent to 5.5 percent) is associated with a roughly 15bp increase in MLR. Accordingly, our price regressor considers both the first and second derivatives of medical inflation.

Payor mix and consumer burden. As we argued for the premium model, health insurers have grown reliant on the ESI market for both total premiums paid in, and the quality of the risk pool. Thus greater employer spending on health insurance is strongly negatively related to medical loss ratio – that is, positively related to insurer gross margins. All the employment related arguments that we’ve made to this point remain applicable, and our results again support the relevant hypotheses: ESI increases the size of the private insurance risk pool, introduces lower cost enrollees, generates revenues through premium priced insurance products offered in lieu of wage increases, and so forth. Our descriptor for payor mix / consumer burden captures the level of employer spending on health benefits relative to other elements of compensation, as well as the consumers’ out-of-pocket burden.

Medical loss ratio history. The cyclicality of MLR over the past forty years is striking; there is a persistent tendency for the loss ratio to mean revert within a tight range around a rolling mean. There was exactly one period (1999 – 2003) during which medical loss ratio decreased for more than 3 consecutive periods; and one period (1995 – 1998) during which medical loss ratio increased for more than 3 consecutive periods. From 1970 – 2008 there were 15 such inflection points – 8 “troughs” (local minima in MLR); and 7 “peaks” (local maxima). The model accurately predicted 12 of these inflections in the year in which they occurred, and another 2 in the year prior to their occurrence for a +/- one period success rate of over 93% (Exhibit 5). Generally, the MLR troughs occurred in periods characterized by CPI-M deceleration and/or improving employment conditions. Conversely, the peaks are almost all in periods with accelerating pricing and deteriorating employment. This is certainly consistent with economic theory and suggests that beyond predicting absolute MLR (or year-on-year MLR changes), the model can help us understand when macro conditions imply a local max or min in MLR. Notably, our model suggests that 2010 MLR is a peak – modestly higher than 2009, based largely on forecasts of continued declines in employment. We recognize that this differs from our belief that 2009 was a peak, and believe this occurs for two reasons. First and most simply, the 2009 flu effect on MLR is much largely than is likely in 2010, and we do not use flu-related regressors in the MLR model. Second, consistent with our belief that rate of change in employment is far more important than the level of employment, our employment regressor is built more around changes in employment than absolute employment. However, our model uses annual observations, which may tend to under-emphasize the MLR effect of very intense but relatively brief (one or a few quarters rather than an entire year) periods of job loss[8]. Note that even a modest improvement in unemployment (from 9.3 percent in 2009 to 9 percent in 2010) would bring the model’s projection in line with our own view. Whether the peak occurred in 2009 or will occur in 2010, beyond this period the model is very consistent with our expectation of falling MLRs as employment recovers.

We believe that our MLR regressions speak to the nature of competition among private insurers. In most service industries we would expect competition among industry players to be a large contributor to cyclical pricing (in this case, MLR) patterns. The fact that we can describe peaks and troughs in the MLR cycle almost wholly as a function of the rate of change in medical inflation and/or employment suggests that competition is relatively weak. This is consistent with our pre-existing views on private insurance; we believe that competition is limited for structural reasons (e.g. the tendency of employers to choose only one underwriter, and for employees to be unable to choose another; the role that purchasing scale plays in accessing discounts; geographic restrictions on competition, and more); and, in that these structural features of the insurance markets are unchanged by health reform legislation, we see no immediate reason to alter our views on MLR dynamics as the reforms unfold.

Projections and caveats

Total premiums

For 2009 – 2019, our baseline (i.e., pre-reform) aggregate health insurance model calls for annual growth in real premiums of 3.5 percent – comparable to recently issued OACT projections of 3.8 percent annual growth. Of this, we attribute just over 1 percent to demographic shift (primarily population growth) and more than 2 percent to real (i.e., ex-CPI-U) medical price growth. Mix growth is a modest drag on total premiums of less than 0.5 percent per year – consistent with enrollees on net purchasing plans with a lower actuarial value. This 3.5 percent annual growth rate is lower than the 6.0 percent growth rate in real premiums from 1970 – 2008, but a reacceleration from the 2.1 percent growth rate over the past five years (Exhibit 6). Our projection of premium growth below the health consumption growth rate implies a shift in payor mix that is consistent with (1) the migration of the aging population into Medicare; and (2) the observed increased out-of-pocket cost burden borne by enrollees. Over the projection period, our 3.5 percent real growth expectation is notably smoother than the growth of aggregate health spending (which we found back loaded to the 2015 – 2019 period); we expect 3.6 percent real premium growth in 2009 – 2014, and before including the effect of reforms, 3.4 percent from 2015 – 2019.

Generally speaking, these projections are driven by a fairly constant rate of inflation in per-unit healthcare prices, gradual normalization of employment levels and critically, conservative assumptions about employers sponsorship of employees’ health insurance.

