RIETI-JSTAR Symposium

# Japan's Future as a Super Aging Society: International comparison of JSTAR datasets (Summary)

## Summary

In order to search for desirable social security policy in light of a rapidly aging population in recent years, it is important to understand the actual situation of the elderly in a more accurate manner. RIETI launched in 2007 the Japanese Study of Aging and Retirement (JSTAR), a large panel data survey of the elderly. Until now, four waves of the survey have been conducted. In this symposium, after reviewing developments in the study, cutting-edge researchers who are playing leading roles in similar studies in such countries as the United States, the United Kingdom and also in Europe provided presentations to help us get the real picture of the Japanese elderly through international comparisons. Participants also discussed how the Japanese social security system should be reformed.

### Opening Remarks

• Atsushi Nakajima (Chairman, RIETI)

Japan's population is aging at the fastest pace among developed countries. To establish a sustainable social security system in the country, multifaceted data that cover the actual situation of the elderly are essential as a policy infrastructure. The Japanese Study of Aging and Retirement (JSTAR), launched by RIETI in 2007, is designed to enable us to make international comparisons with similar studies being conducted in major countries. By analyzing these datasets, the characteristics of the Japanese elderly can be identified.

In this symposium, the actual situation of the elderly in various countries will be revealed from the findings of panel studies and presentations will be provided on how policy is formulated in respective countries based on the results of analysis. There will also be a discussion about viewpoints that are essential to Japan's future social security reform. With population aging in major countries, it is becoming increasingly important to analyze the actual situation of the elderly and make effective policy responses, including the enhancement of the social security system. JSTAR data have been finally accumulated and more studies are expected in the future. I believe that today's discussion will help broaden your knowledge on this subject.

#### Special Greeting

• Masaru Yoshitomi (Special Adviser, RIETI)
Read by: Asako Yoshitomi (Professor, Tokyo University of Foreign Studies, Daughter of Masaru Yoshitomi)

A decade has passed since we launched the "New Economics of Aging" Project at RIETI in 2005. The rapid aging of Japan calls for action to keep the pension account solvent, an issue shared among developed countries around the world. To design a policy response, decisions reflecting people's diversity and incentive mechanisms must be considered. Thus, the first wave of JSTAR was implemented in 2007 to obtain large-scale, multidisciplinary data based on a longitudinal survey with the aim of understanding the decision-making processes of older individuals.

I would like to express my appreciation for the work by the JSTAR team and for the support from colleagues in the Health and Retirement Study (HRS), English Longitudinal Study of Ageing (ELSA), Survey of Health, Ageing and Retirement in Europe (SHARE), Korean Longitudinal Study of Ageing (KLoSA), China Health and Retirement Longitudinal Study (CHARLS), and Longitudinal Aging Study in India (LASI). JSTAR has benefited greatly from the ideas generated by their efforts. I also extend my gratitude to the interviewers, interviewees, and the 10 municipalities that participate in JSTAR, who have gathered baseline data from 8,000 individuals.

Although the initial objectives of JSTAR have been accomplished in that we now have a better understanding of the aging population in Japan, three directions should be emphasized for the next stage of progress. First, more policy-oriented research must be conducted in the real world, especially in Japan. In contrast to the United States where HRS has been directly involved in social security policy making, we see little evidence-based policy making in Japan. Second, more international comparison studies should be conducted. Since data from JSTAR and related surveys now cover a substantial population and are internationally comparable, collaborative research, which would benefit many counties, should be encouraged. Lastly, measures need to be taken to secure more stable funding. Publicizing our efforts and gaining public understanding could help in this respect.

I trust this conference would be a precious opportunity to disseminate the significance of the research endeavors by JSTAR and related international studies.

#### Keynote Speech: Recent Research Development of JSTAR and its Implications for Social Security Policy

• Hidehiko Ichimura (Faculty Fellow, RIETI/Professor of Economics, Graduate School of Public Policy, The University of Tokyo)

I will talk about recent research conducted using JSTAR to explain the contributions that detailed micro-level data with a panel structure can deliver. I will first explain the JSTAR study and then go through three examples. One is on labor supply and retirement behavior, the second is on asset and bequests, and the third one is medical service demand and copayment relationships.

JSTAR started in 2005 under President Masaru Yoshitomi's leadership with much methodological help from the HRS, SHARE, ELSA, and KLoSA group leaders. The first wave in 2007 covered stratified random samples within Takikawa, Sendai, Kanazawa, Adachi, and Shirakawa. The survey conducted in 2009 added Naha and Tosu. The one conducted in 2011 added Chofu, Hiroshima, and Tondabayashi. The fourth round continued with those 10 municipalities. We had difficulty securing funding for the fourth wave, but RIETI generously provided it.

The first wave response rate was around 60% while subsequent rounds were around 80%, the highest among Japanese panel data. It was comparable to the low response rate countries in the SHARE study. However, the response rate is lower than the record-holding HRS study, so more effort is required.

Municipalities were chosen to make sure a variety of municipalities are covered, but also based on whether survey results could be linked with health expenditure records. Municipality-based sampling is advantageous because many individuals face a uniform socio-economic environment that can be identified without asking them. Many research questions relate to how the environment impacts individual decisions, and JSTAR takes a random sample from the same environment within each municipality. However, this design means that a nationally representative sample needs to be reconstructed. To see if this is feasible, so far we have experimented with various weighting schemes using published census data to construct a nationally representative sample.

