Third Meeting,
Thursday, September 7, 2000
Friday, September 8, 2000
Attendance:
Members of the Technical Panel:
Dale Yamamoto, Chairman
Principal
Benefits Practice
Hewitt Associates
Lincolnshire, IL
James Robinson
Associate Scientist
Center for Health Systems Research and Analysis
University of Wisconsin-Madison
Madison, WI
Michael Chernew
Associate Professor
Department of Health Management and Policy
University of Michigan
Ann Arbor, MI
David Cutler
Professor of Economics
Department of Economics
Harvard University
Cambridge, MA
Len Nichols
Principal Research Associate
Health Policy Center
The Urban Institute
Washington, DC
Consultant to the Panel
Sam Gutterman
Director and Consulting Actuary
Price Waterhouse Coopers
Chicago, IL
Staff of the Health Care Financing Administration, Office of the Actuary:
Richard Foster, Chief Actuary
Sol Mussey, Director of the Medicare and Medicaid Cost Estimates Group
Claire McFarland, Actuary
John Shatto, Actuary
Gregory Savord, Actuary
John Wandishin, Actuary
W. Kent Clemens, Actuary
Katie Levit, Director, National Health Statistics Group
Mark Freeland, Deputy Director, National Health Statistics Group
Stephen Heffler, Deputy Director, National Health Statistics Group
Jackie Carroll, Executive Officer
Donna Holt, Secretary
Terry Peach, Secretary
Staff of the Office of the Assistant Secretary for Planning and Evaluation:
Christy Schmidt, Deputy to the Deputy Assistant Secretary for Health Policy,
ASPE, DHHS
Ariel Winter, Executive Director of the Panel, ASPE, DHHS
Gene Moyer, Consultant to ASPE and to the Panel,
DHHS
Others Present:
Mike Andrews,
Social Security Administration
Phil Ellis, Economist,
Treasury Department
Linda Baker,
General Accounting Office
Cynthia Engel,
Health Care Financing Administration(now known as Centers for Medicare and Medicaid Services(CMS))
Noah Meyerson,
Congressional Budget Office
Robert Nguyen,
Congressional Budget Office
Eli Donkar, Deputy Chief Actuary.
Social Security Office of the Chief Actuary
Steve Goss, Deputy Chief Actuary.
Social Security Office of the Chief Actuary
Steve Lieberman,
Congressional Budget Office
John Sabelhaus,
Congressional Budget Office
Bruce Schobel
American Academy of Actuaries
Ralph Monaco,
University of Maryland
Proceedings of the Meeting:
September 7
The meeting was called to order by Mr. Winter, who welcomed the group and thanked them for all their hard work. He pointed out that Mr. Yamamoto would be asking them to approve the minutes from the last meeting and that once they were approved, the minutes would be available for members of the public and staff who desire them.
He listed several hand outs which had been received by the panelists and which were available at the table at the rear of the meeting room. He also asked any member of the public who wished to make a statement at the end of the day to see him during one of the breaks so that he could get them on the schedule. Then he recognized Mr. Foster for some remarks.
Mr. Foster also expressed his appreciation for the panel's hard work and announced that he had a get-well card for people to sign for Ms. Rosenblatt. The card was sent around the table for messages and signatures for Ms. Rosenblatt.
Mr. Winter then turned the meeting over to Mr. Yamamoto.
Mr. Yamamoto asked for and received from Mr. Robinson a motion to approve the minutes from the first meeting. Mr. Chernew seconded the motion. The motion was unanimously approved.
Mr. Yamamoto then recognized Mr. Cutler for a presentation for Subgroup I on Medicare assumptions.
Mr. Cutler divided the assumptions into three groups: assumptions common to OASDI, SMI, and HI, utilization and payment assumptions, and contingency reserve assumptions.
Assumptions common to OASDI, HI, and SMI are basically outside the purview of the panel.
The subgroup was basically concerned with assumptions on the disbursement side - benefit payments including utilization and costs and administrative expenses. Mr. Cutler said he was going to spend most of his time on 3 issues related to the utilization and payment assumptions: What is currently done, an analysis of short and intermediate term forecasting, and some preliminary comments on these assumptions.
The methodology used is that total spending is the product of prices, per beneficiary quantities, and the number of beneficiaries. The number of beneficiaries comes from the Social Security actuaries. Prices and utilization vary with the program: Hospital Insurance (HI) and Supplemental Medical Insurance (SMI). HI is about 60% of Medicare spending.
Some 65% of HI spending is for inpatient hospital spending; home health agencies and skilled nursing facilities make up most of the rest along with a small amount of managed care organization spending.
On the Part B (SMI) side, doctors are the largest part of spending followed by out-patient spending, durable medical equipment, labs, other carrier services and other intermediary services.
Mr. Cutler went through the forecasting of these items sequentially, with a division of forecasting into short-term (about three years), intermediate-term (4 to 25 years for HI and 4 to 10 years for SMI) and long-term (25 years or more for HI and 10 years or more for SMI). He dealt mainly with the intermediate term forecasts.
For HI, prices involve labor and non-labor goods and services so the labor portion is grown at the rate of hourly earnings, while the non-labor goods are grown at the overall rate of inflation. There are small adjustments for skilled nursing and home health, but these are relatively trivial.
Of more importance is the forecasting of the various quantities and utilization measures. This is divided into two parts - admissions or days in the hospital or a skilled nursing facility and visits for home health agencies and measures of intensity of the utilization measures such as case mix for hospitals. In addition, there is an aging term and a managed care shift effect as people move into managed care. The HI forecast is much more detailed than the SMI forecast.
The SMI forecast is made difficult by the fact that hospital outpatient services are as of August 1, 2000 paid on a prospective basis. There is no experience to help determine the forecast for the new prospective payment system. One of the recommendations is going to be that six months or so of research on outpatient prospective payments or other prospective payment systems would add a lot of information and accuracy to the forecasting process.
Outpatient benefits are forecast much as inpatient benefits are with a utilization term and a case mix term. There is no aging term and no managed care shift term. There probably should be.
All other SMI components are done on a residual basis. The residual contains four components: a utilization term, a case mix term, an aging term and a managed care term. Mr. Cutler asked the actuaries if that was correct. Mr Foster replied that if intensity was included, it was correct. Mr. Cutler replied that intensity was included in the case mix term. Mr. Cutler felt that the big driver of the forecast was the quantity term. The behavioral offsets are also important and these are currently taken into account in the forecasting process.
