scieee AI-readable full text Open interactive document viewer

Early Retirement in West Germany: A Sequential Model of Discrete Choice

Siddiqui, Sikandar

Abstract

EconStor is a publication server for scholarly economic literature, provided as a non-commercial public service by the ZBW.

Full text

Siddiqui, Sikandar Article Early Retirement in West Germany: A Sequential Model of Discrete Choice Zeitschrift für Wirtschafts- und Sozialwissenschaften (ZWS) - Vierteljahresschrift der Gesellschaft für Wirtschafts- und Sozialwissenschaften, Verein für Socialpolitik Provided in Cooperation with: Duncker & Humblot, Berlin Suggested Citation: Siddiqui, Sikandar (1997) : Early Retirement in West Germany: A Sequential Model of Discrete Choice, Zeitschrift für Wirtschafts- und Sozialwissenschaften (ZWS) - Vierteljahresschrift der Gesellschaft für Wirtschafts- und Sozialwissenschaften, Verein für Socialpolitik, ISSN 0342-1783, Duncker & Humblot, Berlin, Vol. 117, Iss. 3, pp. 391-415, https://doi.org/10.3790/schm.117.3.391 This Version is available at: https://hdl.handle.net/10419/291891 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Zeitschrift für Wirtschafts- u. Sozialwissenschaften (ZWS) 117 (1997), S. 391 - 415 Duncker & Humblot, Berlin Early Retirement in West Germany: A Sequential Model of Discrete Choice* By Sikandar Siddiqui1 1. Introduction Most of the industrialized countries can today be expected to undergo a dramatic process of population ageing during the next three to four decades. This tendency, which is due to both the decline in birth rates in the 1970s and the rising average life expectancy of individuals, is particularly pronounced in Germany: According to a forecast by the German Federal Bureau of Statistics („Statistisches Bundesamt"; see Sommer, 1992), 34.9 % of the German population will be older than 60 years by the year 2030 (compared only to 20.4 % in 1989). Unless appropriate policy reforms are undertaken, these developments are bound to put the financial basis of Germany's unfunded public pension system under considerable strain. As raising the average retirement age is often considered an appropriate means to cope with this problem, the expected demographic changes have spurred off an increasing interest in the factors determining the labour supply behaviour of the elderly. This becomes particularly obvious when considering the rules being laid down in the 1992 Pension Reform Act, which was enacted by the West German government in 1989: The new German pension law substantially reduces the financial incentives for early retirement which had been typical of its predecessor. Moreover, it prescribes that the minimum retirement age will be raised in quarterly steps from the year 2000 onwards until it reaches 65 by the time of the year 2012. In this paper, I present an empirical analysis of retirement behaviour among male West German employees, using data from the Socio-Economic Panel (SOEP). Panel datasets of this kind provide a particularly useful basis for empirical studies of retirement behaviour because of two main reasons: * Verantwortlicher Herausgeber / editor in charge: J. W. i Roland Berger und Partner GmbH, Bockenheimer Landstr. 42, D-60323 Frankfurt/Main. I would like to thank Wolfgang Franz, Winfried Pohlmeier, seminar participants at Mannheim and Konstanz, and two anonymous referees for helpful comments, the Deutsche Forschungsgemeinschaft (DFG) for financial support, and Monika Buhl, Klaus Kornmesser and Eva Kurz for excellent research assistance. ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 392 Sikandar Siddiqui Firstly, the multi-period structure of panel data enables the researcher to take the possibility of unobserved heterogeneity among individuals into account; and secondly, it allows to treat an individual's decision about the timing of retirement as a sequence of choices among multiple discrete alternatives (defined as the remaining possible retirement ages) under the assumption that a person decides to quit the labour force as soon as the option of an immediate exit from the labour force is more advantageous than any of the remaining possible choices. However, the problem of endogenous sample selection that arises whenever „retirement" is understood as an irrevocable decision (as it is done here) poses an important difficulty in this context. Continuous-time duration models, the use of which has been advocated by Hausman and Diamond (1984) and, more recently, Borsch-Supan and Schmidt (1993), offer a straightforward solution to this problem; yet as there is no obvious way in which the multiplicity of alternatives, which is typical of the retirement decision, can be implicated in such a model, this approach is not adopted here. Instead, the analysis presented in this paper is based on a multi-period multiple-choice model, in which unobserved heterogeneity is captured by time-invariant random effects and the problem of endogenous sample selection is accounted for by regarding the entire sequence of decisions made between the earliest possible retirement age and the one actually chosen as a single observational unit. Compared to the