Healthcare pricing has been inflationary relative to prices in the broader economy in almost every year of our model’s 48 year history (Exhibit 7). 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. As such we feel very comfortable with the assumption that healthcare inflation continues. Given the stable inflationary dynamic we model – and the relatively robust price elasticity of health insurance we previously detailed – we view medical inflation in excess of CPI-U as a drag on total premiums of 4 to 5 percent per year.

As economic growth returns and unemployment falls, our model predicts an associated acceleration in health premiums. The model’s sensitivity to unemployment is illustrated by the projected dip in premiums in 2010 – 2011 which is a direct function of the elevated unemployment rate. Note that this window is the only portion of the projection period in which our model suggests a fundamentally different dynamic in real premiums than OACT (see Exhibit 1). Very simply, our model’s estimate of these years is low because it does not account for the effect of federal COBRA subsidies, which have kept millions of unemployed enrolled in private insurance. As a crude approximation of the impact, we recalculated these projections, replacing actual unemployment rates with values from 2008 (5.8% for full year 2008 – before the subsidy and the most devastating part of the economic crisis). This change brought our projection in-line with OACT, supporting the hypothesis that the policy shock — which the regression misses — is the primary cause of the discrepancy. Note that we rely on CBO projections that unemployment will remain above 9 percent until at least 2012 and reach steady state levels around 2015.

Our assumptions around employer sponsored insurance are the most critical drivers of total premium (and medical loss ratio) projections. Our model is extremely sensitive to group health insurance compensation, and we stress that if employers increase the percentage of total employee compensation via ESI it would be a strong positive for the private insurance market, likely to both increase real premiums and decrease medical loss ratio.

It is important to consider the evolution of employer sponsored health insurance which, in 1960, represented just over 1 percent of total compensation[10]. In 2008, health insurance premiums paid by employers were more than 7 percent of total employee compensation. From 1970 through 2008, while real wages grew by 2.0 percent per year and other benefits (life insurance, pensions) grew by 2.8 percent, real health insurance grew by 5.8 percent annually (see Exhibit 6). Note, however, that this relative growth discrepancy is both front-loaded and inconsistent. In fact, over the past 3 – 5 years, health insurance compensation growth has slowed considerably – actually declining over the 2005 – 2008 period, even as real wages grew modestly. For projection purposes, we hold relative health insurance compensation (as a percent of total compensation) constant through the forecast period – an assumption we believe is reasonable at worst, and more likely tainted by a recency bias and conservative in the long run. We note that despite a very recent downward trend, even over the last 10 years health insurance compensation has grown more than twice as fast as real wages.

Thus our assumptions are conservative, and we see the potential for upside. If the growth rate for health insurance compensation is 20 percent higher than for wages over the next ten years (a very modest assumption; from 1998 – 2008 health insurance growth was 155 percent higher than wage growth), our model suggests an added $70B in 2019 real health insurance premiums (and a total of $341B over the 2009 – 2019 period).

We expect that health reforms will increase total premium spending by 5 – 10% over our baseline forecasts between 2014 and 2019. This very rough estimate is a function of three dynamics which we believe could fundamentally change the private insurance market: (1) increased enrollment; (2) increased pricing; and (3) the aggregate actuarial value of the coverage selected. Enrollment is the easiest to forecast – both CBO models and our own analysis of the uninsured (and who may choose to remain uninsured even with a mandate) project that total enrollment in private insurance (covered lives) will likely rise 7 to 9 percent as a direct result of reforms. In a call earlier this week we detailed our view that the reform package may further accelerate the pricing of health premiums, largely due to underwriting restrictions that increase the risk of an average contract.[11] We also believe that there should be a demand curve effect on total premiums, countervailing the unambiguously positive impact of higher enrollment and prices. We believe that it is likely that insurance pricing dynamics and the nature of the exchanges (including adverse selection), along with the individual coverage mandate cause the aggregate actuarial value of insurance to fall. The primary arguments are two-fold: (1) there will be a subset of enrollees who will simply seek to satisfy the individual mandate by purchasing the lowest-cost (i.e., lowest actuarial value) option; and (2) there will be a subset of employees (especially households with income >$40,000[12]) who will refuse ESI and elect coverage in the health insurance exchanges, trading cash for a lower actuarial value of health insurance. Note that for every percentage point decline in aggregate actuarial value (as a percent of total health costs), we estimate that total premiums would decline by approx. 90bp. On net, we believe that these three dynamics together very roughly could imply an additional 5 – 10% upside to these total real premium forecasts by 2019 once we add in reform effects. Given our understanding of the elasticity of health premiums, we believe it is reasonable to surmise that price growth will be effectively offset by a substitution effect whereby enrollees elect in aggregate less generous coverage, leaving the net impact of reform on premiums as effectively a function of increased enrollment.

Medical loss ratio

Our baseline (pre-reform) models project that the aggregate medical loss ratio will continue on a slightly downward trend – though continue to exhibit normal course of business cyclicality – throughout the forecast period. In short, we see slight gross margin expansion among private insurers. This is in contrast to the fairly rapid (and historically anomalous) MLR declines projected by OACT. Over the 2009 – 2019 period, our model projects an aggregate decline of 1.3 percent compared to OACT’s forecast of a 3.4 percent decline. In addition, our forecast is significantly more cyclical (Exhibit 8).