We looked at household deposits, household salary, and consumption. Regarding deposits, we asked participants to provide an exact number or a range with a lower and upper bound. Using these bounds plus the exact numbers provided, those aged 50 to 59 had a lower bound of 8,368,676 yen and an upper bound of 8,609,847 yen. For those aged 60 to 69, the corresponding figures were 10,237,468 yen and 11,100,772 yen. These figures were in line with the 2009 National Survey of Family Income and Expenditure (NSFIE). Looking at monthly income, the wages and salaries of the household heads were similar to the 2009 NSFIE. The same pattern was found for household consumption. Using individual level data, we will be able to improve this match. Also, if we find some cells that are not contained in our sample, we will then try to sample them by adding new municipalities. We do not have many fishermen in our sample, for example, and if their behavior is different, then we should seek out this data. Doing so, we will be able to further improve our national representativeness.

JSTAR was largely designed after SHARE and is mostly common across HRS family surveys. Section A contains individual and family characteristics including age, gender, marital status, education level, family composition, information on their parents, frequency of contact with parents, and whether care is shared with others. Section B contains cognitive ability including word recall, repeated subtraction, percentage calculations, measures of discount factors, and risk aversion. Section C contains work information including hours worked and earnings, content of work, job satisfaction, whether there is a retirement age, expected retirement age, and the possibility of employment after retirement. Section D covers health including self-assessments of health, activity of daily living (ADL), and instrumental activities of daily living (IADL). Section E contains income, consumption, and durables information. Section F covers grip strength. Section G is about the housing situation and assets. Finally, Section H covers medical treatments and the use of care services. Section H also covers health check-ups, dentist visits, whether care needs are officially assessed, timing and level of care, reason for needing care, care service usage, expenditure, care given by family and acquaintances, and care given to family members.

JSTAR differs from other HRS-type surveys in that food intake is measured with a questionnaire validated for Japan, we can directly obtain health and nursing care usage information from the municipalities, and we can link our data with information from health examinations. What is unique in Japan is that all these pieces of information are linked with socio-economic and family background data as detailed above.

I will now focus on examples of research carried out using JSTAR. Three implications of aging include labor shortages, a deterioration of the pay-as-you-go pension system, and increased healthcare expenditures.

About labor supply, Japanese people tend to remain in the labor market longer compared to other countries in the world. Labor force participation above the age of 60 is much higher than other Organisation for Economic Co-operation and Development (OECD) countries. In 1980, 65% of males aged 65 to 69 in Japan worked, which dropped to 49% in 2007. Although the participation rate has fallen in 27 years, it is still much higher than in other countries. We need to know the reasons behind this. The labor force participation rate for women up to the age 50 is not very high, but above age 65, it is very high.

Examining labor supply, Ichimura and Shimizutani in 2012 looked at the impacts of health, family-related, and socio-economic variables. This is the first time in Japan where the impacts of all three aspects on the retirement process are examined simultaneously. As Banks and Smith showed in 2006, the retirement of elderly Japanese is a process rather than an event where people reduce their hours worked over time and then retire (see Fig. 1 and 2). The process differs mainly due to family factors for men and socio-economic factors for women. We can assess how these aspects affect whether a person is retired in two years' time based on their current employment situation.

Japanese men who are well off in these three aspects have a higher chance of retirement after two years than those who are disadvantaged for the age groups of 50 to 59, 60 to 64, and 70 to 74. The pattern is opposite for men aged 65 to 69 where, if you are not well off, you are more likely to be retired. For women aged 50 to 59 and 70 to 74, if you are well off, you are more likely to be retired. However, for women 60 to 69, the probability of retirement is higher if you are not well off. This can be interpreted that people cannot continue working if they are disadvantaged.

For men, better indices lead to higher probability of retirement at all ages except for those 65 to 69. For women, higher indices lead to a higher probability of retirement at ages 50 to 59 and 70 to 74, but for ages 60 to 69, there is a lower probability of retirement. If the pension eligibility age is raised above 65, men and women with lower indicators in all three aspects will likely be more affected.

When the pension age is raised above 65, is it feasible for people to continue working? Usui, Oshio, and Shimizutani in 2014 have attempted to quantify the work capacity of the elderly in Japan. In their estimation, the retirement rate for males working full time aged 60 to 64 is 20.5% and 17.6% for part-timers. However, based on the health status of people aged 50 to 59, the predicted retirement rate is just 4.7% and the full-time employment rate is 88.2%. For those aged 65 to 69, the current full-time retirement rate is 29.3%, but 84.4% are predicted to be able to work (see Table 1). There is a huge potential in the labor supply if health is the only factor.

Data from women in their 50s may not indicate the feasibility of working. There are many socio-economic reasons why women in their 50s do not work. We cannot force people to work if they are not able to, so the pension system should be designed such that the eligibility age is self-selected. Early retirement penalties should depend on family, health, and socio-economic conditions and not just age.

They also showed that, for those above 70 years of age, 12% have a negative net lifetime wealth, a huge fraction. It is even higher for those aged 50 to 59. This is a serious issue. Looking at the importance of pension payments, raising the eligibility age from 65 by one year means that individuals must save the equivalent of 50% of annual expenditure and this percentage depends on the household type (see Fig 3).

Can bequests relieve the pressure on pay-as-you-go retirement systems? Sanchez-Romero, Ogawa, and Matsukura in 2013 showed that the annual bequest flow as a fraction of gross domestic product (GDP) per year is currently 6% and is predicted to increase to 9% in 2050 (see Fig 4). These figures are not small compared to medical expenditures at 10.3% of GDP, pensions at 11.2%, and care services at 1.8%. The main driver is the death of wealthy individuals as a fraction of the population, which is one benefit of having an aging society since people with wealth die and leave bequests. Therefore, we can expect to have a modest reprieve for the next 50 to 60 years. We need to squeeze from JSTAR just how much bequests can add to tax revenue.

