Centre for Market and Public Organisation What do DC pensions mean for retirement? Evidence from the UK Sarah Smith CMPO, University of Bristol and IFS Background • UK experiencing shift from DB to DC schemes – – • • • • Half of all men contracted out of state secondary scheme have a DC scheme More than half of those currently approaching retirement (50-64) What does this shift in pension provision mean for the timing of retirement? Numerous studies have established the importance of pension incentives for retirement – Pension wealth – positive effect – Accrual – negative effect Stock and Wise, 1990; Gruber and Wise, 2004, for international evidence UK – Blundell, Meghir and Smith, 2002, show that growth of DB occupational schemes can explain part of shift to early retirement in 1980s and 1990s This paper • How do retirement incentives evolve in DC schemes (compared to DB schemes) and what effect might this have on retirement? – • Absence of strong age-related incentives implies profile of retirements likely to be far smoother How do individuals respond in practice to incentives in DB and DC schemes? – – – – Model DB and DC pension wealth and accrual for sample of older workers in English Longitudinal Study of Ageing (ELSA) and estimate effect on probability of retirement DB wealth has positive and significant effect, but DC pension wealth does not DB and DC accruals have negative effect, sig only for DC schemes Results imply that DC schemes result in later retirements Previous papers • • Friedberg and Webb (2005) Estimate separate wealth effects for SS, DB and DC – – • • Estimate accrual effects for SS and DB, but not DC schemes Simulations show that shift to DC schemes results in later retirements – – • • SS and DB have positive and significant effects DC wealth has positive effect, but insignificant DB wealth, but zero accruals DC pension wealth Coile and Gruber (2006) Estimate separate wealth and accrual effects for SS, DB and DC – – – SS and DB wealth have positive and significant effects DC wealth has positive effect, but insignificant DC accrual has negative effect, significant at 10% level The UK • DC pensions took off in 1988 when the government allowed people to contract out of state secondary pension scheme into DC schemes (previously only DB schemes) • For people who have contracted out, the low value of the basic state pension means that their private scheme is likely to be the main driver of retirement • Unique institutional features – compulsory annuitization by age 75 Modelling retirement 60 early retirement age, 60 40 20 30 normal retirement age, 65 20 -20 -10 0 accrual - £000s 0 10 Single period accruals -20 • Earliest reduced form models estimated probability of retirement as a function of wealth and single period accruals (or accruals over a discrete period). Failed to take account of non-linearities caused by early retirement windows in DB schemes Option Value model – modelled individual retirement decision as a comparison of PV of retiring now with PV of retiring at all possible future retirement ages accrual - £000s • 50515253545556575859606162636465666768697071727374 50515253545556575859606162636465666768697071727374 Option value model • Value of retirement in period r: r 1 S s t s r Vt (r ) s tU w Ys s tU r Bs r U w Ys Ys s • U r Bs kBs s Value of postponing retirement until r r 1 S s t s r S r 1 Gt (r ) s t Et Ys s t Et kBs r s t Et kBs t Et s t s s • s t s t Assuming values for , k, and that expected random components are equal to zero, the probability of retirement as a function of G(r*) can be estimated using a probit model Peak value model • OV dominated by future earnings – – • • These are uncertain They may also capture individual heterogeneity Coile and Gruber, 2002, proposed simpler Peak Value Assuming = k = 1, OV can be re-written as a pure revenue function Gt (r * ) s t Et Ys s t Et Bs r * s t Et Bs t r 1 S S s t s r s t Peak value • • OV should better reflect underlying incentives, but PV may give better approximation PV better suited to identifying effects of separate elements of pension system Accrual – DB and DC schemes compared Delaying retirement in DB scheme Delaying retirement in DC scheme • Increase value of lump sum and pension by increasing final salary and years’ service (up to max) • • Entitlement to early retirement Increase value of pension fund by receiving additional contributions from the state/ employer and getting another year’s return on accumulated fund • Loss of pension income (after normal/ early retirement age) • Loss of one year’s pension (after age 50); change in the annuity rate with