dynamic programming models employed, e.g., by Rust (1989) and Berkovec and Stern (1991), and to the option value approach developed by Stock and Wise (1990) and recently extended by Pohlmeier (1993), the model applied here is set in a rather simple utility maximization framework. This is due to two main reasons: Firstly, both dynamic programming and option value models require that the optimization problem which the individuals are assumed to face is solved for every person in every iteration of the maximum likelihood algorithm anew, which greatly increases the computational burden of the respective model's empirical implementation. Secondly, and more importantly, virtually all of these approaches are based on rather restrictive assumptions regarding the correlation pattern of the error terms entering the time- and alternative-specific utility levels. The model presented here circumvents these difficulties; yet one feature it has in common with the above-mentioned approaches is that it describes the retirement decision as a trade-off between the gain in income which can be obtained by postponing one's exit from the labour force and the gain in leisure which is implied by early retirement. It is therefore capable of integrating possible labour (dis)incentives implied by the rules of the relevant pension system into the empirical analysis of retirement behaviour. The outline of the paper is as follows: In section 2, the assumptions of the model employed are described in detail. Section 3 contains a brief descrip- ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 Early Retirement in West Germany 393 tion of the data being used and of the way in which the endogenous variable is defined. The empirical results are being discussed in section 4, and the paper ends with a couple of final remarks and a brief summary. 2. The Theoretical Framework Following the model set up by Fields and Mitchell (1984), it is assumed that the utility level U that individual i can enjoy at time Vi > t if he (or she) chooses to retire at time can be expressed as a function of the expected present discounted value of all present and future income streams (which will be denoted by Y^n) here and in the following), the expected length of his (her) retirement period, Lit(ri), and a disturbance term ^(n). For the alternative-specific utility level Uu(ri) which is associated with retirement at time r*, the following functional form is specified: (1) Uitin) = InYititi) + PitlnLit{n) + eit(n) The parameter flu is a taste parameter which measures the relative intensity of an individual's preference for leisure and is allowed to vary among individuals as well as over time. It is assumed to be a linear function of a vector 7 of unknown parameters and of a set of exogenous variables, which are collected in the vector xu and represent certain socio-economic characteristics of the individuals observed: (2) fa = xit '7 If 7Tj(s|t) denotes the individual-specific probability that a person who was alive at time t is still alive at time s > and if Si is the last period in which 7Tj(s|t) is positive, the expected present discounted value of present and future income levels, Yit(ri), can be expressed as follows: (3) Yit(n) := J^Etiw^-TrMt) • (1 + p)<- + £ JEt(Mn)) • *i(s\t) • (1 + p)<- s=i s=r In this equation, wis denotes an individual's labour income at time 5, the variable biS stands for the amount of retirement benefits received by the respective person, Et is the expectations operator given the information set of time t. (Its use in this context is, of course, not meant to conceal the fact that the model described here is essentially set in a perfect-foresight world and does not account for the impact of income-related uncertainty on the utility levels associated with different alternatives). The age-specific survival ZWS 117 (1997) 3 26 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 394 Sikandar Siddiqui probabilities 7Ti(s|i) were computed from the (estimated) population averages published by the German Federal Bureau of Statistics in 1990. This implies the assumption that the mortality risk does not vary among individuals of the same age, which is, of course, somewhat implausible since differences in health, wealth, and possibly even education and marital status probably exert considerable influence on the life expectancy Yet as there does not appear to be sufficiently exact information on the magnitude of the impact of these factors, the above-mentioned way of proceeding might nevertheless be regarded as a reasonable simplification. The discount factor p is set to 0.03 for the sake of simplicity. Clearly, this is one of the most questionable characteristics of the model under discussion, since p mirrors the relative importance ascribed to present and future income streams by an individual. Nonetheless, the introduction of a fixed discount factor is a rather common practice in studies directed at assessing the relative financial attractiveness of different retirement ages under a given pension schemes (see, e.g., Schmidt, 1995). In this paper, it is adopted for the sake of computational simplicity in order to avoid