Looking at the demand of medical services if the health insurance co-payment is lowered from 30% to 10% when a person turns age 70, Ibuka in 2014 found that a reduction leads to a service use increase for specific illnesses. For those with joint disease, service usage increases by three times per month if the co-payment is reduced. Increases are also found in liver disease, ear disease, diabetes, high blood pressure, and hyperlipidemia. The extent of the increase differed depending on the patient's income. The results suggest that increasing co-payments will decrease medical service demand.

In summary, the research illustrates how JSTAR has been used. However, these results are tentative and more work needs to be done. However, I believe that the results give us an image of how social security policy should take shape. They help us go beyond concerns about budgetary implications of an aging society and to begin to think about how we should face an aging society.

#### Research Findings from International Comparisons with SHARE

• Axel Börsch-Supan (Director, Munich Center for the Economics of Aging (MEA) at the Max-Planck Institute for Social Law and Social Policy)

SHARE is different from HRS and JSTAR because it is a multi-country study designed to compare the economic policies of 21 countries across Europe including Israel. European countries have different policies so one can learn much by comparing them. On an individual country level, the methodology is very similar to HRS and JSTAR in that we collect health, socioeconomic, and social participation data.

The main design challenge of SHARE is that it includes different countries. Institutions, languages, and interpretations of words all differ. The organization is complex with five management centers, a survey agency in each country, and 2,000 interviewers. We tried to reduce this complexity by using information technology tools and by measuring as objectively as possible. This is important because perceptions of health are overrated in some European countries and underrated in others.

Young unemployed people show health effects in 10 or 20 years. This effect is worse in countries without unemployment insurance. SHARE looks at the current state of people, their upbringing, and looks retrospectively at their lives. Retirement depends on health, social, and work environments. People retire later if they enjoy their work. Women in countries with maternity benefits retire later than in countries without them. Countries where the language lacks a future tense have different savings patterns as compared to those with a future tense. Also, as European countries have aged, there is a trend toward less dependence on the younger generation by the old.

SHARE data have refuted a number of prejudices. One is that health declines dramatically after the age of 60. However, looking at self-rated, functional, or objectively measured health, the decline is slow and smooth. Indeed, the variation within each age group is much larger than the decline between the ages of 60 and 69. The healthiest at age 69 are much healthier than those who are unhealthy at the age of 59. Some diseases are more prevalent among the uneducated. Indeed, for diabetes and stroke, it is 60% more prevalent among less educated women, and 90% more prevalent among less educated men. The socio-economic and biomedical data from SHARE allow for this kind of analysis. There is positive correlation between the amount spent on healthcare as a percentage of gross domestic product (GDP) and the level of health of a country. However, some countries spend more on healthcare than others at the same level of health, which indicates inefficiency in the healthcare system.

A second prejudice is that it is best to retire as early as possible. However, cognitive scores show that those who retire early have worse scores than those who retire later. Employment status and the number of friends are also positively correlated with cognition. Changes in retirement policy in different countries can show effects on memory ability, allowing for determination of the causal direction. Various contributing mechanisms could mean that retirement is bad for the brain. If one does not use his or her brain at work, the biggest effect, at 64%, shows that memory ability is likely to be lost. Also, friendships are lost faster if memory ability is lost. One should think twice about retiring too early.

The third prejudice is that there will be a war between generations with the young wanting more from the elderly who have taken pension expenditures. However, there is a lot of intergenerational exchange in Europe indicating that this war is overhyped. The data of children who say that they have conflict with their parents do not match areas where parents say they have conflict with children. Looking at 22 dimensions of family interaction and social and intergenerational cohesion, those that are in conflict are a minority. Cohesion is even stronger in the older countries, so there is no war between generations in older/ailing Europe.

Another prejudice is that families care less for themselves if they receive help from the state. However, there is almost always a significant positive correlation between spending on families and the families taking care of themselves, which contradicts economic intuition.

In summary, SHARE is particularly helpful because it looks at various policies in various countries of various political leanings, which allows the direction of causality to be determined. The data also contradict prejudices that people may have about an aging population. If these trends inside Europe can be shown, then the same trends throughout the globe can also be seen.

#### The Lifetime Risk of Nursing Home Use and Lifetime Out-of-pocket Spending

• Michael Hurd (Director, RAND Center for the Study of Aging)

The Health and Retirement Study (HRS) interviews around 20,000 people over the age of 50 every two years with more added after six years to keep it representative of the population. The HRS is a rich long-running dataset that can increase our understanding of what happens in old age while also addressing policy problems.

In Japan in 1970, between 6% and 7% of the population was above age 65, but this is projected to be around 38% in 2050. In comparison, the United States has benefited from higher fertility rates and more immigration, but the number of elderly will increase due to the baby boomers aging as well as reductions in the fertility rate. Therefore, long-term care will increase.

