Mind the Gap! The effect of an increased UK State Pension Age on expected working life of employees

Transcription

1 Mind the Gap! The effect of an increased UK State Pension Age on expected working life of employees Ricky Kanabar 1,3 and Adriaan Kalwij 2,3 DRAFT- PLEASE DO NOT CITE June 29, 2018 Abstract We model how individual s retirement expectations adjust in the wake of significant reforms to state pension age introduced in the 2011 and 2014 Pension Act in the UK. Results show that men do not significantly adjust their expectations upward as would be consistent with the policy objective. Instead we find that the effect of an increase in state pension age widens the gap between an individual s expected retirement age and his state pension age. For women, who experienced a more rapid increase in their state pension age over the sample period, this gap also widens but some adjustment takes place: the expected retirement age increases by 0.3 years for a one-year increase in state pension age. We control for a rich set of individual and household economic and sociodemographic characteristics, including having an occupational pension plan. The widening of the gap between employees expected working life and UK State Pension Age may suggest the need for improved pension communication to avoid employees insufficiently preparing for retirement. JEL classification: J26 Keywords: Retirement, Expectations, United Kingdom Household Longitudinal Study 1 Corresponding author: Department for Social and Policy Sciences, University of Bath,Bath, BA2 7JP,. United Kingdom. R.Kanabar@bath.ac.uk 2 Utrecht University School of Economics, P.O.Box 80125, 3508 TC Utrecht. The Netherlands. a.s.kalwij@uu.nl 3 Network for Studies on Pensions, Ageing and Retirement (Netspar; 1

2 1. Introduction One of the most important decisions an individual makes with respect to the labour market is the decision when to retire and therefore understanding when individuals expect to do so is crucial. Since the 1990s the retirement decision, often defined as a gradual reduction or complete cessation of labour supply (Denton and Spencer, 2009) has gradually come to the forefront of the policy agenda in developed countries due to the widespread recognition of population ageing. In 2018 the UK government released it s Industrial Strategy, a key policy document stating the four most important issues (Grand Challenges) facing the UK; the ageing society being one such Grand Challenge. The fiscal implications of an ageing population have a significant impact on current and future government budgets, for example keeping the State Pension Age (SPA) fixed at 66 in the UK (as opposed to increasing as legislated to 68 by 2045/46) has been estimated to cost 250 billion (DWP, 2017); in response governments have sought to introduce a range of policy reforms to reduce the burden on the state. One of the most widely implemented measures has been to raise the age at which individual s become eligible to claim their state pension (i.e. the State Pension Age, SPA); the principal aim of this policy is to extend individual s working lives (DWP, 2011). Alongside other OECD countries such as Italy the UK started making reforms to the SPA during the 1990s; the first piece of legislation enacted was to equate (and hence increase) female SPA to that of males. Subsequent reforms were then introduced which increased the pace of equalisation and also raised the SPA of both men and women to 66/67 depending on date of birth. A key question therefore is how do individuals intend to respond to such policy changes? 1 Expectations data is one way to analyse such a question and have been shown to play an important role in shaping major economic decisions in the future including the decision to retire (Disney and Tanner, 1999; Chan and Stevens, 2004). We analyse how the 2007, 2011 and 2014 State Pension Acts in the UK affected the gap between men s and women s Expected Retirement Age (ERA) and State Pension Age (SPA) using the United Kingdom Household Panel Study. Due to the timing of the reforms and the cohorts affected (with the exception of certain older women see Cribb, Emmerson and Tetlow, 2016) very little is known about how men and women have or intend to react in terms of their actual labour supply behaviour in response to higher SPA in the UK. Our results suggest individuals do not fully adjust their expected retirement age in response to changes to SPA, moreover the gap between Expected Retirement Age (ERA) and SPA widens in response to an increase in SPA. The magnitude of this widening is almost one for one in the case of men and 0.7 for females, suggesting women make a partial adjustment. We find the results are driven by younger sample members. We control for a rich set of individual and household level characteristics and find occupational pension, income, 1 Another key question is how have individuals responded to changes in SPA in terms of labour supply? Cribb et al. (2016) addresses this issue (for cohorts of females born in the early 1950s only) and finds a positive impact in terms of the employment rates of affected cohorts and also additional spillovers effects in terms of partners labour supply. 2