age • Delay in receiving pension • Delay in receiving pension • Probability of dying without ever drawing a pension • Probability of dying without ever drawing a pension early retirement age, 60 20 60 Defined contribution 10 0 accrual - £000s 20 -20 -10 -10 0 accrual - £000s 0 10 40 20 30 normal retirement age, 65 -20 accrual - £000s Accrual – DB and DC schemes compared 50515253545556575859606162636465666768697071727374 50515253545556575859606162636465666768697071727374 DB accrual rate = 1/60th Age-earnings profile estimated using FES data 1968-2002 Nominal discount rate = 5%; inflation rate = 2.5% Survival rate = CMIB, PNML02 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 DC contribution rate set at 13.5% to yield same pension wealth at age 65; nominal investment returns = 6% Predicted retirements – for illustration .3 .2 .1 0 Probability of retirement .4 Take simulated wealth and peak values, apply previous estimates of pension incentive effects (Blundell, Meghir and Smith, 2002) and predict retirements under two types of pension 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 DB early retirement DC What happens in practice? • • • • English Longitudinal Study on Ageing (ELSA) Panel study of 11,400 individuals aged 50+ Interviewed every two years from 2002; waves 1 and 2 currently available Detailed information on: – Pensions – current two and past three schemes – Other wealth (net financial wealth, owner occupied housing, other physical assets) – Income, demographics, economic activity, health • Here, focus on sub-sample of 1,478 men who are – Aged 50 – 64 and employed in wave 1 – Survive to wave 2 – Exclude high pension/ non-pension wealth and those who don’t know pension type • Model retirement between W1 and W2 as a function of (modelled) DB, DC, state pension wealth and accrual Ignore possible selection effects into pension type • Modelling pension wealth • • • • • • • • • DB pension Wealth based on self-reported years in the scheme (or, if missing, years with employer) and self-reported scheme accrual rate. Assume uprating in line with 2.5% inflation and 50% survivor benefits Future wealth based on accrual rate (up to max years) and constant, nominal earnings. Early retirement with 4% reduction in pension value for each year DC pension Self-reported fund value at wave 1, converted into wealth using the second best available age-specific annuity rate (FSA). Future wealth assumes contributions remain at their current rate and the fund attracts a nominal 5% annual return. Both use age-specific life expectancies for the cohort from the Government Actuary’s Department and a 5% nominal discount rate Modelling pension wealth • Two features of UK pension system • It’s complicated • Current DB 26.9% Current DC, past DC 3.4% Current DB, past DB 4.7% Current DB and DC 3.0% Current DB, past DC 0.8% Past DB 6.3% Current DC 34.5% Past DC 1.9% Current DC, past DB 9.1% No private pension 9.4% Lots of people have DC schemes with very small amounts in them (£’00,000) Value of DB pension Value of DC pension Mean 25% 50% 75% N 2.19 0.74 1.88 3.06 670 0.49 0.07 0.18 0.46 746 Focus on main pension = private pension wealth > state pension wealth Distribution of pension wealth, by earnings quintile Mean value (£’00,000) Proportion with pension main main any 1 2 3 mean of dbpen 4 5 4 1 mean of dcpen mean of nopen 2 3 mean of dbmain mean of spmain DB DC No private pension 4 5 mean of dcmain 1 0 0 0 0 .1 .2 1 .2 .4 2 2 .3 .6 3 3 .4 .8 .5 4 any 1 2 3 mean of spval mean of dcval0 4 5 1 mean of dbval0 2 3 mean of spval mean of dcval1 State pension DB DC 4 5 mean of dbval1 Sample characteristics DB DC - employer DC - individual State pension 33.8% 3.6% 7.4% 55.3% Total pension wealth 3.26 2.05 2.22 1.38 State pension wealth (£’00,000) 0.47 0.59 0.45 0.63 DB pension wealth (£’00,000) 2.59 0.07 0.54 0.03 DC pension wealth (£’00,000) 0.03 1.29 1.16 0.10 Financial wealth (£’00,000) 0.45 0.61 0.65 0.39 Total non-pension wealth (£’00,000) Age 1.46 3.18 3.11 1.73 55.0 55.2 55.1 56.6 £27,965 £34,357 £41,006 £17,113 Job tenure 16.0 19.6 16.1 10.9 % doing manual work 0.25 0.26 0.35 0.46 % self-employed 0.04 0.02 0.56 0.31 % who retire 0.17 0.09 0.09 0.15 % sample Gross annual earnings Regression analysis • Estimate the following model using a probit regression: Ri 0 1SPWi 2 DBWi 3 DCWi 4 SPAi 5 DBAi 6 DCAi 7 NRAi 8 ERAi 9 AGEi 10 ln Yi 11 ln Y X i ' ui 2 i R = binary variable if individual has left work by wave 2 SPW = state pension wealth at W1 for everyone DBW, DCW = DB, DC wealth at W1 for main DB, DC pension SPA, DBA, DCA = accrual (of total pension wealth) for those