In the United States, nursing homes are expensive at $85,000 per year. We would like to know the chance of ending up in a nursing home and out-of-pocket costs. Nursing home use is likely to be the same in Europe and the United States. Also, nursing homes are all but required for those with advanced Alzheimer's, so rates in the United States and Japan should be similar for that group. The United States has Medicare, which is almost universal health insurance for those over 65, but it only partially pays for short-term hospitalizations. Medicaid pays for long-term care, but people must pay out of pocket until their assets are depleted unless they have insurance. At baseline, the HRS does not sample from those in nursing homes, but as the HRS follows everybody regardless of where they live, as time passes, all those who will enter a nursing home will be interviewed by HRS, and HRS will represent the nursing home population. HRS also has exit interviews, so we can find out information about nursing home use up until the time of their death. We use the Assets and Health Dynamics Among the Oldest Old (AHEAD) and HRS cohorts in our analysis for a total of around 18,000 people. For those above the age of 90, nearly 20% moved into nursing homes between waves one and two. By wave three, 25% of those aged 90 or above had moved into nursing homes. The rate remained about the same for subsequent waves indicating stabilization. The percentage in nursing homes increases with age with practically no nursing home use for those below the age of 55, but with more than 50% use between interviews for those aged 95 and above. Use increases closer to death, so without exit interviews, the data would be underestimated. In the end, we find a nonparametric estimation that the lifetime risk beyond age 56 of any nursing home stay is 56%. A model-based simulation had a result of 57.7%. Therefore, both nonparametric and model-based simulations are consistent. The mean number of nights spent at a nursing home is quite high at 283. Women are more likely to spend time in nursing homes because they live longer than men. People with less education or who are smokers spend less time in nursing homes because they die earlier. Regarding out-of-pocket nursing home costs, its population average, and distribution, we used the same methods as for time spent in nursing homes. We found that there is a skewed distribution with some paying almost nothing out-of-pocket, while others paying a lot. At the younger ages of 70 or 71, about 70% is covered by insurance since they are short-term stays covered by Medicare. That percentage declines at older ages. Two-year out-of-pocket spending sharply increases with age. Those aged 95 or above spend over$10,000 per year. With a discount rate of 0% to age 57, a person's average rest-of-lifetime out-of-pocket spending in real 2013 dollars is $18,000. At the 95th percentile with 0% discounting, the amount paid out of pocket is more than$100,000.

Around 27% of people at age 56 should expect to have a long-term stay at a nursing home that is longer than 100 days. 7% would likely experience two long-term stays, and 2.4% have three or more. Women are at greater risk because they live longer and are more likely not to have a spouse to look after them.

In conclusion, the lifetime risk of nursing home use in the United States is between 56% and 58%. The most frequent pattern is of the wife taking care of the husband, and after widowing she lives independently until she enters a nursing home and spends down her savings until she is eligible for Medicaid. This makes Medicaid a better long-term choice than long-term care insurance.

#### Research Findings Using HRS-typed Datasets: The Case of Asia

• Albert Park (Professor, The Hong Kong University of Science and Technology)

I will talk about the China Health and Retirement Longitudinal Study (CHARLS). Nearly 25% of Chinese elderly live in poverty, and China is rapidly aging. By 2050, over one-third of the people will be over age 60, and 10% will be over age 80.

The national baseline survey in 2011 included 18,000 respondents. The data from a follow-up national baseline survey in 2013 will be released soon. We searched for eligible households with at least one member age 45 or older. Our response rate is 80% at baseline, and the follow up rate is close to 90%. The CHARLS dataset has received a lot of attention because it is one of the few in China that is collected in accordance with international standards and is publicly available. A compiled report included descriptive numbers of consumption and poverty based on income by age. For the 60 or older age group, poverty rates were higher than for the 45 to 59 age group. The national baseline survey received much attention because the Chinese government has never released poverty rates by age group.

Additionally, the facts about the wellbeing of the elderly are fascinating. Only 46.9% of elderly people received transfers from non-coresident children. This counteracts mythological ideas that families always take care of the elderly in East Asian societies. For those receiving transfers, the average amount is 1,700 renminbi per year and accounts for 37.3% of elderly household expenditure indicating that transfers target the poor elderly. We found that Chinese elderly are in very poor health: 38% percent have difficulty with daily activities; one-third report bodily pain; 32% report poor self-reported health; and 40% of Chinese elderly experience depressive symptoms, with women experiencing especially high rates. These rates are much higher than in the Organisation for Economic Co-operation and Development (OECD) countries. China has expanded health insurance programs covering 66% of the population and covers nearly all rural people. However, the quality of benefits provided is relatively modest.

The New Rural Pension Program (NRPP) had not been expanded to the whole country when the baseline was taken, so we could compare areas that received benefits with those that did not. Pension payments increased the probability of not working by 25% while alleviating poverty. Weak evidence shows that it has crowded out private transfers and reduced depression.

In many countries, health expenditures increase dramatically with age. In the urban areas of China, healthcare expenditures increase rapidly as people age, but rural spending actually declines even though individuals become less healthy as they age. Elderly living with a child in the house are even less likely to spend as they get older, suggesting that they choose not to use scarce resources on their own healthcare. With good quality insurance, there is not as great a fall in expenditures as a person ages in rural areas, but, even controlling for that, the difference between urban and rural persists, raising concerns that rural people are not benefiting from healthcare services as their health deteriorates with age.

We are working to compare the wellbeing of elderly in China, Korea, and Japan. We focused on depression because China and Korea have some of the highest elderly suicide rates in the world. The frequency of elevated depressive symptoms is highest in China, second in Korea, and third in Japan. However, the United States has elevated depressive symptoms at between 11% and 13%, making it close to Japan. Looking at factors explaining depressive symptoms, there are inconsistencies across countries. In Korea, there is a sharp increase in depressive symptoms as one ages, but it is the opposite in Japan. Physical health also provides the largest explanation for depression across all three countries, along with economic factors. Education is a predictor of depression in China and Korea, but less so in Japan because its elderly are more educated.

Finally, we compared China and England. Interestingly, for physical health, the gradient for wealth inequality was much stronger in England than in China, which could be related to China's socialist legacy. However, we found that wealth had a larger impact on subjective wellbeing in China than in the United Kingdom. Relative wealth may be more important than absolute wealth with subjective wellbeing more reflecting current circumstances that change based on recent experiences.