3 education, job occupation and sector of work are important for determining the gap between ERA and SPA. The findings are of policy relevance and indicate that individuals do not adjust their expected retirement age in line with reforms to the SPA. If the intention of policymakers is to keep individuals in employment for longer or, put another way, delaying retirement, and should they be successful in doing so, then government need to design strategies to improve communication of changes in SPA, for example via information campaigns, to avoid insufficient preparations for retirement. 2 The retirement decision has a wide range of implications for both the individual in terms of consumption, savings and living standards at older ages; and for the state from a taxation and welfare spending perspective (Manski, 2004; Cribb et al. 2016). Therefore understanding how SPA and different economic and sociodemographic characteristics affect retirement expectations is crucial if policymakers wish to ensure that individual s adequately plan for retirement. The rest of the is paper is set out as follows. Section 2 provides an overview of the literature regarding the role of expectations data and the small literature devoted to retirement expectations. Section 3 summarises the main features of the UK pension system. Section 4 describes the United Kingdom Household Panel Survey. Section 5 describes the methodology followed. Section 6 presents estimation results. Section 7 discusses the implications of the estimation results. Section 8 concludes. 2. Literature review The use of expectations data has been shown to play an important role in explaining major lifecycle decisions related to consumption, savings and of relevance for this paper, retirement decisions (Manksi, 2004; Chan and Stevens, 2004; Benitez-Silva and Dwyer, 2005; Botazzi, Japelli and Padula, 2006). The seminal work of Bernheim (1989) using the US Retirement History Survey (RHS) showed that the arrival of new information close to retirement leads individuals to revise retirement age expectations in a manner which is consistent with the standard lifecycle model. In related work, Disney and Tanner (1999) used the UK Retirement History Survey to analyse individuals expected versus actual retirement ages and studied whether expectations data plays a role in improving the accuracy of retirement models. Their findings show that most individuals reported their expected retirement age (ERA) to be the SPA at the time (and did subsequently retire at that age) highlighting evidence of strong cultural norms associated with reaching SPA and the notion of retirement in the UK (Kohli, 1991). Consistent with the findings of Bernheim (1988), Disney and Tanner (1999) also find that individuals report their most likely retirement age as opposed to their mean (expected) retirement age. 2 This implies that the notion of retirement in the UK is associated with a reduction in hours worked or complete exit from the labour market. Whilst the default retirement age was scrapped in the UK in 2012; there remains a strong association between reporting being retired and number of hours worked or exit from the labour market (Banks and Smith, 2006). 3

4 Building on the work of Bernheim (1989), Benitez-Silver and Dwyer (2004) use the US Health and Retirement Study (HRS) to investigate the role of information in determining retirement expectations and how expectations evolve over time. 3 4 Their findings show that respondents updated their expectations according to the newly received information, consistent with the rational expectations hyptothesis. The authors highlight, however, that not all changes are anticipated: in particular health and employment shocks lead to deviations from expected retirement age consistent with the findings of Disney and Tanner (1999). Chan and Stevens (2004) also using the HRS study the effect of pension incentives on retirement, instead of focusing on age of actual labour market exit the authors use ERA. Their findings highlight that ignoring individual s preferences for retirement leads to overstating the effect of pension incentives on retirement decisions. The authors also find a strong correlation between the retirement decision and individual characteristics such as pension wealth, earnings, assets and work expectations; and that expected retirement age is a strong predictor of actual retirement age. Recent studies have emphasized the prevalence of co-ordination in the timing of retirement and the role of spousal effects, often referred to as joint complementarities in leisure (Lundberg 1999; Coile, 2004; Michaud and Vermeulen, 2011). Ho & Raymo (2009), using the Health and Retirement Study (HRS), analyse joint retirement expectations and realisations among dual earner couples. Their findings show that couples who expect/plan to retire jointly are significantly more likely to actually do so than those who do not; moreover age, health and spouse s relative earnings in addition to retirement expectations play an important role in predicting observed behaviour. Relatively few studies have analysed how changes in SPA directly affect retirement expectations. Notable exceptions include Botazzi, Jeppelli and Pudula (2006), De Grip, Fouarge and Montizaan (2013) and Copolla and Wilke (2014). Botazzi, Jeppelli and Pudula (2006) investigate what effect a series of reforms introduced in Italy during the 1990 s to raise the SPA and reduce the replacement rate had on self-reported expected retirement ages and pension wealth. 5 Their findings suggest individuals adjusted in the expected way: males increased their ERA by two years and females by three years. 6 However, the authors also find that there is higher offset between private wealth and perceived pension wealth among individuals who are better informed about their pension wealth; underlining the importance of 3 For early work analysing the role of retirement expectations using HRS, specifically for the case of females see Honig (1988) who finds it is women s own characteristics and that of their spouse s which is important in addition to a range of sociodemographic and economic characteristics which determine whether she will work past In a related paper Cobb-Clark and Stillman (2006) using the Austrailian HILDA study analyse changes in ERA over time and consistency of answers and non-random non-response. Their findings highlight the importance of defined pension benefits and standard retirement ages (within a job), and the importance of labour market position in order to minimize group heterogeneity. The authors advocate the use of using expected retirement ages as opposed to subjective probabilities based on the reasoning that (i) any non-responding individuals may in fact be facing greater uncertainty about their retirement options and (ii) asking only about the likelihood of working at various ages suggests workers see retirement as something distinct to the complete cessation of employment. 5 The policy also reduced the replacement rate of younger workers compared to older workers. The actual increase in SPA depended on whether an individual is working in private or public sector or alternatively selfemployed. For more information see Table 1 in Botazzi et al. (2006). 6 See also Mastrogiacomo (2004) who the first of these policy reforms and whose findings are consistent with Japelli et al. (2006) and similar to Disney and Tanner (1999) emphasizes the importance of accounting for sample attrition and don t know responses. 4