Mr. Robinson asked whether intensity affected prices or quantities. Mr. Gutterman said it could affect either, depending on the quantity units used. Mr. Robinson agreed.
Mr. Cutler indicated that the forecasting technique was sort of an educated extrapolation of the historical record which contains some very weird experiences. What one does about the weird experiences is very important. What the actuaries have done is draw pictures and try to tell reasonable stories based on an examination of the pictures. They have not used very sophisticated econometric techniques as yet. Use of these techniques might not pass the "laugh" test, but they might be helpful in looking at these assumptions anyway.
Mr. Cutler then went through six pictures showing the time trajectories (actual and forecast) of the market basket and prices received by hospitals. Mr. Chernew pointed out that the actuals included the effects of legislated changes. Mr. Cutler agreed, but indicated that the actuaries forecast the market basket and on top of that the effects of changes in the laws. While the forecasting may not be correct, it is probably done as reasonably as one can expect.
Mr. Foster commented that what the actuaries try to do is to compare actual hospital labor and non-labor prices paid to standard price and wage increases and try to understand the differences over recent time. Currently these differences are negative indicating that hospital prices and wages are growing more slowly than the standards. Mr. Nichols suggested this might be happening as hospitals lay off nurses. Mr. Chernew and Mr. Cutler then had an exchange about how price changes might be a weighting problem as the proportion of nurses and other supplies changed over time. Mr. Foster pointed out that the re-weighting of the market basket every five years accounts for many of these weighting changes.
Mr. Freeland said that the market basket is insensitive to these quantity changes because often hospitals are changing the share of wages between RNs and LPNs, while total wages paid are unchanged. The shift to a different priced resource, then, shows up in quantities rather than in prices.
Mr. Cutler then moved on to the forecasting of quantities. He showed the history of inpatient admissions, of skilled nursing facility days, and of home health visits. There was a discussion of the effects of the aborted Medicare Catastrophic Coverage Act on days as well as the spillover effects of the inpatient hospital prospective payment system and the passage of the Balanced Budget Act.
Mr. Cutler then turned to measures of intensity - case mix for hospitals and skilled nursing facilities. Their time paths are complicated by initial changes, by actual increases or decreases, by gaming, and by efforts to root out fraud and abuse. Further, some advances in medical techniques, especially surgeries, are likely to increase the case mix, while others, such as new drug therapies, may not change it at all. Mr. Gutterman pointed out that mortality rate changes also affect case mix since if the person has died, he is not available to use more intensive services. Mr. Robinson also pointed out that cohort aging also affects these results. Since aging in the model is applied only to utilization and not to case mix, it gets lost in the aggregation of units and the cost per unit analysis. Mr. Foster said that this was accurate, but it is not clear what to do about it.
Mr. Cutler completed his material on the HI forecasts by saying that the issues brought up by Mr. Chernew concerning the changes in the case mix resulting from changes in the types of technological change needed to be looked at and perhaps incorporated into the forecasts.
Mr. Cutler then turned to an SMI picture with five lines showing prices of various SMI goods. The time path for physician prices is very confusing, largely, suggested Mr. Nichols, because of the low volume payment for physicians. These payments for low volume show up in physician prices. Since these are nominal prices, prices for physicians in real terms are falling and will continue to do so for the next 75 years. This, however, is set in legislation so that the model is basically required by law to forecast this way. Other prices each trend from their latest historical level to the long term assumption.
Associated with the payments are behavioral offsets. These offsets suggest that physicians will perform more services as prices are held down. These offset assumptions are reasonable, so long as the price is above marginal cost. Otherwise, physicians would perform fewer services to minimize their losses.
His tentative conclusions are that the price assumptions are generally reasonable except for the effects of legislation which sets prices or sets specific ways of paying for services. The case mix index assumptions are probably too low. The managed care assumptions are not handled particularly well. There probably should be some assumption about recoveries from fraud and abuse. There is a lack of consistency in the HI and SMI assumptions which becomes important now that some HI payments for home health visits are being transferred to SMI. Finally there are the age-sex assumptions in HI, but not in SMI, and issues about technological change. He then opened the floor to questions and comments.
Mr. Robinson asked when studies and demonstrations on the prospective payment systems (PPS) including those done for PPS for inpatient hospitals would be available.
Ms. McFarland indicated that the home health demonstration report should be available in the next 6-8 months and that should provide data not currently available. She also indicated that Sharman Stephens of HCFA(now known as CMS) is heading up a group to monitor home health experience studies around HCFA(now known as CMS) and that may produce some data or other information.
Mr. Cutler asked the group how it felt about the issue of the 10 year (for SMI) and 25 year length (for HI) of the long term projections. Mr. Robinson felt that ultimately they should be changed to be the same length as a result of making the entire model more unified and consistent.
Mr. Gutterman said that he was not sure if the projection of case mix is going to be any different for 25 years than it is for 10 years. Understanding the interactions between and among the projections should precede consistency. Mr. Chernew agreed that accuracy was sufficiently weak that understanding what is going on is more important than forcing the long term to be the same.
Mr. Chernew mentioned that his methodology group was going to have some comments on consistency. Mr. Cutler thought that the discussion might be deferred until after their report.
Mr. Yamamoto asked if there was anything in the law requiring the 10 year and 25 year projection difference. Mr. Foster said the law was not very explicit and the 10 year/25 year dichotomy was more historical accident than anything else.
Mr. Cutler asked about the specific method of making recommendations by the entire panel. Mr. Yamamoto said that each subgroup would present recommendations and seek consensus from the whole group.
Mr. Yamamoto then called a five minute break.
After the break, Mr. Robinson made a presentation on projection methodology. After mentioning pieces he had sent to the panel, he began with a summary table he had put together for this presentation. The table referred to the intermediate forecast, up to 10 years for SMI and 25 years for HI.
Mr. Gutterman asked about the 12 year time span of the table and was told that while the Trustees Report presents ten years, the spreadsheet includes the extra two years for budget estimates.
Mr. Robinson split the population projection methodology between nonmanaged care and managed care. The population projection is also split among various eligibility groups or subgroups of interest. These groupings differ for the HI and SMI models.