#### Policy Effectiveness: The Case of the United States

• Robin Lumsdaine (Professor, American University)

In addition to the more than 2,500 academic publications that have been based on HRS, the data also have been useful for informing U.S. policy. In the United States, there has been much policy debate focused on how changing demographics will impact the government and budgets. Increasing longevity will impact the acquisition of sufficient social security funding. Some early proposals to address this problem proposed shifting the responsibility for savings to the individual, while others sought to increase the retirement age. A number of studies conducted by the U.S. Government Accountability Office (GAO), an independent, nonpartisan agency that assists Congress in evaluating how the government spends taxpayer dollars, studied this topic.

A study based on the original HRS cohort showed that although women live longer than men, they receive significantly less social security and therefore are more likely to be in poverty. A GAO study using the HRS found that women's benefits tend to be lower due to lower wages and fewer years in the workforce. Based on these findings, the study argued that proposals to encourage retirement savings through accounts based on a proportion of salary could exacerbate this gap if in addition there are differences in investment behavior between men and women. Recommendations were made to provide information about investment objectives and financial planning, consider mandatory annuitization to reduce the possibility that a person outlives their savings, and to use lifetables (in the calculation of annuities) that do not distinguish between men and women so that those with comparable savings at retirement receive comparable monthly benefits from their annuities.

A second study looked at whether changes in social security requirements have encouraged later retirement. Using the HRS study, the GAO found that policy creates mixed incentives and recommended encouraging work at older ages. A benefit of the HRS study is that one can compare differential effects across programs and see how they interact with each other.

A third study looked at the likelihood that people would continue working if the early retirement age is raised. Current policy in the United States has the normal retirement age increasing from 65 to 67 while the early retirement age remains unchanged at 62. Increasing the early retirement age would affect a small group of people who cannot work past age 62 and would then apply for disability insurance, which is more expensive to administer, thus offsetting some of the savings implied by increasing the retirement age.

In 2003 and again in 2011, the GAO considered whether individuals are equipped to face increasing responsibility to ensure retirement income security. They found although much information is provided by private pension plan sponsors about payout options, individuals could not interpret the associated risks and implications for lifetime income adequacy. The GAO recommended increased public awareness (e.g., via something equivalent to the annual mailings from the Social Security Administration (SSA)) and improved financial literacy.

Two GAO studies in 2001 and 2005 emphasized that older workers are an important resource of which few companies take advantage. The GAO also recommended focusing on retention and recruitment of older workers, as well as accommodations that allow them to continue working. The Department of Labor agreed and instituted a number of initiatives to encourage employment of older workers.

In 2006, the GAO answered the question as to whether baby boomers shifting their savings from stocks to bonds would cause a stock market meltdown. They found that a crash is unlikely, one reason being that many boomers have few financial assets to shift. Another is that increasing life expectancy meant that boomers would work longer and spread out their asset sales over a longer period of time. The GAO also noted that the decision to retire is affected not only by social security benefits, but also on the availability of pension benefits and retiree health insurance.

While at the Federal Reserve, I learned the importance of having available a large longitudinal dataset (i.e., one that contains information about the same people over a long span of time). Having data from "normal" times to compare with times of crisis enables us to isolate the effects of a recession from other changes over time. For example, the HRS allows us to look at changes between permanent and temporary layoffs, firm-induced changes in compensation, how policies for older workers affect employment, pension plan changes or terminations, and the effects of involuntary reductions in hours. We also are able to evaluate how changes in health or family care responsibilities can affect labor force participation. In addition, the consistency of questions across HRS, ELSA, SHARE, and JSTARS enables us to conduct important cross-national comparisons of such effects.

The HRS dataset also has been useful for analyzing the financial effects of the Great Recession on the elderly in the United States. Although many initially feared that the declines in the stock market and housing prices would have a large effect on the elderly, analysis of the HRS found that many elderly were cushioned from the effects, because 29% of their wealth was in social security and another 17% in defined benefit pension plans. In addition, most near-retirees were not affected by declining home values because they no longer had mortgages.

Analysis of longitudinal data can identify data needs and identify policy concerns before they become a problem. In March 2003, the GAO published a report using the HRS that found that more data are needed about administrative and employer records, as well as linkages to other datasets. Their recommendations have helped enhanced HRS, through private employers' summary pension plan descriptions and linkages with SSA records. Enhancements such as these have improved our ability to look at how social security and private pension plan provisions affect retirement decisions.