5 individual s understanding and knowledge of the social security system (Botazzi, Jeppelli and Pudula, 2006). De Grip, Fouarge and Montizaan (2013) using matched survey-administrative data carry out a similar exercise in the case of a reform introduced in the Netherlands, which increased the SPA from 65 to 66/67 (66 from 2020 and 67 from 2025). The authors found individuals affected by the first reform (born ) only partially adjusted (increasing their ERA by 3.6 months) and found stronger effects (an increase of 10.8 months) for those born from 1960 who saw their SPA increase from 65 to 67. The study finds that the changes in ERA were driven by highly educated females who also have higher pension wealth; suggesting policymakers should investigate why similarly educated men and less educated individuals were not adjusting their expectations in the manner expected. De Grip et al. (2013) also found that individual s in manual occupations did not revise their expectations to the same extent as individuals in professional occupations, possibly due to health reasons. The authors suggest the reform had an income effect (due to a change in lifetime income) and noted that individuals who had made more years of contributions to their occupational pension were less sensitive to the announced policy changes made to the state pension. In closely related work, Coppola and Wilke (2014) analyse the 2007 German Pension Reform which raised the SPA from 65 to 67, like the Dutch reform the policy affected younger cohorts (born after 1967). Consistent with De Grip et al. (2013) the authors find that lower educated individuals failed to revise their expectations in the manner expected, however in the German case there was no differential response by gender. Whilst there was some evidence of individuals adjusting their ERA, after controlling for within group cross sectional correlations, the authors found there was lack of adjustment in ERA. Consistent with other studies these findings highlighted the need for policymakers to promote the policy more effectively using high quality information campaigns and marketing. There is surprisingly little evidence for the UK on the role retirement expectations have in shaping the retirement decision, particularly given that population ageing is a key policy priority for government; underlined by the number and magnitude of policy reforms introduced in recent years. The one exception we are aware of is Disney and Tanner (1999) whose data cover to the late 1980 s/early 1990s, is prior to any kind of reform to SPA and is based on a selected sample. We attempt to fill this knowledge gap using a large scale nationally representative survey, the United Kingdom Household Panel Study (UKHLS) which contains a rich set of economic and sociodemographic factors including spousal characteristics; in addition to full DOB information and pension membership status. 7 We exploit the fact our sample period covers corresponding to a time when three different State Pension Acts were in operation allowing us to investigate the relationship between SPA and ERA. 3. Overview of Pension Policy in the UK The pension system in the UK is principally made up of three pillars. The first of being the (Basic) State Pension. The new State Pension in operation from April 2016 is a single tier 7 We use the secure version of the data which contains individual s full DOB. 5

6 flat rate basic pension based on a Pay as You Go funding structure. 8 Individuals in the UK are eligible to claim their State Pension when they reach the so-called eligibility age and not before. The level of State Pension benefit an individual is entitled to depends on the number of years of National Insurance Contributions made when in employment and an individual s DOB.. 9 In the 2018/19 tax year the maximum State Pension an individual can receive under the new single tier system is per week. 10 Expenditure on state pensions represent a significant proportion of welfare expenditures: in the tax year 2017/18 this was forecasted to be 93.8 billion or just under 5% of GDP in GB (OBR, 2018). In terms of its generosity the UK has the least generous public state pensions among OECD countries (OECD, 2017) underlining the importance of additional voluntary retirement saving through occupational and private pensions (the second and third pillars respectively). A recent report compiled for the UK Department for Work and Pensions (DWP) highlights that 1.1 million single pensioners rely on their State Pension alone as their sole source of income (Just, 2017). 3.1 State Pension and Pension Acts in the UK The basic structure and features of the UK State Pension system were first introduced in the Beveridge Report published in Between 1948 and 5 th April 2010 the SPA for women (men) remain fixed at 60 (65); however various other changes were enacted such as the abolishment of high marginal tax rates on employment earnings post SPA (see Bozio et al.,2010, for a detailed description of the history of UK State Pension until 2010 ). The first change to SPA was made in the 1995 Pension Act, which legislated to equate the female SPA to that of males, specifically female SPA would be increased by one month every month for those individuals born after 6 th April 1950, starting from April The full effect of the reform when originally legislated would not be complete until March The introduction of the 2007 Pension Act raised the SPA for both men and women to 66 between 2024 and 2026 (increasing by one month, every month), to 67 between 2034 and 2036 and to 68 between 2044 and 2046 (Bozio et al., 2010). The rational for the reform was to rising individual life expectancy observed since the 1950s and due to advances in healthcare (ONS, 2015), which had not been matched by an equivalent or proportional increase in SPA thus having significant fiscal implications for the exchequer and calling into 8 For individuals who reached SPA prior to this date under the `old system, the State Pension system has two tiers, a flat rate basic pension and an additional pension related to earnings. For a comprehensive description of system in place prior to April 2016 see Bozio et al. (2010). Our sample is composed of individuals for whom the rules of the `old system determine their level of state pension whilst for younger individuals the new system is the one that matters. 9 A recent OECD report showed that gaps in employment have limited impact on retirement income in terms of state pension accrual due to welfare policies in operation in the UK (OECD, 2015). 10 In the full basic state pension amounted to per week under the old system, however the amount calculated was a function of numerous factors such as the `class of the contribution, see Bozio et al. (2010) for more details. 6