On the HI side, the managed care population for the aged is split by age and sex. The disabled population is not split in this way. Each managed care group is then subtracted from the total to arrive at the nonmanaged care population, which receives most of the attention.
Mr. Yamamoto asked whether there is research to cause the actuaries to include the ESRD population in the general population. Mr. Mussey said that when they needed to do so, they break out the ESRD groups and handle them with separate analyses.
Mr. Robinson said he had a section on the three major groupings of the model: the population model, the utilization model, and the pricing components all of which make up the general methodology. There are several buried assumptions in the models, the major of which is that the number of persons remaining alive is prorated over the age groups which in essence assumes that survivorship does not vary by enrollment status within the age-sex cohorts. He did not go through all the steps in his presentation, but just showed the structure of the documentation. The service utilization breakdowns, the managed care shift effect, and ESRD models only exist in the HI model, but not in the SMI model.
Age-sex and enrollment utilization rates are multiplied by historical quarterly utilization rates to produce historical utilization rates by historical quarter for each sex, age group, service, and enrollment status. This produces a starting base. Then the growth factors are applied year by year to get future utilization rate assumptions. These rates are then multiplied by the population projections to arrive at units of utilization. Finally, to arrive at incurred benefits, cost per unit values are multiplied by the units.
A separate analysis is done for HI managed care. On the SMI side, there is a separate managed care spreadsheet which follows a reconstruction methodology in which historical trends are decomposed, projected into the future and applied in a way similar to that used for nonmanaged care.
The documentation ends with the projection of incurred benefits in the future.
Mr. Robinson ended with an appeal for the actuaries to look over his work and comment on any oversimplifications or misstatements before the panel tries to use this, perhaps as an appendix to the final report.
Mr. Chernew then continued the projection methodology presentation by discussing the inconsistencies between the two models or at least areas in which they are not coordinated. The two major inconsistencies are that HI is age/gender adjusted while SMI is not and that the SMI and HI methodologies do not have common components. Whether this is important or not is an issue for the panel to decide and to determine whether to recommend that this be fixed.
Mr. Chernew then went through seven recommendations: Add utilization measures to the SMI model, include age and gender utilization patterns for all services in the SMI model; make distinctions between spending for decedents and survivors; make assumptions regarding shifts in the site of care explicit; treat managed care utilization and case-mix consistently; make assumptions more explicit and transparent and integrate them with each other; and the actuaries should work more closely with research organizations, especially with RAND. There was considerable discussion of these recommendations.
Mr. Cutler felt that some of these could be handled right away and that some would take longer. Mr. Chernew replied that this might be true, but the software was such that any of these tasks were daunting and that incorporation of many of these was likely to be very difficult.
Mr. Foster said that some of these changes could be made rather easily and that others, such as going to a full-blown Markoff model, would be much more difficult. There might be some way to do this progressively or perhaps to reshape the system to make it more amenable to modification. He felt the RAND model might be very helpful in all this. When the Office of the Actuary did switch over to the current models, it took a long time to get appropriate data and to get the people free from other important tasks to allow them the time to make the changes. He agreed with the suggested changes, but figuring out how to do implement them might be difficult.
Ms. Levit had two comments. First, she asked that consideration be given to Medicare home and community-based waivers. Mr. Chernew replied that they had not given a lot of thought to this level of detail in the model. Ms. Levit also asked about the desirability of breaking managed care into services rendered to beneficiaries. Mr. Chernew replied that when people were paying a fixed premium and utilization didn't matter, it was not clear that one could easily incorporate these into the model. In the long run, however, utilization will matter. In current law, however, expenditures may be more important than utilization.
At this point (12:10) Mr. Yamamoto adjourned the meeting until 1:00 P. M.
Mr. Yamamoto called the meeting to order again at 1:10 P. M. Mr. Cutler commented that he agreed with Mr. Chernew's statement that managed care payments matter more than utilization in the short run, but asked Mr. Mussey if it was not true that plans are now required to report more about services supplied to members.
Mr. Mussey replied that plans are beginning to report hospital utilization for risk adjustment payments and that they will report more as the risk adjustment model becomes more comprehensive. Mr. Cutler said that they are beginning the more comprehensive data collection this year (2000). Mr. Robinson questioned whether the data would be reliable or not.
Mr. Yamamoto then turned the meeting to the long range growth assumptions group, beginning with Mr. Chernew. The charge of the group was to look at the assumptions for growth between 25 and 75 years from now. Mr. Chernew asked the group whether the Panel's charge included recommending an expected value growth rate. Consensus seemed to be that it was, but Mr. Gutterman said that there was more than just suggesting a number. Rather, the charge includes examining considerations, methods, and future research into the formation of the number. Mr. Chernew agreed, but felt that such considerations would probably be under represented in his discussion.
He began by asking if HI age-adjusted cost growth was equal to wage growth per worker. Mr. Foster agreed, but indicated that it was average hourly earnings, a slightly different concept.
For SMI, the growth rate is basically equal to real GDP per capita growth. Per Capita nominal GDP is assumed to grow at 4.4 percent and CPI is assumed to grow at 3.3 percent per year, which means that real per capita GDP growth is 1.1%. Mr. Chernew intended to focus on the difference between health care cost per capita growth and real GDP per capita growth.
There was a discussion about whether total or Medicare cost growth should be used. Mr. Freeland felt that depending on definition, both sides of the discussion could be right.
Mr. Chernew indicated that health expenditures showed a strong upward trend for 35 years and relatively flat for the last 5 years. The question is how much one weights the experience of the last 5 years as an aberration or not. One might think about the future of medical technology, health status, and how each is impacted by the other and by managed care. Benefit design is also important.
He discussed an inverted "U"curve which represents the time path of costs of technology for any given disease. Initially, there is no treatment and the disease is very cheap. As one develops treatments for the disease, especially invasive ones, the cost of treatment rises quickly. Then as physicians refine the treatments and learn more about them, they can treat the disease in cheaper ways and bring costs back down. In the process of all this, more and more people are brought into the system and this also increases overall costs.
Mr. Chernew did a regression on the difference between health expenditures and GDP as a function of GDP growth and a time trend. The time trend did not matter, but GDP growth mattered a lot. As GDP grew faster, the gap got smaller.
He mentioned several other variables which may have an affect on the future. He hoped that everyone would pass to him a favorite number for the growth rate.