For example, recent work has used the linkages with SSA data to help identify the extent of public misinformation. Only about 50% of respondents could identify their pension plan type or age of eligibility. Similarly, only 50% could make a guess about the amount of their expected social security benefits, and only half of those could guess within $1,500 of their actual benefit. The policy response is for increased information, financial literacy initiatives, and annuitization. In conclusion, in its 22-year history, the HRS has been invaluable in identifying areas that need policy response, as well in evaluating the effects of policies that have been enacted. Finally, the longitudinal aspect is critical, especially as the population ages. #### Policy Effectiveness and Usage: The Case of the United Kingdom • James Banks (Professor, University of Manchester) The English Longitudinal Study of Ageing (ELSA) started in 2002. Half of the funding comes from the government agency running policy, so in implementing ELSA, we talked with the government agency about what it would like to know and how it would use the information. The ELSA sample is representative of the population aged 50 and over, is a longitudinal study, and is open access. Every four years, we have a one-hour nurse visit to evaluate health and collect biomarkers. So far, we have interviewed about 18,000 respondents. ELSA has been used by the government to make calculations, inform inquiries on designing policy, and provide a base for research papers. In the United Kingdom, the responsibility for retirement savings has shifted from the state to the individual. We are also increasing the retirement age, restructuring disability benefits for older workers, and removed mandatory annuitization. Pension reforms have cut the generosity of public pensions making the aging population more affordable than elsewhere in Europe. Over the next 10 years, the retirement age will rapidly increase for women to bring their retirement age in line with men. However, many women do not understand the details of the reform, and 80% could not correctly state their pension age. Indeed, some women who could answer correctly one year did not answer correctly the next. There is a lot of confusion. ELSA was the first study in the United Kingdom to measure all aspects of wealth in order to forecast retirement income. One benchmark the UK government used for retirement income adequacy is two-thirds of one's working income. If everyone between age 50 and retirement age stopped working immediately after ELSA wave 1, then 40% could not have adequate income, but this depends on what assets are counted as generating income. We collect social security numbers and link government records to provide an optimal model of savings and spending. ELSA is used to measure healthcare quality. The percentage of people receiving adequate care varies by disease type, but rates remained constant over time. Some conditions such as pain, osteoarthritis, falls, and vision are poorly managed. We provide good care for medical conditions, but provide poor care on geriatric conditions. Our healthcare system provides financial incentives to doctors for providing treatment for certain diseases, particularly medical conditions. For these diseases, the quality of care is better. The British healthcare system quality of care does not really vary by wealth. For diabetes care, those in the poorest wealth group have the same quality of care as the wealthiest group. One reform is called "Digital by Default" where government services and welfare payments are to be delivered online rather than in benefits offices. For 50 to 54-year-old men, 36% at baseline did not have access to the internet or email. However, over the course of the panel, 70% had gone online. For women aged 65 to 69, 77% had no access, and only 30% of them went online. It will be difficult to provide online services. Our Office of National Statistics (ONS) measures wellbeing and quality of life in addition to finances and health. About loneliness, poor people spend the most time alone, and women spend more time alone than men. Comparing happiness and loneliness, the richest are the happiest when they are alone, meaning that the poorest are the most likely to be alone, and are also the most likely to be unhappy when alone. Wellbeing and life enjoyment are strong predictors of survival. Those with the lowest enjoyment have a 27.6% chance of dying over 10 years, as compared to 9.9% for those with the highest. Descriptive analysis is unsatisfactory in designing new policies since the government needs to know the effects of certain policy changes. However, a fully structured model of behavior is often not detailed or specific enough, and also sometimes beyond the capacity of government officials to understand. One middle ground is dynamic micro-simulations which can be used to look at mortality, health, disability, income and poverty in retirement. A simulation of the population going forward can be done to compare different policy regimes. We are predicting a large rise in the number of workers due to the increase in the retirement age, and that, between now and 2020, the income of the richest group will increase substantially and pensioner poverty will fall. In conclusion, ELSA is successful because it is being used to inform policy because links between statistics are just as important as the statistics themselves. In Britain, our government is constantly updating policies so there is always a need for more evidence. As the survey progresses, it can inform more effective policy. #### Panel Discussion: Evidence from Social Security Policy--Implications for Japan's Social Security Policy • Moderator: Yasuyuki Sawada (Faculty Fellow, RIETI/Professor, Graduate School of Economics, The University of Tokyo) Sawada: Professor Mitsuhiro Fukao will present on the Japanese experience and policy issues. Mitsuhiro Fukao (Program Director, RIETI/ Professor, Keio University): The Japanese government-sponsored pension system has been mostly stabilized by cutting benefit levels and by switching benefit increases from a wage inflation index to a price inflation index. However, by cutting pension payments for the very old, people may rely on social assistance programs. Health insurance costs for the elderly are between five and seven times higher than other age groups, so the system may not be sustainable. Long-term care insurance also has a skewed benefit distribution. If a doctor removes the feeding tube of a patient who is long unresponsive, he or she may be arrested for homicide, leading to prolonged care. We need to change the practice of medical intervention in very old people. The older generation is placing a heavy burden on the young. The working age population of 15 to 64 will decline by 17% over the next 20 years, whereas the old-age population above age 65 will increase by 35%. The current administration is trying to increase labor participation for female workers while increasing the fertility rate, but it would take 20 years for children to participate in the economy. Immigration is another possibility, but we only accept about 30,000 people per year because they have to be either rich or well-educated. Germany accepts more than 10 times this number, and its economy is growing rapidly. If a person's mental health deteriorates, it is their spouse or families who must make decisions about placing them in a nursing home or for certain medical interventions. Therefore, we have to educate and change the understanding of death and aging in the general public. Sawada: Politicians should stand up and ask how poor people can become healthier without spending more while also questioning the policy impacts of changing pension eligibility ages. What concrete policy instruments should politicians pursue to make the elderly healthier? Also, what happens to elderly people if we increase the retirement age from 65 to 70? I ask each participant to answer these questions as well as the questions from the floor. Ichimura: A question from the floor points out the importance of having third party evaluations of government numbers. One important aspect of HRS-like data is that they are available publicly, so anyone can evaluate government numbers. Having access to panel data allows us to move beyond focusing on the budgetary aspect of aging towards concrete policy steps. That is illustrated in the heterogeneity in the retirement process. Usui-san, Shimizutani-san, and Oshio-san talked about labor shortage issues, and if their calculation is correct and we take the scenario for men, the 65 to 70 age group provides potential for expanding the labor supply. Women also represent an underutilized workforce. Börsch-Supan: There is no single policy that will do the trick. There are three distinct causes of the demographic revolution. One is increasing longevity, another is declining fertility, and the third is the effect of the baby boom/bust transition. Each cause requires a different policy response. If we live longer and stay healthier, then retirement age should go up. Also, if health and longevity both go up proportionately, then problems such as the uptake of disability insurance are second order problems. Therefore, I propose a proportionality rule where the share of time spent working stays constant with respect to the time spent in retirement in a 2:1 ratio. Fertility rates are difficult to change through public policy, and there is little evidence as to which policies work and which do not. Therefore, one should aim for quality over quantity by improving education. Unfortunately, we see in Europe the exact opposite with cuts for education without touching pensions, which is perfectly wrong. About the baby boom/bust, economics tells us that a small cohort of children and a large cohort of retirees mean that retirees will have to fund a substantial part of their own pensions. That has been done in countries such as the United Kingdom, but Japan and others have done little. Sawada: Data from HRS-like surveys are indispensable in crafting policy. Börsch-Supan: It is indispensable that one understands the side effects of a shift in retirement age. A shift in the retirement age from 65 to 67 does not mean that disability claims will triple because people are healthy at those ages. Germany's pension rates are being reduced by 15%, so those close to poverty will drift into it. Data show how many need poverty prevention programs. Understanding policy effects is important in gaining voter support. Hurd: Answering a question from the floor, our dementia estimates come from the Aging, Demographic, and Memory Study (ADMS), which is the only study in the United States that is population representative. Answering another question from the audience, there is much variation in the annual cost of nursing homes. In New York, it might be$120,000, while in Arkansas it might be \$40,000.