7 question the fiscal sustainability of the UK state pension system going forward (Lassila and Valkonen, 2017). The introduction of the coalition government in 2010 saw another Pension Act being introduced in 2011, which sought to bring forward the rise in the female SPA. The equalisation of SPA would now be achieved by November Moreover, the increase in the SPA from 65 to 66 would be bought forward to take place between December 2018 and October 2020, from The magnitude and far reaching extent of the 2011 reforms were non-trivial: it has been estimated that 5 million individuals were affected (House of Commons library, 2017). It is important to note that the exact increase in SPA under the reform is not uniform between men and women or within gender and depends on an individual s date of birth. For example, the 2011 reforms partially affected cohorts born between April 1953 and October 1954, the magnitude was such that SPA was increased between 2-3 months (individuals born 6 th April th May 1953) and 16 months (individuals born 6 th November th December 1953). Individuals born between 6 th December 1953 and 5 th October 1954 saw their SPA increase by between months. Individuals born after 5 th October 1954 would reach SPA when they turned 66. It is therefore crucial to have information on individuals exact date of birth to accurately analyse the effect changes in SPA have on ERA. The final State Pension Act we consider was introduced in 2014 which bought forward the increase in the SPA to 66; prior to this all individuals born from 6 th April 1968 onwards had a SPA of 66 this would now hold for all individuals born after 6 th April For individuals born between 6 th April 1960 and 5 th March 1961, SPA is 66 plus 1 month every month (for example it is 66 years and 1 month for an individual born between 6 th April 1960 and 5 th May 1960). For persons born after 5 th March 1961 the SPA is 67. The 2014 Pension Act also noted that the SPA would be reviewed on a periodic basis with the first report due by 7 th May In March 2017 the Cridland Review was published. The review made a number of recommendations in relation to state pension in the UK. In terms of SPA it was recommended the increase in the SPA from 67 to 68 which had originally been set to take place between 2044 and 2046 should be bought forward by 7 years to ; this change affects individuals in their late 30s and early 40s (at the time it was published) and therefore does not affect our sample respondents who are at least 45 years old. Figure 1: UK State Pension Acts and Men s SPA 7

8 Pension Act Reforms & Men's SPA Pension Act 2007 Pension Act 2011 Pension Act 2014 Figure 2: UK State Pension Acts and Women s SPA Pension Act Reforms & Women's SPA Pension Act 2007 Pension Act 2011 Pension Act 2014 The changes to SPA legislated in the State Pension Acts are summarised visually in Figure 1 for men and Figure 2 for women, each figure shows the age at which an individual is eligible to claim their SP depending on the Pension Act in place at the time of their survey interview. Specific details of the changes made to SPA in each Act are detailed in Appendix A. The Figures show there that has been a significant increase in SPA for individuals in the affected cohorts over a relatively short period of time. The largest increase for men is 1 calendar year, whereas for women it is 2 years. The changes made to SPA for women has been criticized by the so-called Women Against State Pension Inequality (WASPI) group which argues that females in the most affected cohorts received insufficient advance notification of the changes they faced (House of Commons library, 2017). In terms of actual labour supply effects, Cribb et al. (2016) analyse the increase in female SPA from 60 to 62 and found the reform increased the employment rate of affected individuals by 6.3 percentage points. At the time of 8