Mr. Gutterman suggested that the analysis might help one determine the limits to growth. Mr. Nichols returned to Mr. Chernew's graph and said that with a 1% growth rate, health expenditures would reach 20% of GDP by 2040. This he felt was reasonable since David Cutler's boss predicted at one point that it would reach 18% by this year.
Mr. Foster indicated that there might be a reaction by society when society decided that health costs had reached too large a proportion of GDP. He gave as an example the early 1990's when employers felt that health costs were too high This led to the managed care revolution. Mr. Cutler said that such reactions tended to be temporary. After employers or others in society lose interest, prices and expenditures tend to go up again. He mentioned a time in the early 1980s in which there was much discussion of controls over the medical community and for two or three years prices rose very slowly. After that, prices and expenditures continued their rise as before. Managed care is probably somewhat the same thing.
Mr. Gutterman mentioned the beneficiary share of the Medicare premium which may represent the amount beneficiaries are willing to pay for their benefits. The relationship between the SMI premium and the minimum OASDI benefit may also represent a kind of ceiling on the SMI premium payment
Mr. Yamamoto also mentioned that there were many cost controls built into the Medicare program. He wondered how those controls interacted with the uncontrolled portions of national health expenditures.
Mr. Goss suggested that if the graph had been drawn defining the gap as the difference between the growth rates of expenditures and GDP, it would have reached 100% asymptotically and would have looked very different and perhaps make one think that this might work out over time.
Mr. Chernew agreed, but said he had drawn the graph in this way since it seemed to mirror what was used in the model.
Mr. Foster said that the actuaries might be able to help with improving these graphs using techniques currently in the Trustees Report. He also said he would like to hear Mr. Cutler say more about the role of technology in growth. What work he had seen on the subject suggested that technology explained about half of growth. He wondered if such information might lead to a number. Mr. Cutler said that this was exactly what had been in the back of his mind.
Mr. Foster went on to say that people will continue to want the new technology until someone tells them that the society can no longer afford it. Mr. Cutler then asked who would tell them since we have no experience in doing that with medical care.
Mr. Gutterman suggested that the provider side is also important. If the provider is paid too little, they may cut utilization to minimize their losses even if there is new technology to be used.
Mr. Cutler suggested that today technology is much less hit or miss. Rather, new developments are targeted very precisely so that they are less invasive and cheaper and a doctor can give them to more people. Bypass surgery is not going to be done on people who have no need; staton drugs, on the other hand, may be given to people who have mild symptoms where we would never consider them for something surgical.
Mr. Foster said that while expenditures might rise from the current 13.5 or 13.6% of GDP to 20% or so, there has been zero enthusiasm in the past ten years for increasing social insurance premiums by even a fraction. This may eventually enforce a limit.
Mr. Cutler mentioned the budget surplus. Mr. Nichols agreed, but also felt that we might not always have the current mind set about the value of government services.
Mr. Foster did not know what the tolerance level for health care spending is, but even if we eventually reach a limit, we may not manage it in the time frame we are talking about. Mr. Nichols, however, gave the analogy of military spending. No one in 1970 would have believed that it would be this low by 2000 and rumor has it that it is going lower. Mr. Cutler suggested a happier version. The elderly are healthier and they may either decide to work more or we may force them to work longer. It may not then be a transfer from society to the elderly.
At this time, Mr. Yamamoto called for a five minute break before the presentation on uncertainty.
After calling the group together again at 4:10, Mr. Yamamoto asked if there were any other comments on the last presentation.
Mr. Ellis said that he had observed that the prospective payment system for hospitals is not as prospective as one thinks in that what a hospital does to people affects how much they get paid. It does not have the strong incentives one would expect. High reimbursement procedures are the ones which are growing the most. The same thing may happen with the outpatient and SNF prospective payment systems.
Mr. Cutler said that he thought Medicare Payment Advisory Commission (MEDPAC) might have some information on that.
Mr. Nichols said that he thought what Mr. Ellis was saying was "How binding was the PPS constraint?" He felt that a barometer of that was how much lobbying providers do to postpone implementation of the PPS system. The group felt that they might do well to hear someone knowledgeable on that, perhaps MEDPAC.
Ms. McFarland said that the Home Health PPS system is not wholly prospective. Providers cannot increase the payment by doing more for the patient, but they can say the patient got sicker and so reclassify him for an increased payment.
The Panel thought that Mr. Foster and Mr. Winter might decide on someone to brief them on this issue. Alternatively, they might find some reading material on the issue. Mr. Foster thought there might be something available from within HCFA(now known as CMS). Mr. Robinson thought that he might have some materials from a conference held recently at the University of Wisconsin.
Mr. Freeland felt that there had not been sufficient thinking about the issue, but that rather than a single number, they might consider a Gomperts curve which rises as trade-offs become more and more difficult. He mentioned that patients view the trade-off as between expenditures on health and their quality of life; If you are sick, you may not enjoy a Mercedes Benz or a gourmet dinner or anything else. It depends on their perceived connection between spending and what they are receiving in return. Similarly, one can look at the connection between revenues and spending. The elements brought up at this meeting might make a big contribution to knowledge if someone put them together in a reasonable way.
Mr. Gutterman said that the 1991 technical panel had gone through some of these trade-offs, but it never got into the issues of technology. Looking at that report might be helpful.
At this point the Panel turned to the subgroup on uncertainty and stochastic forecasting.
Mr. Gutterman began by indicating that he had more questions than answers in his presentation. Perhaps there are no answers when one talks about uncertainty. First, uncertainty is a vehicle for indicating how comfortable one is with or how certain one is about any single number presented. It forces one to focus on the whole dynamics of the system. It does not, however, deal well with structural changes.
Mr. Robinson then gave a brief overview of the SMI simulation currently in use. For each of 40 combinations of service and enrollment groups, the model calculates average growth rates for each quarter from 1991 to 1999 as well as variances and covariances about those growth rates. These are then aggregated to the five enrollment groups and include managed care under the assumption that it has the same growth rate and the same variance within enrollment groups. These are then aggregated to a single rate and variance for all services and averaged over the quarters to annual values. He indicated that he did not agree with how this was actually done, but that he would discuss this later.
Given these variances and growth rates one can arrive at averages or distributions by multiplying them by historical per capita values. Many of these are displayed in the SMI Report, appendix C.