The cost of dementia study featured on the front page of the New York Times led policy makers to understand the need for research in how to prevent dementia and how to care for people cost-effectively. Countries with aging populations will experience escalating costs since people with dementia cannot be cared for in the home.

The Gruber-Wise project established that public policy has a large effect on retirement behavior, and well-founded scientific studies influence public policy. The "lump of labor fallacy" states that an older worker has to leave for a position to open up, but it is not true in the economy.

Although the labor supply of the older population in Japan has been falling, in the United States, it has increased substantially. The cause is unclear, but it shows that labor force participation can increase in a population with stable health.

Obesity is a problem worldwide leading to health problems such as diabetes. Policy has a role to play, similar to how it affected smoking rates, which have declined leading to improvements in health. We need public policy to address health problems to allow people to work longer.

Park: A comment from the audience suggested that the distance to a hospital might affect rural and urban populations differently. However, we have controlled for that, so this explanation cannot explain our findings.

The second comment was about the high female depression and suicide rate in China. Empirical analysis suggests that it is due to educational gaps between women and men in rural China, as well as gaps in physical health. Evidence suggests that differences between women and men are declining, and many attribute that to higher labor force involvement and wealth.

There are two worlds of rural and urban China. Labor force participation and healthcare in urban areas resemble European countries, while rural China looks like a developing country. The government wants to reduce those differences. The retirement age in urban China must increase, where for women it can be as low as 45.

China needs to reduce inequality. To expand rural coverage, China instituted universal healthcare not linked to employment, whereas in urban areas you have employer-provided pension and health programs. For a worker migrating from rural to urban, it creates a disincentive to participate in employment-based programs. It is going to be a slow process moving towards universal healthcare not linked to residence status.

The impact of the rural pension program has been presented to Chinese government officials. Many government officials want data to assist fine tuning policy, so they are receptive to sound analysis. The World Bank is also pushing its own results to the Chinese government, so they are hearing it from multiple sources.

Lumsdaine: Increasing the retirement age in the United States would lead to an increase in disability applications, and this has informed policy. Changes in the retirement age interact with the social policies of each country, and studies show the heterogeneity of experience.

The HRS has also informed U.S. policy in the interaction between retirement policy and caregiving. During the recent financial crisis, many grandparents cared for grandchildren to allow the middle generation to participate more fully in the workforce. Using HRS data on how caring for grandchildren affected a woman's retirement choices, it was found that family characteristics affect retirement decisions, whereas for the caregiving decision, financial aspects are not influential. Policies aimed at affecting retirement decisions may have less of an effect if caregiving is driving retirement decisions.

Sawada: I have an interesting question from the audience. Is there any evidence using ELSA that explores the impact of the Summer Olympic Games in London on the wellbeing of the elderly? I think this question was asked because Tokyo will host the Olympic games.

Banks: We have not yet looked at the Summer Olympic Games in London, but our study is partly funded by the Department for Culture, Media & Sport. We do know that that the Olympics had a positive effect with a temporary rise in national wellbeing across all population groups.

One issue with policy for aging populations is that one policy cannot do everything. Different policies are needed to address the problems of current generations of elderly, as opposed to the current younger generations who will become the future generations of elderly.

With regard to health, there is no difference in the treatment rates between rich and poor, but there are differences in mortality, which is linked to education, health behaviors, occupational health risks, cognitive capacity, family background, among other things.

About adequate retirement income, I provided a number of 67% because the government decided on that number, not because it is optimal. To determine "optimal," a model full of dubious assumptions is needed, so the first step should be to ask policymakers what they consider optimal and then make calculations based on that.

Sawada: What model of dissemination is effective?

Banks: We held workshops and tried to attract policy analysts to inform them on how to use the data. We also had senior principal investigators providing seminars at government departments.

Lumsdaine: It is similar in the United States. The data are publicly available, but as they become increasingly longitudinal, they become complicated. There are a number of workshops and seminars to help analysts use HRS. There are also a number of harmonization efforts underway to facilitate cross-country comparisons.