the necessity of computing alternative-, individual- and time-specific values of Ya(-) in every iteration of the estimation algorithm anew. Due to a lack of appropriate data - the corresponding statements in the German Socio-Economic Panel, which is the data base used here, were found to be rather imprecise and incomplete -, the amount of assets held by an individual is not included as an argument of the utility function within the framework of this model. As a consequence, the coefficients referring to variables correlated with wealth will „absorb" some of the unobserved wealth effects, which has to be kept in mind when interpreting the estimation results. Assuming time intervals of unit length and recalling that the terms 7rj(s|t) denote age-specific survival probabilities computed from official estimates of the corresponding population averages, the expected duration of a person's retirement period can be expressed as (4) MrO = X>(s|i) s=r, Let tio denote the period in which person i first has the opportunity to retire and r™** the latest possible year of retirement for the same person. (For reasons that will become obvious later, ti0 is assumed to be the year of the corresponding individual's 58th birthday). This implies that an individual who has not retired until the beginning of period t can either opt for immediate retirement or plan to retire at one of the (r-™**^ - t) remaining fu- ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 Early Retirement in West Germany 395 ture years of retirement. Thus, an individual can be said to have the choice between a total number of (r-™8*'^ - Uo + 1) different retirement ages at time tio. These alternatives are henceforth numbered in consecutive order beginning with zero. In order to retain relative computational simplicity while adding some stochastic flexibility to what otherwise would be a restrictive multinomial logit or independent probit model, a so-called „one-factor" decomposition (Amemiya, 1985, p. 323 f.) is chosen as to the stochastic error terms £it{ri): (5) £it(ri) = 6rlt-ti0 -V + UitiTi) The error components Uit(ri) and V{ are assumed to be normally distributed with mean zero; the factors 80 to 6 (max) , are unknown parameters which T i tiO have to be estimated. From an economic point of view, the random effects Vj represent unobservable but time-invariant individual characteristics, while the 6 coefficients capture their influence on the utility levels associated with different retirement ages. In accordance with the general conditions for the identification of parameters in multinomial choice models, which are summarized, e.g., in the paper by Borsch-Supan et al. (1990), the following normalizations are introduced: (6) E[uis(ri)vi] = for all r*, (7) E[uis(rl)2} = 0.5, (8) ElwsMuisiri ')] = 0 for all n, n ' if s + s', (9) E[vi 2} = 1, and (10) ¿o=*i = l. If the number of panel waves, T, falls short of the number of possible retirement ages or if not all retirement ages are observed with sufficient frequency, even further equality constraints as to the covariance structure of the error terms have to be added. Following a suggestion made by Pudney (1989, p. 127-131), an individual is assumed to retire at time t if this is the period in which the utility level associated with the option of immediate retirement exceeds the maximum ZWS 117 (1997) 3 26* OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 396 Sikandar Siddiqui utility level that can be achieved by postponing retirement for the first time. Let fi denote the year of retirement that has actually been chosen by individual i (which is observable for the econometrician if it lies within the sampling period) and r* be the corresponding person's optimal year of retirement (which is a discrete random variable from the econometrician's point of view because it is a function of the unknown parameters and the stochastic components of the individual's utility function). Then, the probability that this individual retires at time t given he/she has not retired until t-1 and conditional on v can be expressed as follows: Pr(r* = t\r* > tl;t>) (max.) (11) = 'J] Pr[Utt(t) > Uit(r)\r; > t - 1;»] r=t+1 r(max.) = H $ [In Yit(t) - In Yit(r) + zit '7 • [lnLit(f) - InLit(r)\ + [<5t_ii0 - ¿>r_tJ -v] r=t+1 Here and in the following, $(•) denotes the cumulative distribution function and </>(•) the probability density function of the standard normal distribution. It follows from equation (11) that the unconditional probability of individual i retiring at time t is 7 t-i (12) Pr(r* = t) = Pr(r* = t\r* > t - l;u) • JJ [l - Pr(r* = s|rj > s - 1; v)]<P(v)dv -00 S=i'° (The probability of a person retiring before is zero by definition). As the term (r^ - £ii0) measures the length of the time interval between the earliest possible retirement age and the one actually chosen, this model can be understood as a duration model in which time is measured in discrete intervals. Seen from this angle, the right-hand side of equation (11) equals a hazard rate (conditional on v) with regard to the transition from work to retirement. Together with the alternative-specific income variables introduced in equation (1), the one-factor structure chosen as to the error terms renders this hazard rate dependent on age, allows for a large variety of correlation patterns among the hazard rates referring to different retirement ages and thus mirrors the fact that apart from the explanatory variables in use, individual-specific but time-invariant factors can influence the probability of an immediate exit from the labour force differently at distinct ages. It becomes obvious that the unknown parameters of the decision model described here can be estimated by the method of maximum likelihood. ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 Early Retirement in West Germany 397 Suppose that the individuals in the sample are observed during a period of T consecutive years. In this case, there will be some persons who did not choose to quit the labour force during the sampling period although they were already entitled to retire. Taking this phenomenon (which can be thought of as a kind of right-hand-side censoring if this model is understood as a duration model in discrete time) into account, and recalling that the date of retirement actually chosen is denoted by fiy the contribution of individual i to the sample likelihood function (At) can be expressed as follows: ( Pr(r* =?i) if ri < T («=> observed) I [1 - Pr(r*i = slr** > s ~ ^v)](f>{v)dv if n > unobserved) S=ti0 Then, the log-likelihood function for an entire sample of N individuals simply is N (14) lnA = ^lnAi i=1 and can be maximized by means of conventional numerical optimization techniques. 3. The Data 3.1 The definition of „retirement" In previous micro-econometric studies of retirement behaviour, a variety of different definitions of the endogenous variable have been used. To a certain extent, this is due to the fact that the questions of main interest within this field vary considerably among researchers. Another reason is that the informational content of the datasets in use differs from case to case. The striking differences among the results of previous empirical investigations concerning the labour force exit behaviour of elderly Americans can, in part, be attributed to these factors. Hurd and Boskin (1984), for instance, who use the „Retirement History Survey" (RHS) as a data base, only consider a person as retired if he or she has left the labour force finally and irrevocably. Contrary to that, in a study by Burtless and Moffitt (1984), which is based on PSID data, an elderly employee's retirement age is defined as the age in which a sudden and pronounced decline in the individual's number of working hours occurs. If the incentive effects of firm pension plans are being analyzed, it is most appropriate to equate a person's date of retirement with the period when he or she first receives pension benefits, as it is done by Stock and Wise (1990). ZWS 117 (1997)3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 398 Sikandar Siddiqui In an empirical study of retirement behaviour in West Germany conducted by Börsch-Supan (1992) using the 1984 cross-section of the SOEP, the period of retirement is defined as the year in which the number of a person's working hours per week first falls short of fifteen. However, as „retirement" is understood as an absorbing state in the model described in section 2, it seems more reasonable in this context to equate a person's date of retirement to the moment in which he/she first declares himself /herself retired. This is possible with data from SOEP because it contains a very detailed set of information on a person's present and past labour force status in which „retirement" is a separate category. One of the main goals of the empirical study described here is to explain the effect of Germany's pension system on retirement behaviour. Nevertheless, it is assumed here that individuals can opt for retirement up to two years before they reach the age of 60, which is the minimum age for the receipt of old-age pension benefits in this system. The reason for this assumption is that, according to the rules of § 105c of the German Employment Promotion Act („Arbeitsförderungsgesetz" = AFG), unemployed individuals aged 58 or above can receive full unemployment benefits without having to declare explicitly that they are willing to work as long as they are ready to apply for the receipt of pension benefits at the earliest possible age (see Bundesminister für Arbeit und Sozialordnung, 1993, p. 111). As an evaluation of the labour force histories of all male SOEP respondents born between 1915 and 1925 reveals, less then 3 % of all persons in this category who were either unemployed or out of the labour force at the age of 58 returned to employment afterwards. This indicates that setting the minimum retirement age („retirement" being defined as a state in which a person does not intend to return to paid employment) to 58 is a reasonable choice. 