Mr. Robinson said that while the method took the square root of each variance and then averaged them, it would have been better to average them and take the square root of the average. To do so would have resulted in a somewhat smaller variance. Also, he questions the amount of autocorrelation among the quarterly amounts over the year. The method assumes positive autocorrelation. He feels the question needs examination. He provided many examples of results depending on the size and sign of the autocorrelation, as well as whether it is a stationary structure, especially for adjacent years.
Mr. Foster mentioned that there were several articles and books dealing with autocorrelation in the last ten years. Unfortunately, they may require more data than are available to the Office of the Actuary.
Mr. Robinson, Mr. Chernew, and Mr. Gutterman then had an exchange about what could be done to solve the autocorrelation problems in the data. This included the problem of removing the effects of any new legislation in the data as well as using the data for a single state as a control.
Further, Mr. Robinson went on, if one tries to extend the method to HI, one has to worry about whether one does the two models separately or together. One has to worry about whether the variation varies with economic and demographic model components. Finally, the 25 year intermediate period in HI is likely to cause the SMI problems to be even more pronounced. Mr. Gutterman said that if one tried to extend the SMI model to 25 years, he would have the same problems. Also, he said that in SMI the premium provides a self-correcting factor which would not be present in HI. There was considerable discussion about these variances, their calculation and their use in the model.
Mr. Robinson then made one final comment about the shape of the assumed distribution and that the resulting distribution of benefits might be closer to log normal than normal. If this is the case, the intermediate case will be more like the median than the mean.
There was a discussion of what the alternative I, II, and III forecasts meant. Mr. Foster said that the alternative II was the Trustees' best estimate and that I and III represented 2% higher and 2% lower growth rates relative to something (not specified). Mr. Robinson indicated that if all the variation were put in, there was a danger that the interval would in future years become so large as to make the results unusable.
Mr. Goss then commented that the 1991 panel had addressed this question somewhat and had reported that for the OASDI program they had run a 95% interval and had got a set of numbers very close to the I and III alternatives.
Mr. Cutler made the point that the stochastic model was a pioneering effort for the Trustees and that while the panel might quibble about the details, they feel that the effort should be extended to HI and made a regular part of the reports each year.
Mr. Gutterman asked the panel for help in setting the direction for the subgroup.
Mr. Foster asked that the panel think about whether the stochastic model was more important than doing something like the OASDI programs which set alternatives I and III by a set of explicit high and low cost assumptions. Is either of these worth the considerable work involved in constructing and using these uncertainty models?
There was considerable discussion on this issue, but Mr. Gutterman elected to go on with his presentation and delay the decision until later.
His next topic was the actuarial balance calculation and whether it needs to have some uncertainty interval connected with it. For SMI, should there be an interval around the premium calculation?
Mr. Foster indicated that the actuarial balance concept preceded the easy graphing capability available with today's computers and that it was a handy and relatively simple way of summarizing the long tables as a single number which one could easily understand. The Trustees later based their test of long-range financial balance on the concept for that reason.
Mr. Foster also said that the Trustees, largely because of the efforts of Steve Kellison, pointed out that HI had got all the attention over the years because it was always going broke. SMI has grown faster than HI in almost every year, but SMI has received little attention.
There was discussion of other possible tests and of the need for further research.
Mr. Gutterman then asked for comments from anyone.
Mr. Ellis commented that the SMI status was a "hot-button" political issue and a point of contention in the current Medicare reform debate, but this should not dissuade the panel from dealing with it.
Ms. Schmidt said that as one who has to explain these things to various Secretaries, she would hope for a test for SMI which was similar to that for HI and one which would be more easily explained as opposed to a test of affordability, which would be a very difficult measure.
Mr. Gutterman asked about whether a test of affordability was possible. It was a little like the use of the word "Fair" which is almost impossible to define.
Mr. Foster said that the Trustees would no doubt welcome something which illustrated the implications of high SMI growth rates. Those which they have put in recent years have been relatively obscure because the trap of saying that the fund was out of balance was difficult to avoid in a simple manner.
After considerable discussion of the issue, Mr. Gutterman said that the panel should continue to focus on the issue over the next few weeks and that he would be raising more questions and asking for feedback.
At this point, Mr. Yamamoto called for a break.
After the recess, Mr. Nichols began a presentation on Presentation Issues/Research and Data.
He and Ms. Rosenblatt broke their task into two parts so that they would have separate recommendations on research and data and on presentation issues.
On presentation issues, he felt that the Report of the Public Trustees might be a good model and asked that copies be given to the Panel. Mr. Foster said that he would make them available.
Then Mr. Nichols went through five recommendations with sub-recommendations. Only the major recommendations are listed here:
The data used by the Office of the Actuary needs to be prepared on a quarterly basis in accordance with a specific schedule.
The staff of the Office of the Actuary should be sufficient to allow quarterly analysis to test actual versus expected and to test trends at levels of detail beyond what is now analyzed.
The annual Trustees Reports should use language that can be better understood by the public.
Redesign certain aspects of the Trustees Reports to improve clarity.
Add new sections to each report dealing with uncertainty and past forecast errors.
There were several comments about these recommendations. One in particular by Ms. Schmidt dealt with whether the recommendation on the Office of the Actuary's budget should include some specific percentage increase and that the percentage should be substantial.
Mr. Chernew suggested that the recommendation should include analysis of managed care on a geographic basis. Mr. Nichols agreed, but wondered if it could be justified in terms of producing the Trustees Reports. He also brought up the question of whether the research budget was too small in terms of the entire HCFA(now known as CMS) or DHHS budgets. There was also the question of many pockets in those budgets which might be tapped. Ms. Schmidt replied that even within the HCFA(now known as CMS) budget there were many pockets which were harder to see under the new organization than they had been under the old.
Ms. Schmidt also asked whether the quarterly data were desired because of timeliness or for some other reason. She felt that the timeliness issue needed to be brought out more explicitly.
She also mentioned that there was a new group within DHHS trying to find research gaps and do what it can about filling those gaps. The Panel was very interested in this new group. Ms. Schmidt said she would try to get them at least draft materials on the new group.
On presentation issues, Mr. Nichols said that the and Ms. Rosenblatt had asked themselves who reads the reports or what is the audience for them. His view, but not necessarily that of Ms. Rosenblatt, was that they should be written for junior Hill staffers. One objection he had to the current reports was that they would seem to be simply written and then would use an arcane term which was not defined and which lost him as a reader. Finally, he felt that more emphasis should be placed on cost per beneficiary or per worker.