9 writing the reforms to men s SPA had not yet taken effect. Indeed, the fact that many of the changes to SPA will not take place for some time and have only been implemented gradually underlines the importance of understanding whether individuals have made adjustments to their ERA. 3.2 Occupational and personal pensions in the UK (Second and Third Pillars) Occupational pensions. Occupational pensions which are a voluntary form of retirement saving between an employee and their employer and make up the second pillar of the UK pension system. These have traditionally been Direct Contribution (DC) or Direct Benefit (DB) type pension schemes; individuals are taxed on receipt of income from their occupational pension not at point of contribution. Occupational pensions (and personal pensions) play an important role in providing income in retirement, given that the level of state pension is relatively low in the UK (OECD, 2017); for example, average net replacement rates increase from 29% (state pension alone) to 62.2% after accounting for the presence of an occupational pension (OECD, 2017). From April 2002 individuals have been able to claim their state pension from the age of 55 (up from 50); highlighting the significant age gap between claiming occupational and state pension in the UK for both men and women. A recent DWP report highlights that only 62 % of pensioners expect to receive income from an occupational pension therefore the state pension is an important source of income for a non-trivial proportion of the retired population (DWP, 2015). In terms of affecting individual s retirement expectations, this may partly explain the high incidence of retirement transitions at or close to SPA and the relatively low incidence of actual or expected retirement at the age of 55. In an effort to increase occupational pension coverage and hence retirement saving (but not retirement age per se), the UK government in April 2012 introduced a policy whereby employees were automatically enrolled into a workplace pension (Auto-Enrollment, AE), evidence has shown that such nudge behaviour has been effective in terms of changing individuals behaviour in a particular way (in this case minimising opt out) given a particular policy objective (Cribb and Emmerson, 2016). 11 Early evidence from the introduction of AE suggests it has been largely successful in terms of coverage but issues such as increasing contribution rates conditional on participation still remain (Cribb et al. 2016; DWP 2017). Alongside AE the government made significant changes to the way in which retirement saving via a DC pension can be accessed and used. Until March 2015 individuals could draw down up to 25% (tax free) as a lump sum and annuitize the remainder; however, following the introduction of the Pensions Schemes Act 2015 in April 2015 individuals are free to use their retirement savings as desired. 11 Auto enrollment was gradually phased in starting with larger employers before moving on to smaller employers, note that it does not cover the self-employed and we do not include such individuals in our sample for analysis purposes due to the way the UK pension system has historically treated self-employed versus employees, in addition to issues such as selection and unobserved heterogeneity. 9

10 A priori it is not clear how this policy change may influence retirement expectations. One possibility is that given it came into effect soon after the 2014 state pension act which introduced further increases to SPA, the 2015 Pensions Schemes Act could have mitigated upward revisions to ERA; for example if individuals were likely to be liquidity constrained the policy reform meant this would no longer be the case and ceteris paribus induce individuals to retire earlier Personal pensions. Personal pensions are another type of voluntary retirement saving vehicle arranged between an individual and a provider (usually an insurance firm); and make up the third tier of the UK state pension system. Personal pensions are usually a DC type scheme and are treated in the same way as occupational pensions from a taxation perspective. 12 Individuals must be aged at least 55 before they are eligible to claim their personal pension; the Pension Freedom reforms introduced in 2015 affected personal pensions in the same way they did occupational pensions. The widespread availability and uptake of occupational pensions since the 1950s and the fact that the UK labour market has historically been made up of employees means the proportion of individuals who have a personal pensions is higher among the self-employed, a group of individuals who we do not consider in this study. 13 This also implies that when analysing retirement age expectations among employees it is relatively more important to control for the presence of an occupational pension rather than a personal pension Abolishment of the mandatory retirement age Until October 2011 existing legislation meant that employees had relatively few employment rights after the age of 65; for example, employees were required to request permission to work past this age and therefore 65 became the default retirement age (Lain et al., 2017). This changed with the introduction of the 2011 Employment Equality Regulations (EER) which scrapped the so default retirement age and allowed employers to retire employees only if there as a justifiable legal basis for doing so. The abolishment of mandatory retirement did not correspond with a sudden increase in employment rates among older workers, however there has been a general increasing trend in participation rates among individuals aged 65 and over since the early 2000 s (ONS, 2018) An individual can have an occupational and personal pension although certain limits exist in terms of total contributions made within a tax year. a tax year runs from 6 th April each year to 5 th April the following year. The annual pension contribution allowance on a personal pension is set by government; in the tax year it is 40, Self-employment rates have increased in the period after the Financial Crisis; nonetheless the self-employed still only make up 15% of the labour force in the UK (ONS, 2018) ries/lfk6/lms 10