3.2 Explanatory variables The intensity of an individual's preference for leisure, which is measured by the parameter within the framework of the model described in chapter 2, is assumed to be influenced mainly by the person's current health status and educational background. In addition to these variables, the dummy variable PUBLIC, which indicates whether a person is or was a public sector employee, is included in order to control for the influence of possible differences between the pension systems applying to public and private sector employees, respectively. From a psychological point of view, it might be interesting to find out whether married persons exhibit a higher preference for leisure than others; therefore, a corresponding dummy variable (MARRIED) is taken account of. ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 Early Retirement in West Germany 405 Table 3 Parameter Estimates Variable Sample I: German citizens only Sample II: Germans and foreigners Estimate t-statistic Estimate t-statistic CONSTANT 2.3975 0.856 -1.5596 -0.729 D_DISAB 4.8436 4.232 4.4203 4.912 HAUPTS 2.1553 5.122 1.6587 4.115 REALS 2.8181 4.882 2.0863 3.655 ABI 1.2155 1.158 0.7307 0.702 UNI 2.2651 1.771 1.5887 1.328 MARRIED -0.4826 -1.193 0.5803 1.911 PUBLIC -1.7346 -3.246 -1.4168 -3.756 CH_MAL 1.2083 3.233 1.0620 3.137 UR -0.0619 -0.243 0.2959 1.535 FOREIGN - - 0.9819 2.772 ¿2 -2.8447 -1.351 -1.5456 -1.370 ¿3 -2.2364 -1.104 1.3586 13.769 1.1895 7.474 -0.7519 -0.717 ¿5 1.4450 8.029 1.4268 9.376 -2.6571 -1.162 -1.4604 -1.132 Mean 0 5.0754 4.5659 Mean Log Likelihood # of cases -1.1165 419 -1.0658 586 Unobserved wealth effects might also be one of the reasons why the preference for early retirement appears to be significantly lower than average among public sector employees. Another possible explanation for this phenomenon is that, for the vast majority of the public sector employees in Germany, the probability of involuntary early retirement following a dismissal by the employer is virtually reduced to zero by the relevant legal regulations. But also readers who simply consider public sector employees as overpaid and underworked on average compared to their private sector colleagues might regard the above-mentioned result as a confirmation of their verdict. The estimation results displayed in table 3 can be used as a basis for estimation of age-specific retirement probabilities. In figure 1, the estimated ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 406 Sikandar Siddiqui cumulative retirement probabilities are compared to the observed cumulative frequency distribution (as it can be computed from the observed agespecific hazard rates with regard to retirement) of the retirement ages of a group of male SOEP respondents born between 1917 and 1932, of which the persons whose data were actually used for estimation simply constitute an adequately selected sub-sample. age estimated actual Figure 1: Cumultative retirement probabilities (German und foreign males) Apparently, the model underestimates the probability of immediate retirement at the earliest possible retirement age, while otherwise „fitting" the data rather well. The fact that the probability of retirement at 65 is also underestimated is probably due to the problem of right-hand-side censoring mentioned above. 5. A Policy-Related Simulation Experiment Being deduced from a micro-economic decision model in which the individuals are assumed to balance the gain in income attainable by postponing retirement against the increase in leisure implied by opting for an early exit from the labour force, the results of the investigation presented here can be used as a basis for simulation experiments aimed at forecasting the manner in which assumed changes of the current pension system affect the frequency distribution of retirement ages. This is done in the following for two different cases. The first experiment relates to Germany's most recent Pen- ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 Early Retirement in West Germany 407 sion Reform Act, which was enacted on January 1, 1992 but of which the main constituents will only become effective after the turn of the century. One of the main goals of this new law is to reduce the financial attractiveness of early retirement by introducing an extra deduction of 0.3% from annual pension benefits for every month that lies between the date when a person first receives them and his/her 65th birthday. The second policy scenario of which the simulated outcomes are presented here is the introduction of a hypothetical, ,,non-distortionary" social security system in which the expected present discounted value of net transfer payments (pensions plus unemployment benefits minus taxes) is independent of the retirement age chosen. (As in the econometric model of section 2, the discount factor for future income streams is set to 0.03). In figure 2, the pension system prevailing during the sampling period is contrasted with the one coming into force after the completion of the 1992 pension reform and with the supposed nondistortionary system. This is done by comparing the age-specific expected present discounted values of the net transfer payments, computed under the assumptions of chapter 2 for a person aged 58 and with average earnings, corresponding to each of these three systems. 