Many of the key drivers of MLR dynamics within our regression model framework are very similar to those that are most impactful on the total premiums model – notably, pricing, employment and the ESI market. Additionally, our consideration of the MLR’s historical tendency to regress to a (rolling) mean helps to assure that forecasts remain reasonably range-bound.

As we have outlined in detail, a projected return to economic growth (and the eventual normalization of unemployment rates) have a significant role in reducing MLR – from the CBO-forecast peak of unemployment of 10 percent in 2010, a return to 6 percent unemployment is associated with a 120bp decline in MLR. Further, medical inflation and the acceleration and deceleration of medical inflation (roughly speaking, the first and second derivative of medical price with respect to time) are key cyclical drivers of MLR. And though we model projected medical inflation as smooth (that is, ex-surprise price changes) we estimate that the stable annual increase in medical prices that we include increases MLR by approximately 70bp in each year of the forecast period.

Again we emphasize that employer sponsored insurance trends are as critical to MLR as they are to total premiums. We’ve argued that the risk-pool efficiency generated by insuring workers through their employers helps minimize cost of care (and improve insurer margins). For projection purposes we conservatively hold health insurance compensation (as a percent of total compensation) constant, and stress that this assumption raises the projected MLR rather significantly. Even if health insurance compensation grows only modestly faster than wages over the next ten years (e.g., 20% faster v. 155% faster from 1998 – 2008), projected 2010 MLR would be 10bp lower and 2019 MLR would be 45bp lower than we currently project.

Keep in mind that for the moment, our MLR model, having been built on historic data, speaks to the employer-dominated private insurance market, but not the exchange-based private insurance market that begins in 2014. Nevertheless, many of the underwriting restrictions in the reform bill will affect employer-sponsored plans. In particular, restrictions regarding pre-existing conditions, rescissions, etc. serve to increase underwriting risk. We believe that the impact of health reform on MLR is likewise negative (underwriters demanding more margin because of more risk) – thus our comfort with forecasting slight declines in MLR as well as MLR volatility[13]. Exacerbating this view, and reinforcing the projected downward trend, is the fact that ESI enrollment (traditionally the most efficient pool) as a percent of total private insurance enrollment drops as a result of reform (from both new individual HIE enrollees and current ESI enrollees who elect to switch to HIEs) from 85 percent to just 76 percent.

In addition, our conjecture is that by disassociating insurance from employment – both by subsidizing premiums for some enrollees, thereby mitigating the income effect on the decision to carry coverage; and by removing the employer as the source of coverage for some HIE enrollees – that the MLR trend will become at least marginally less sensitive to employment cycles. Of course, we would also caution that while we believe that these views are sound theoretically, that we do not have the benefit of 40 years of health expenditure data to support our reform hypotheses.

  1. See our February 16, 2010 call “Introducing our Health Demand Model”
  2. We emphasize that our regressions are built on a history that of course does not include reform effects, and that our estimates of reform effects are therefore our judgments and estimates, rather than de facto outputs of our regressions.
  3. The usual caveats apply throughout all of our models: as with any projections based upon historical regressions, the driving assumption is that relationships are constant through time. This assumption becomes even more critical given the changing nature of healthcare, so we carefully consider how changes in healthcare are likely to drive actual premium and MLR results away from the estimates our models produce. The 40+ year look-back gives us some confidence in the long-run relevance of the coefficients we’ve calculated, but unforeseen shocks may impact any year, or series of years. The impact of such shocks on our projections is inherently difficult to predict.

  4. http://www.census.gov/hhes/www/hlthins/hlthin08.html
  5. With the important proviso that beneficiaries can only enroll during the annual enrollment period
  6. NHE data include both the total premiums received by private insurers as well as the administrative expenses / net income (i.e., a close approximation of the difference between total premiums and the medical benefit expense). This data allows us to calculate an MLR (i.e., 1 – net / total benefits).
  7. See for example our November 2, 2009 call “2010 HMO and PBM GMs…”
  8. And to carry things a step further – we presume the relationship between rate of job loss and accelerated claims costs is non-linear. This implies that an annual average rate of job loss for a year that had an intense period of job loss would tend to underestimate the acceleration in claims cost.
  9. See Evans’ “Health and Capital;” 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: www.sector-sovereign.com.
  10. We define “employee compensation” as the sum of wages, and employer contributions to health insurance premiums, life insurance premiums and pension / retirement funds.
  11. March 29, 2010 “Three Reform Realities that Aren’t Priced In”
  12. Ibid note 6.
  13. We do not view the MLR floors introduced by the legislation as a rate-limiting constraint and have detailed our view that private insurers likely already meet the threshold loss ratio, as calculated by the law (See ibid note 7).
Print Friendly