11 On balance we do not expect the 2011 EER to affect the impact SPA has on ERA in a systematic way, due to the longstanding relationship between reaching SPA and reducing work hours or exiting the labour market in the UK which had been linked to strong cultural norms and social security incentives (Kohli, 1991; Blundell, Bozio and Laroque, 2013). This contrasts with evidence from the US which found positive effects in terms of participation rates following the introduction of a similar policy (von Wachter, 2009). 4. Data We use data from waves 2-6 ( ) of the United Kingdom Household Panel Study (Understanding Society) the largest annual longitudinal household survey in the UK, providing information on around 40,000 households (Buck and McFall, 2011; Knies, 2017). Understanding Society contains detailed information relating to a range of individual and household economic and sociodemographic characteristics. The secure version of the study contains individual s day, month and year of birth which is needed to construct the exact age at which an individual is eligible to claim their State Pension. This is important as previous research has shown that reaching SPA in the UK is associated with an exit or significant reduction in labour supply (Banks and Smith, 2006; Blundell, Bozio and Laroque, 2013). We follow previous research and assume that individuals refer to retirement as a reduction in labour supply; this could be a reduction at the intensive margin (hours worked) or at the extensive margin (cessation of employment). We make use of the retirement planning module available at every wave of the survey which contains a range of questions relating to individual s subjective expectations in relation to events associated with retirement. The specific question of interest in relation to expected retirement age is the following: There is a lot of policy interest in how people are planning for their long term future and retirement. At what age do you expect you will retire or will consider yourself to be retired? Respondents provide an integer value or don t know. The retirement planning module is age-triggered only individuals aged 45, 50, 55, 60 and 65 who consider themselves not retired are asked the battery of questions in the module. We exclude 65-year-old men and 60-yearold women from our sample due to selection; it is likely that these individuals are systematically different (in an unobservable way) from the rest of our sample as they are working past SPA at the time the survey was administered. Appendix B describes the key economic and sociodemographic characteristics used in the analysis. 15 Appendix C provides summary statistics on these same characteristics by gender. 15 The descriptive statistics are based on the sample used for the full model which includes controlling for whether an individual has an occupational pension. 11

12 When interpreting the results, it is important to note that for analysis purposes we drop individuals whose response is don t know. Evidence has shown that individuals who have difficulty in answering retirement planning questions are a non-random sample; these individuals face more uncertainty with respect to the retirement decision, retirement income, are lower educated and less financially literate (Disney and Tanner, 1999; Lusardi, 2007; Mastrogiacomo, 2004; van Santen, Alessie and Kalwij, 2012; and Copolla and Wilke, 2014). We do not attempt to correct for this selection bias due to lack of available instruments, however do compare whether such individuals differ in terms of observable characteristics from those who gave a valid response. Descriptive statistics (not reported) based on observable characteristics by whether an individual gave a positive response versus answering don t know indicated that the latter group contained a higher proportion of individuals with lower levels educational attainment and public-sector workers and a higher proportion of single individuals. In addition, this group also reported on average a lower level of individual and household income. We restrict our sample to all individuals at each wave of the survey who are eligible to answer the survey. We then combine wave specific observations such that we have one observation per individual. 16 Our sample consists of 2,708 males and 2,405 females for which we have complete information. We focus only on employees given the historic differences in terms of pension provision and retirement savings relative to the self-employed; in addition to issues such as selection and unobserved heterogeneity. For similar reasons including the fact the SPA is different between men and women over our sample period we analyse results separately by gender. Before turning to the methodology section we first document the average changes in expected retirement age over time and by gender. We split our sample period into three separate subperiods defined by the State Pension Act in operation at the time of the individual s survey interview. For each gender we split our sample into (roughly) single year age bands based on SPA and compute the average gap between ERA and SPA for each SPA (band)- period combination. 16 In a small number cases there was more than one observation per individuals, whilst Understanding Society is an annual survey, in some cases the survey took place just after an individual s, say, 45 th birthday and just before their 46 th birthday. We only keep one observation (the first) for each individual. 12

13 Table 1: Mean difference in expected retirement age and state pension age (ERA-SPA) by period for male respondents. Period Table 2: Mean difference in expected retirement age and state pension age (ERA-SPA) by period for female respondents. cells: mean ERA (median) Jan Nov May13 May SPA (banded) Nov Dec All [65,65.5) [65.5,66.5) [66.5, 67) No observations All cells: mean ERA (median) Period SPA (banded) Jan Nov Nov May May Dec All [60, 60.5) 3.8 No observations No observations 3.8 [60.5,61.5) No observations 2.6 [61.5,62.5) No observations 1.9 [62.5, 63.5) No observations 0.9 No observations 0.9 [63.5,64.5) [64.5, 65.5) [65.5, 66.5) [66.5, 67) No observations All Tables 1 and 2 highlight that one common finding across genders is that younger cohorts report (on average) a lower ERA and hence the difference between ERA and SPA is greater for individuals with higher SPA. Turning to gender specific findings, for males we find that within the same SPA band individuals who are surveyed later in the sample period report on average lower ERA. For women, we find that older individuals who have a lower SPA band (between 60 and 62.5) and are surveyed earlier in the sample period (between January 2010 and May 2014) report (on average) higher ERA and hence the gap between ERA and SPA is positive. There is also some descriptive evidence of the ERA-SPA gap narrowing (ERA increasing) between the first and second period for women. 13