Börsch-Supan: There must be a mix of dissemination policies including academic journals and media exposure. It helps to write newspaper articles and sit in policy advisory positions. The European Union requires a dissemination plan when applying for funding, and we have to write about five pages about it in the proposal.

When the German government decided to reduce the retirement age by four years, SHARE data were readily in place to predict dropout rates and labor force participatio.

The data show the importance of fixing structural problems to help fix economic problems. Greece has the highest replacement rate of pensions in the EU along with the highest elderly poverty rate. A few people receive several huge pensions, while others receive nothing. No monetary policy will help to solve such structural problems.

For Japan, without the third arrow of Abenomics, the first two arrows are unlikely to ever work.

Hurd: RAND received a large grant from the SSA for HRS, and making the data easy to use has been critical in its success there. They became advocates for HRS, and that has spread to other parts of government. There are many studies based on the HRS that would not have been possible without making the data easy to use.

Ichimura: We have not established strong links within the government. We need to understand routine questions that come up in the daily operations of the government and use JSTAR-like data to answer them.

Börsch-Supan: The "routine questions" comment by Professor Ichimura is important. Using SHARE micro-data, we found that in some municipalities the young take care of the elderly for 11 months out of the year, and then the municipalities support the young for one month to give them time off. The family can help more if the municipality helps the family.

Sawada: Related to this comment, there are three actors in improving the livelihood of the elderly in general: market mechanisms, government, and community. Markets often fail and governments come to the rescue, but then governments fail as well. Therefore, community mechanism including family supports should be strengthened to amend market and government failures.

Next, I would like to ask each panelist to address the future challenges or prospects for improving HRS-like datasets. Perhaps we should also trace younger cohorts to talk about people's entire life course. We also face challenges in continuing data collection due to funding constraints.

Banks: The main challenge we face is funding, as the UK government is going through substantial public spending cuts which hit our government funding departments. Funding will become harder because one will be asked, "What can you learn from 12 years of data that you could not from 10?" The second equally important challenge is keeping high scientific quality and high response rates needed to convince funders of the value of these datasets. We need to provide value for money in order to make sure the case for funding is as strong as possible.

The third main challenge is with regard to the role of evidence more generally in impacting policy. Evidence may not sway policymakers to fix long-term problems if they have political reasons for acting in the short term, and this is particularly relevant for issues relating to aging populations and future government liabilities. In the United Kingdom, more than half of eligible voters are over the age of 50, and voting rates are higher for the elderly and it is hard to make reforms if the government feels it has to protect the living standards of older people, particularly since the recent financial crisis and recession.

Lumsdaine: I think that in the United States, there are also large funding challenges. Another challenge is in keeping the sample representative. There are especially challenges with the SHARE data because managing a dataset with 21 countries while ensuring comparability is difficult.

Participants in the HRS receive feedback and thank you letters explaining how the dataset has informed policy, which helps keep respondents engaged and reduces attrition.

Although it would be useful to track a cohort from the time they enter the workforce, it would take too long for the data to become useful in informing policy. Instead, life histories are being collected to help inform policy in the shorter term.

Park: The CHARLS survey is a high quality survey, and, with China's rise, there is much interest in the data. However, we are struggling with funding as we do not have strong ties to government funding departments. Given our high scientific standards and the importance of the CHARLS dataset over time, I am confident that funding problems will be resolved. We received tremendous support from the international community in maintaining our high scientific standards. Our advisory board consists of directors of HRS, ELSA, and SHARE.

I view challenges as opportunities. There is a real need for information in China because the aging problem is hitting fast and is at an early stage of development. Like JSTAR, we are a young survey, and I encourage you to exploit international expertise.

Hurd: It is important to find cognitive measures that allow us to detect early decline. We need to analyze prevention before it can start. We need to look at people in their 50s or 60s, but our data are only useful for those in their 70s and 80s.

We also need linkages to other sets of data. People in the United States sign up to check their data, so you can then validate your own data and relieve the burden on your respondents. For now, there is no substitute for household surveys.

Response rates will become a problem in the future, so we need to know how to better tailor our survey. Using the internet is satisfactory for certain topics or certain populations, but the technology is not ready for what we are doing now. It may become a bit cheaper in the future, but right now it is not all that different.

How we adjust to changing science is what has led to HRS having a three-hour interview. We collected DNA from 3.5 million HRS participants, and this practice was encouraged by the Human Genome Project. We must prevent the survey from going stale as science changes, but we still have limitations on money and time. Keeping the survey relevant longitudinally while still evolving is the biggest challenge that HRS faces.

Börsch-Supan: Everybody yearns for stability in funding, but it actually is a nightmare. Some people suggested ideas such as going to the internet to save money, but it is hard to measure grip strength that way, and it will not save all that much.

Regarding science, the life course approach is absolutely important. We also need data that are as objective as possible. That is why we must include biomedical measures along with socioeconomic measures. We must also link other data with our surveys, such as pension records or medical records.

Ichimura: We also experience funding issues. We would also like to link our dataset to social security data, but the data are stored in Hitachi language so that no one in the ministry can read it. Our problems also include low fertility, so we need to collect data on younger cohorts, but they are struggling to get a 20% response rate.

We also have difficulty in piquing politicians' interest in the survey's scientific aspects. We must maintain the current population to measure the effects of policy. We should make the use of JSTAR data a requirement in evaluating the potential effects of future policy changes. This would help in securing funds because the need for JSTAR data would become automatic.

Sawada: In closing, I will restate the three directions posed by Dr. Masaru Yoshitomi. First, more policy oriented research is needed. Second, more international comparisons are needed. Finally, data collection should be continued despite the challenges.