with main state, DB, DC pension NRA, ERA = binary variable if individual reaches normal, early retirement age by W2 Age = linear age term Y = earnings X includes log of non-pension wealth, whether the individual is working full-time at W1, whether the individual is self-employed at W1, whether they are in manual work at W1, their job tenure at W1 (and its square), whether their spouse is in work at W1, whether they have college education and indicators for a range of health problems at W1 Plus, dummies for main pension type • Estimate pooled model, plus separate regressions for each of three main pension types Results Any wealth Main wealth Main wealth Main wealth Sep accrual .0362** (.0095) -.01773 (.0356) .0531 (.0469) -.1011 (.1271) -.2351 (.2071) .0503 (.1035) .0374** (.0095) -.0213 (.0467) .0516 (.0467) -.1042 (.1269) -.2344 (.2071) .0544 (.1036) .0361** (.0094) -.0216 (.0484) .0463 (.0476) .0366** (.0094) -.0224 (.0497) .0572 (.0478) .0380** (.0103) -.0363 (.0440) .0511 (.0478) -.0018 (.0023) -.0148** (.0070) .0003 (.0016) -.0087* (.0051) -.2330 (.1611) -.0020 (.0040) .0091 (.0055) .2251 (.1611) .0020 (.0047) 1395 -469.68 DB wealth (£’00,000) DC wealth (£’00,000) SP wealth (£’00,000) DB peak (£’000) DC peak (£’000) SP peak (£’000) DB acc2005 (£’000) DC acc2005 (£’000) SP acc2005 (£’000) DB acc2004 (£’000) DC acc2004 (£’000) SP acc2004 (£’000) N 1395 Log likelihood -472.38 1395 1395 -.0075 (.0050) -.0227* (.0127) .0008 (.0040) .0071 (.0052) .0116 (.0135) -.0006 (.0041) 1395 -471.92 -469.98 -467.72 Results • • • • • Wealth DB pension wealth enters positively and significantly in all specifications State pension wealth enters positively and significantly in single regression DC pension wealth enters negatively and insignificantly in all specifications Including DB and DC wealth only for main pension results in slightly better fit than including for all, but little effect on the results • • Accrual DB pension accrual has expected (negative) sign, but is insignificant in most specifications. Not capturing full range of early retirement incentives? State pension accrual is insignificant, but little variation across individuals DC pension accrual has expected (negative) sign and is significant in most specifications. Variation across individuals reflects contribution rates. Endogenous? Single period accruals result in slightly better fit than peak value • • • Why no DC wealth effect? • Researchers are better at modelling (formulaic) DB pension wealth? – But DC pension wealth comes from individuals’ reported values • Individuals have better understanding of what DB pension wealth means for retirement income? • DC pension wealth has very different risk properties to DB pension wealth? – • DC pension wealth may be endogenous for some? – • Jan 2003 onwards, period of rising stock market, but after-effects of earlier crash But bias should go the other way DC pensions are more flexible and imply a less close relationship between retirement and pensions? – – Part of accrual in DC scheme is independent of working Individual can draw a DC pension and work Retirement and pension receipt employer pension 0 .2 Density .4 .6 personal pension -5 0 5 -5 0 Difference between age started to draw a pension and age stopped work Graphs by pen Source: British Household Panel Survey 5 0 .1 .2 .3 .4 .5 .6 What does this mean for retirement? Predicted retirement hazards 50 51 52 53 54 55 56 DB predictions 57 58 59 60 61 62 DC predictions DC pensions associated with later retirement…. 63 64 0 0 .1 .1 .2 .2 .3 .3 .4 .4 .5 .5 .6 .6 What does this mean for retirement? Predicted retirement hazards 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 DB predictions DC wealth, DB paras DC predictions 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 DB predictions DB wealth, DC paras DC predictions … this has little to do with lower levels of wealth, but largely driven by differential response to pension incentives Conclusions • Shift from DB to DC schemes has implications for the timing of retirement • • DB schemes: Provided a tool for employers to manage labour market exits (through early retirement windows). Strong age-related incentives and clustering of retirements • • • • DC schemes: Age profile of retirements is (likely to be) smoother. Estimates suggest weaker link between wealth and retirement, consistent with greater flexibility. But accruals (contributions?) still appear to be important. Results suggest that DC schemes result in later retirements. • Important unresolved issues include – – Selection into pension type Endogeneity of pension wealth and contributions