14 5. Methodology Figures 1 and 2 shows that the change to SPA is gradual and relatively small for some cohorts; whilst for others it is much larger (up to two years), in addition the magnitude of the changes differ by DOB in different calendar years. This complex relationship between SPA on the one hand and calendar time and cohort on the other, makes it difficult to use standard econometric frameworks of regression discontinuity designs (RDD) or regression kink designs (RKD) for analysing the relationship between the SPA and the ERA (Card et al., 2015; Card et al. 2017). Nevertheless, these studies show that if one controls for flexible (smooth) functions of the assignment variable, then one can nonparametrically identify the effect of SPA on ERA by exploiting the discontinuities as shown in Figures 1 and 2 =. One possibility is to analyse the SPA-ERA relationship around the kink and jump points separately, however the low number of observations around these points makes this impossible. We, therefore, take a parametric approach in which we control for polynomials in the assignment variables, which in our case are date of birth and calendar (interview) time. SPA varies over time and across cohorts. Our dependent variable is the difference (or gap) between ERA and SPA (i.e., GAP=ERA-SPA). The assignment variables (determining SPA) are date of birth and time of the interview and are denoted by, respectively, d and t. We estimate the following equation by Least Squares for each gender: (1) GAP i = ω 0 + ω 1 SPA i + ω 2 d i + ω 3 d i2 + ω 2 t i + ω 3 t i 2 + µx i + δh i + ƺ i The individual is indexed by i, d i is birth date in deviation of the sample minimum and t i is calendar time in deviation of the sample minimum. X i refers to individual characteristics which are time invariant and finally H i refers to household characteristics (including that of the spouse). ƺ i is an error term. In addition to individual and household characteristics the i key coefficient of interest is that associated with SPA i. If ω 1 is equal to zero, ERA increases with the same number of months as the increase in SPA, leaving the gap between ERA and SPA unchanged. 6.Estimation results We present results in two stages. The first stage is based on two specifications (i) without controlling for whether an individual has an occupational state pension and (ii) controlling for occupational state pension. We focus on (i) and discuss whether controlling for the presence of an occupational pension changes the qualitative nature of the results. The second stage uses the estimated coefficients to make gap predictions, or the difference between individuals ERA and SPA at particular covariate values to highlight the main empirical findings. 14

15 6.1Men Table 3 indicates that an increase of one year in SPA reduces on average the gap between ERA and SPA by 0.97 years, put another way, conditional on other factors controlled for this implies that males do not adjust their expectations to an increase in SPA but actually reduce their ERA. This latter reduction is not statistically significant. In terms of cohort effects younger males report a lower ERA (or the effect of being younger makes the gap negative) holding all else constant and that this effect increases at an increasing rate. There evidence of strong positive time effects which increase at a declining rate. Turning to economic and sociodemographic characteristics we first discuss individual and then turn to household level factors. Individuals in poor health report on average a higher ERA, or put another way this has a positive effect (of 1.2 years) on the gap between ERA and SPA relative to individuals in excellent health. We find evidence education effects; relative to individuals who hold a degree having a lower level of educational attainment has a negative impact on the gap between ERA and SPA. The size and significance of this effect varies, having an A-level compared to a degree lowers the gap by around 0.5 years and is even larger for individuals with other types of qualification. 17 <Table 3 here: currently awaiting release from UKDA> There is no significant difference between ethnic minority groups relative to the white majority except for Black African and mixed white/black African; being a member of these group lowers the ERA-SPA gap by approximately 2 years. However, we interpret this result with caution: this ethnic group represent a relatively small proportion (2%) of our sample. Job characteristics, specifically occupation has a significant effect on the ERA-SPA gap. Compared to individuals whose occupation is considered professional ; belonging to any other occupational groups has a negative impact on the ERA-SPA gap, the magnitude of the belonging to a group varies however in most cases it is around one year or more. In addition to occupation, sector of work was also found to be important. Working in the public sector lowered the ERA-SPA gap by 1.4 years We also control for individual and household income. In both cases the results suggest that higher levels of household and individual income lower the ERA-SPA gap, put another way, higher individual and household income leads individuals to report on average earlier retirement ages. Turning to other household characteristics we find that the presence and number of dependent children in the household raises the ERA-SPA gap, after controlling for cohort and time effects. Housing tenure was also found to be important, relative to individuals who owned their home outright; owning with a mortgage or renting had a positive impact on the ERA- SPA gap. It is likely both dependent children and types of housing tenure have financial 17 Other groups refers to relatively basic skills. 15