Mr. Foster said that they had tried to write the overview section in the last revision for the junior staffer. For the rest of the report they had a tougher standard in mind, but they have offered prizes for the rewriting of selected paragraphs which were particularly obtuse. Still, is there a different target audience for the remainder of the reports? He said he would like the Panel's view on this.
Mr. Gutterman said that perhaps the overview should be the Report and other sections should be a series of appendices.
Also, the Trustees have formed a work group to revamp the OASDI report. The Office of the Actuary is part of that effort and the effort will probably affect the HI and SMI reports as well.
Ms. Rosenblatt felt there should be more graphs, but neither of them could think of any specific graphs to be added.
Mr. Cutler asked if they could hire an editor. Mr. Foster said they had hired several and had hired a Ph. D. in literature and writing to help them.
Mr. Andrews said that the Office of the Actuary had in the past tried to simplify the reports and someone suggested a wizard to help make the reports more readable. The wizards they hired, however, disagreed with each other and their recommendations could not be reconciled.
Mr. Nichols said that neither he nor Ms. Rosenblatt could write this for the actuaries, but that this was their reaction to the report as it stands. The two of them, however, felt strongly about the suggestion that the audience could handle a more sophisticated and explicit treatment of uncertainty. One thing that could be done is to look at what was predicted 5, 10, and perhaps 20 years ago and to see how close the predictions had come to reality. Ms. Rosenblatt felt strongly that the 5 and 10 year comparisons were especially important. Mr. Nichols also felt that there should be a discussion of the confidence interval along with the discussion of alternatives I and III.
Mr. Foster said that the problem was that they get a lot of static every time they make changes in the report. The discussion of stochastic methods was in an appendix because it was less controversial than trying to put it into the body of the report and have people think that it was effecting changes in growth rates or something like that. In addition, putting it into an appendix got stochastic methods into the report without upsetting the people who do not want any changes at all. This was a foot in the door.
Mr. Nichols said that both he and Ms. Rosenblatt recognized that and they applauded the actuaries for getting the foot in the door. Mainly, they simply would like the actuaries to kick harder.
Mr. Cutler agreed and also said that the web site had been very useful. The actuaries were very good about helping people learn about concepts in the report. He did not think we should forget that as we suggest ways to make a very complicated subject easier to understand.
Ms. Schmidt also suggested that there had been congressional interventions during the past 5 or 10 year periods and these have moved things all over the place. When one talks about, say, 2025, there will be perhaps eight changes by then. The panel should consider this when making their recommendations and give it some of their expert direction.
Mr. Nichols agreed and said that they had not intended this to be a score card on how well the actuaries had done in the past. Rather, they felt it would be more of an illustration of how hard the task is precisely because Congress won't leave things alone.
Mr. Freeland commented that most people ignore the bulk of the report anyway and only talk about and comment on the year of exhaustion. If it changes as it has recently, analysts writing in the popular press tend to imply that the actuaries do not know what they are doing. He suggested that a revenue and expenditure chart might help show that a small change in either can have a large affect on the year of exhaustion. Alternatively, there might be a cumulative revenue and expenditure graph which might show the same thing.
Finally, the discussion of the summary's importance indicates that working to improve the summary by putting some of these graphs in it might have the largest payoff in public knowledge.
Mr. Gutterman said that one problem with uncertainty is that this is not really statistical deviations from a mean as much as parameter risk. It is much more uncertainty about what any parameter will be than variation caused by small numbers or a small population.
There are, then, a lot of hurdles in explaining forecast errors, especially when these errors are complicated by intervening events including legislation. So while making things clearer is a good thing to aspire to, it may be very difficult to do in a meaningful way.
Mr. Foster said that while there was a large potential for improving the reports, it would not be easy to accomplish. He recalled President Carter's statement in 1978 that the future of Social Security had been secured for the next 50 years. Had there been some uncertainty measure attached to the predictions, the President might have been spared this embarrassment.
Some years from now, he would like to see a very different Trustees Report which reads differently and conveys information in a way that people can understand. While he would like to see it, he fears that it is not likely to happen. And if it did, what is the range of things that would appear in it and what is the likelihood of different parts of that range?
Mr. Nichols said of all the themes they offered in their recommendations, the one that they feel strongest about is that the report can do a better job of presenting uncertainty in the forecasts.
At this point, Mr. Yamamoto asked if the group was ready for adjournment. After some discussion, he adjourned at 6:05 P. M. until 9 A. M on September 8.
September 8
Mr. Winter called the meeting to order a little after 9 and turned the meeting over to Mr. Foster to introduce the several presentations he had arranged for the morning.
The first presentation will be given by Ralph Monaco of the Inforum group at the University of Maryland on the LIFT model. Mr. Monaco is President of Inforum and an adjunct professor at the University. Mr. Foster said that the Office of the Actuary has had a contract with Inforum for some years and has found the model very useful for ascertaining relationships between macroeconomic factors and the health sector of the economy.
After Mr. Monaco's presentation, John Sabelhaus will give a presentation on the Congressional Budget Office long term model. After that, Bruce Schobel of the American Academy of Actuaries, will, among other things, summarize a recent report by the Academy on Social Insurance Financing.
There are other approaches such as the macroeconomic sector input/output model mentioned by Mark Freeland yesterday. The Office of the Actuary is very interested in the views of the Panel on the comparative advantages on the various approaches.
Mr. Monaco then began his presentation on the LIFT 2080 model (Long-term Interindustry Forecasting Tool with forecasts ending in 2080), which tries to find the distributive effects across industries and spending components of having health grow at a rate considerably above the rate of growth of GDP. It further examines whether high spending on health care affects international competition of selected industries, what happens to the uninsured, and how high health spending can go. Initially, this was a more short term model; the long term concerns arose somewhat later.
The model includes two modules or sub-models, the population module (DPM) and the economic module (LIFT 2080). These models are run independently, but they talk to each other and pass information back and forth. The model includes information on 14 other countries as well as for the United States.
Mr. Gutterman asked how often these models run back and forth. Mr. Monaco said that they are basically run separately, usually with the DPM run first and the results input into the Lift 2080. There are no strong economic/demographic links as yet except for a fertility link and the beginning of a mortality link.