16 implications in terms of staying in employment which leads to individuals reporting on average a later retirement date. Whilst we control for a range of spousal level characteristics (not reported) we do not find any factors which are statistically significant. Finally, we also control for geographical region and find that living in the West Midlands, London, South East and Scotland increases on average the ERA-SPA gap by over one year. When we also control for whether an individual report s being a member of an occupational pension scheme, conditional on all other factors this has a positive effect on the ERA-SPA gap increasing it on average by 0.79 years. Qualitatively all other findings remain unchanged. 6.2 Women Table 4 shows that females make a partial adjustment with respect to the age which they expect to retire in the face of a change in the SPA. Specifically, the ERA-SPA gap reduces by 0.67 years for a one-year increase in SPA. Thus, consistent with the behaviour of men, women also report lower ERAs. In terms of cohort effects, we see a similar pattern to that found for men, namely that younger cohorts of females are reporting lower ERAs (at an increasing rate). We also find strong time effects (at a decreasing rate) noting that women have seen their SPA rise more rapidly over the sample period relative to men. We do not find evidence of education effects except for individuals who had an A-level (equivalent to 13 years of full time education), which reduced the ERA-SPA gap by around 0.6 years compared to those with a degree. We find some evidence of differences in ERA of women from ethnic minority groups; relative to the white majority Indian or Black African females and mixed white and Black African females reduced the ERA-SPA gap by on average almost and 2 and 3 years respectively. Again, we interpret this result with caution due to relatively small sample sizes. <Table 4 here: currently awaiting release from UKDA> Housing tenure is also important in determining the ERA-SPA gap, like men, females who report owning their home with a mortgage or renting report on average a higher ERA-SPA gap relative to those who own their home outright. Table 4 shows that females living in higher income households reported on average a lower ERA-SPA ceteris paribus. We find some evidence that spousal characteristics such as health and job occupation are important. Having a spouse who reported having fair health compared to excellent raised the ERA-SPA gap by over one year. Whereas having a partner who had a skilled non-manual occupation decreased the ERA-SPA gap by just under on year. The presence of an occupational pension was found to be important. Individuals who reported being a member of their employer s scheme reported an ERA-SPA spa which was on average 0.69 years higher than those without, holding all else constant. However, unlike in the case of men the sector of employment was not found to significant (not reported); in all other respects the results did not change qualitatively in the extended specification. 16

17 6.3 Gap predictions Using the estimates based on our preferred set of models it is possible to generate predictions of the average gap between ERA and SPA at covariate values (of statistically significant factors) for men and women to identify possible vulnerable groups in the sense of having a large (negative) gap. 18 We first consider the effect of SPA on the ERA-SPA gap. For men we consider the effect of SPA being set at 65,66 and 67. Irrespective of whether we control for occupational pension the trend remains the same: higher SPA is associated with a bigger difference between ERA and SPA. This is consistent with Tables 1 and 2 which shows that the youngest male cohorts in our sample (who have the highest SPA) report on average lower ERA. In the case of women, we predict the gap based on SPA varying between 61 and 67; in this case the pattern is even more pronounced: the gap is initially positive for SPA=61 and SPA=62 (0.77 and 0.10 years respectively) at successively higher ages after this point the gap increases, indeed for SPA=67 the ERA-SPA gap is years. Therefore, consistent with their male counterparts and Table 2 women from younger cohorts report lower ERAs. Table 5 shows that for men higher levels of individual income tends to have a negative impact on the ERA-SPA gap, for example focusing on the specification which does not control for occupational pension moving from the 50 th to 75 th percentile of the individual income distribution increased the difference between ERA and SPA from to years. The magnitude of this effect hardly changes even after we control for whether an individual is a member of their employer s occupational pension scheme. We find education effects only for men. Relative to degree holders those with lower levels of educational attainment report a larger gap between ERA and SPA; this finding also holds after we control for pension membership. Similarly, we find that job occupation (of current job) is only important for men. Relative to an individual s whose job is considered professional (as defined by the Registrar General s Social Class index) the effect of having a job in a lower social class has a negative effect on the ERA-SPA gap, for example being in an unskilled occupation lowers the ERA-SPA gap by 1.63 years, again the qualitative effects do not change after controlling for pension membership. In addition to job occupation we also find sector of work is important in determining the ERA-SPA gap; the results indicate that working in the public sector has a relatively large negative impact on the ERA-SPA (more than two and a half years). Table 5: Effects of economic and sociodemographic characteristics on the ERA-SPA gap. Covariate Specification not controlling for occupational pension Estimated gap (ageret-spa) measured in years Specification controlling for occupational pension Estimated gap (ageret-spa) measured in years 18 We focus only on factors which are statistically significant at conventional levels for either one or both genders. 17

18 Males Females Males Females SPA SPA=61 N/A 0.77 N/A 0.72 SPA=62 N/A 0.10 N/A 0.02 SPA=63 N/A n/a SPA=64 N/A n/a SPA= N/A N/A SPA= SPA= Income 25 th percentile 50 th percentile 75 th percentile Education Degree ( min. 16 years f/t education) A-level (13 f/t education) No qualifications Base group Base group Base group Base group N/S N/S N/S N/S Social class of current occupation Professional Base group Base group Base group Base group Managerial N/S N/S and technical Skilled nonmanual N/S N/S Skilled N/S N/S manual Partly skilled N/S N/S Unskilled occupation N/S N/S Sector of work Private Base group Base group Base group Base group Public N/S