Mr. Cutler asked if they forecast health status other than mortality and was told that they would like to do so, but have not yet figured out how.
The problem is that the fertility rate has a considerable effect on various labor force and income variables and these also influence fertility. The effects of these on future fertility is not large enough to make much of a difference in the results. If the modelers could come up with a large enough fertility/economic link, they could model more appropriately the OASDI and Medicare long-term imbalances.
The LIFT 2080 model is basically an input/output model and macroeconomic in that they look at aggregates rather than individuals.
Mr. Chernew asked about the extent of dynamics in the parameters of the model. Mr. Monaco said that this was a big concern of the modelers. They had an 85 by 85 industry model in the shorter term model, but after 2015, they aggregated the model into an 11 by 11 industry long-term model. When Mr. Chernew asked a more specific question about energy in the transportation industry, Mr. Monaco said that they allow, for example, the energy use per unit of steel to change over time, but after 2015, they become agnostic about the question.
They basically try to get the broad trends right from 2015 to 2080 by, as Mr. Nichols said, "holding everything constant after 2015," depending on the interest of the client.
Mr. Cutler asked if one of the industries was the health care industry. He was assured that it was not only one of the 85 industries, but one of the 11 used after 2015. Mr. Cutler also asked if there was an exogenous trend increase and was assured that the consumer spending equation had three factors: a term which captures the aging of the population, a price elasticity of about -22, and a third which was not made explicit. Mr. Cutler further asked if medical expense was a constant share of income and Mr. Monaco agreed.
Mr. Monaco then indicated that none of the other industries are using medical services as input so that only consumer demand determined medical service demand. In response to a question from Mr. Gutterman, however, Mr. Monaco said that on the cost side, medical care as a percentage of employee compensation was an industry input.
There was considerable discussion about the model. Mr. Sabelhaus and the CBO model were also brought into the discussion. Mr. Monaco was asked if he could put together a paper describing the model and perhaps answering some of the questions addressed by the panel. He agreed that he would do so.
Then Mr. Foster introduced Mr. Sabelhaus to discuss the CBO Long Term Actuarial Model, LTAM.
He began by saying that he wished to break the presentation into three parts: an overview, a discussion of their goals, and comparing the model to dynamic micromodeling, which they are also working on.
The population model is essentially cell-based with 100 age groups, gender, and 4 marital status groupings for 800 total cells.
Mr. Cutler asked if the model forecast health status and was told it did not, but that the micromodel in planning will simulate health status outcomes.
The current model is basically one which will forecast new OASDI worker benefits using the continuous work history file from SSA. It will not replicate the SSA Actuary's model, but will rather use a lot of their data to solve for OASDI benefit amounts under alternative assumptions and policy rules.
They are doing this model because Congress requested that CBO project OASDI beyond the 10 year budget window. In addition, the model contains three major innovations. It allows CBO to replace SSA equations with alternatives while holding the rest of the SSA model constant. A second innovation is that it has built in an endogenous macro economy with feedback.
Mr. Chernew asked how they get around the fact that the model "blows up" (stops running and fails to produce estimates) under the baseline.
Mr. Sabelhaus said that the model which blows up is the budget model which the macro group is doing. No one in CBO really knows why that model blows up, but he thinks that the model should specify behavior differently and that if it did, they might see what the problems in the model are and be able to fix them. He feels that the model blows up because the ratio of taxes to GDP is kept constant. If taxes were allowed to rise, it might not. The model he is talking about here has not so far blown up.
Mr. Gutterman asked if the growth of Medicare would change the long-run economic picture. Mr. Sabelhaus said that when the endogenous macro economy is built in, it will.
The third innovation is that the model be able to do stochastic simulations. They currently can re-solve the model under alternative I and III assumptions and they are working on a Monte Carlo type of simulation. Part of the Social Security Office of the Chief Actuary's problem is that they have one person doing demographics, another benefits, and still another doing payroll. CBO has collapsed it into a single simulation model so that they can change the assumptions, re-solve the model, and do it again and again.
He also talked about the microsimulation model they are developing. There is a document out about it, but he did not bring copies of it with him.
The two models are similar after the micro model has computed taxes and benefits and summed them to the cell based level. Parallel development saves a lot of work since a lot of the cell based work carries forward to the individual micro model.
The things they can change for re-solving the system are demographic assumptions, economic assumptions, social security policy parameters, some additional trust fund assumptions, and behavioral parameters.
Mr. Cutler asked then if they were testing alternative assumptions or if they were mainly trying to see if the actuary's assumptions are reasonable?
Mr. Sabelhaus responded that the first step was to get the model to work and to make sure that their results were the same as those of the actuary when specific changes were made. Further, they were trying to match the CBO baseline for the first ten years and then to extend the baseline to 75 years out. There are two major activities in constructing this baseline. One is to alter the model so that it produces more reasonable outputs, possibly with macro feedbacks. The other is to try to make the assumptions more reasonable.
These activities involve considerable internal debate so that if they go to Congress with a new long-run view, they can be confident that they have a better view than the one that this one replaces.
Mr. Lieberman interjected that this presentation describes a work in progress. They are telling people that they do not expect to get reasonable results until later in the winter, next year. There are some completed modules, however, and a working model that they are relatively pleased with.
Mr. Cutler asked that they skip over the social security aspects of the model and turn to Medicare. He also asked if there was a document which indicated the assumptions they chose which were different from those the Trustees chose.
Mr. Lieberman said there was one still in the review process. Mr. Cutler asked when it was likely to be ready and was told that it was not clear when it would be available.
There was a discussion about the controversy around the document within CBO. The controversy apparently centered around the Medicare growth rates and the mortality assumptions as well as the economic assumptions.
Mr. Sabelhaus said that he plans to introduce Medicare into the model using the population from the SSA actuaries, putting in values for the important inputs, probably exogenously, and solving for the trust funds.
Mr. Cutler asked how they were going to decide on the cost and intensity rate assumptions.
Mr. Sabelhaus said that they would examine the equations looking for stability and trends. Then they would think about technology changes and everything else.
Mr. Lieberman said that they really were in the preliminary stages and that their main objective initially is to replicate what the actuaries do. Without good reasons for doing otherwise, they would want to control the model to the actuaries' values. They have not at this time done sufficient thinking about the Medicare questions.
Mr. Sabelhaus then went through some things the OASDI model can do at the present time. He actually ran the model while the group watched and commented on the outputs produced as he went. First, he ran a default model which calibrates the forecasting models. These default models replicate the actuary's results within about 1%.
Having calibrated the initial results, the model runs through eight modules where it solves for demographic and macroeconomic outcomes, for covered workers, taxable payroll, the number of beneficiaries, average benefits, and finally, the trust funds. They then have two outputs, a baseline case and an alternative for 25 years, 50 years, and 75 years out. There are several tables.
They continue to try to align the model to the CBO 10 year outcomes, to develop the macro economic sector to arrive at some general equilibrium, and additional development of stochastic simulations. These things are what they will be working on during the coming winter.
Mr. Gutterman asked about the distributions for the stochastic simulations. Mr. Sabelhaus replied that they are relatively simple and ran a stochastic simulation to illustrate. When it was done, Mr. Foster asked if the stochastic simulation they used was to draw the ultimate assumptions randomly and Mr. Sabelhaus agreed. Further, he and others had a question about the 95% confidence interval. It seemed too low given the standard deviation specified. Mr Sabelhaus indicated that it was in fact an interval around the mean and so the standard deviation was turned into a standard error of the mean.
There were questions to clarify the time frame for development of the model, especially for the Medicare portion of the model. The answer was that March seems a reasonable possibility, but that there were only six or seven people working on the model.
At this point, Mr. Yamamoto called for a break.
After the recess, Mr. Foster introduced Mr. Bruce Schobel, the Vice President and Actuary for New York Life Insurance and Chairperson of the American Academy of Actuaries' Social Insurance Committee. He also chaired the subgroup of the Academy's Task Force on Medicare. The subgroup dealt with Medicare financing issues and financial statements. The result of this work was to be the topic of his talk.
Mr. Schobel began by describing the Academy as the public policy arm of the profession. It basically issues policy statements and disciplines actuaries who fail to follow the standards of practice.
The Medicare Reform Task Force issued a report on evaluating Medicare solvency in July of this year. The Academy started the task force because it felt to do otherwise in the face of all the reform proposals was not responsible. The task force was divided into three subgroups on drug coverage, on competition and issues like contracting out, and (the one that Mr. Schobel chaired) on Medicare solvency.
First, the solvency subgroup asked if the crisis was real and decided that it was. There was a desire on the part of some members to look at the assumptions in the actuaries' model, but the group decided not to try to act as a technical panel. They did, however, look at the process of evaluation over the years as reflected by previous technical panels and by the process by which the Trustees choose assumptions. They decided there were sufficient checks and balances that the process was a good one and likely to come to reasonable conclusions. Among the points made by the panel which bear repeating are that no one knows what the next 75 years will look like and it is not possible to predict the next 75 years with any certainty. Still, the law requires that the forecasts be made.
The Trustees have had technical panels every five years or so and they make changes to the process, which moves generally in the directions suggested by the recommendations of the panels.
He then moved to his own views based on the work the subgroup did. The major one is that while the subgroup felt that there should be more work on introducing stochastic processes into the models, he does not. He mainly disagrees with the subgroup because he feels it is difficult enough to come up with point estimates. It would be even harder to come up with distributions around those estimates. In addition, there has been no discussion of the covariances. He felt that to ignore the interaction effects will probably produce a process which is less than optimal. There is a further danger that with today's powerful computers, people will treat the results as gospel and they are not. The subgroup was deeply divided on the subject, however.
The Panel asked him to identify the members of his subgroup and which ones agreed with him. He did so to the extent he could remember.
Mr. Nichols asked if his opposition to the use of stochastic processes stemmed from a fear that they will be misused or from a profound antagonism about separating independent effects from policy acts.
Mr. Schobel said that he did not think it could be done right because one does not know the probability distributions of these variables.
Mr. Nichols asked if it was worse than using alternatives I and III. Mr. Schobel did not precisely answer, but rather said that because we have to guess about distributions, stochastic processes do not contribute to the policy development process.
The subgroup also considered the relative power of the actuaries and the Trustees in determining assumptions used. They decided that the actuaries sign the statements of actuarial opinion, but that the Trustees actually determine the assumptions. There were people who felt that the actuaries were not following standard processes because of this, but in the end the subgroup decided that they were following standard procedure.
He also mentioned an actuary who wrote to his Congressman about the GDP growth assumption. The Congressman asked the General Accounting Office to investigate. They used Price Waterhouse as a contractor to do the actual work. Their report said that the details of the process could be argued with, but that the entire process was reasonable.
At this time, the Panel turned to a discussion of the next meeting which they decided would be on October 11 and 12. There were several questions about whether there was time to have another meeting because of the need to write the report in October, but they agreed that they needed another meeting. Mr. Nichols suggested and everyone seemed to agree that the subgroups come to the meeting with their recommendations and try to decide on the ones with which everyone could agree. Then during the rest of the month of October, the Panel could try to write the first draft of the report.
The group asked Mr. Foster if he agreed with this general strategy. He said they should try to wrap things up as soon as they could. Some of their recommendations will be fuzzy, but they might suggest additional investigation of these. If the report were not completed until mid-November, this would not be a problem. If it is not done until January, this is more of a problem. Mr. Nichols suggested that everyone come up with a set of bullets and paragraphs to go with them and that they share them with each other before the next meeting. Mr. Foster said that given their progress up to now, they would probably be pleasantly surprised at what they come up with. They further discussed the papers they had requested at this meeting.
They also discussed the problem of Ms. Rosenblatt's illness. She had requested that she be dropped from the Panel, but the Panel decided to send her a package of papers and leave the final decision up to her.
In addition to the October meeting, the Panel decided on a one day meeting on November 15th.
There was a question about how the report is to be presented to the Trustees. Mr. Cutler asked if Mr. Yamamoto was to present it to the Trustees. Mr. Foster said that this is the first technical panel which is to report to the Board of Trustees. Therefore, he thought it was a good idea to present it at the Fall meeting in November or December and that all of the Panel attend.
There was also a question about whether this Board or the Board chosen by the new administration elected in November be the one to receive the report. Mr. Foster felt the new Board should be the one, but there was no resolution on this point.
The Panel unanimously thanked the people who made presentations this morning. Mr. Winter suggested that each be sent an official letter of thanks signed by all members of the panel.
At this point, the meeting was adjourned at 1:05 P. M.