450000 -I 400000 3500001 300000 - 250000 II • • • llr^M W • • DM 200000 - 150000 • 100000 - 50000 - 0 It i i i r i i f i 58 59 60 61 62 63 64 65 N 17 Age at Labour Fore« Exit Figure 2: Expected present discounted value of net benefits depending on retirement age If, as it is frequently done, the expected present discounted value of net benefits is understood as a measure for the relative financial attractiveness of different possible retirement ages, one can conclude from this comparison that the financial incentives in favour of early retirement which have been typical of the prevailing pension system will be significantly reduced, but not completely eliminated by the 1992 Pension Reform Act. ZWS 117 (1997) 3 current system —<>— after completion of 1992 pension reform —non-distortionary system OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 408 Sikandar Siddiqui By replacing the original values of the income variables with the ones corresponding to the two alternative policy régimes studied here and subsequently re-estimating the age-specific retirement probabilities on the basis of these newly generated arguments, an estimate of the effect these policies can be obtained. In figure 3, the result of such an experiment is summarized. current system after 1992 pension reform » *> - non-distortionary system Figure 3: Estimated cumultative retirement probabilities (German an foreign males) Changes in the incentive structure of the underlying pension scheme can obviously exert considerable influence on the age-specific retirement probabilities. A replacement of the current German pension scheme by the one that will be in operation after completion of the 1992 pension reform raises the estimated average retirement age of men in Germany by 1.2 years, whereas the introduction of the non-distortionary system described above will result in an increase by 2.1 years. The question in how far such an increase might help to compensate the additional financial burden levied on Germany's social security system by the process of population ageing that is expected to take place during the next five decades, however, cannot be answered at this stage. A few words of caution ought to be said with regard to the simulation exercise of which the results are gathered here: As can be concluded from Lucas' (1976) critique of econometric policy evaluation, the validity of its outcomes depends critically on the correctness of the assumption that the true parameters of the decision models set up in section 2 are independent of the rules governing the current labour market and pensions policy. Moreover, as was pointed out by a referee, the fact that the income variables used in this ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 Early Retirement in West Germany 409 context are generated regressors rather than directly measured explanatory variables might affect the precision of the resulting estimates detrimentally. These problems undoubtedly are of high importance, yet there does not appear to be a manifest solution to them. 6. Summary and Conclusions In this paper, the retirement behaviour of male employees in West Germany was analyzed empirically using a sequential multi-period decision model for multiple discrete alternatives. The estimation results reveal that health-related factors play a key role in determining the age a person chooses for the transition from work to retirement. Education, too, does matter; yet the influence a person's educational status has on the probability of an early exit from the labour force seems to follow a rather complicated, non-linear pattern. The hypothesis that the tendency toward early retirement is significantly higher in times of high unemployment could not be supported from the data. A policy-related simulation experiment leads to the conclusion that a removal of the financial incentives for early retirement implied by the current German pension scheme would raise the average retirement age of men by 2.1 years. The main advantage of the approach presented here is that it adequately reflects the sequential nature of the retirement decision and allows for unobserved heterogeneity among individuals. One major point of criticism, of course, is that its relative computational simplicity necessitates the introduction of rather restrictive assumptions concerning the utility maximization framework it is based on. Moreover, the treatment of individual expectations as to future income streams and survival probabilities and the manner in which they are being formed has only been very sketchy in the model under discussion and clearly deserves more attention in future work. Another drawback might be that the approach presented here does not explicitly account for the possibility of rationing on the labour market. In spite of the fact that it is doubtful whether the informational content of the database used here is high enough to solve this problem in this particular case, the author believes it is the last-mentioned aspect that deserves most attention in future empirical research on retirement behaviour. ZWS 117 (1997) 3 27 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 410 Sikandar Siddiqui Appendix Table A.l Definitions of explanatory variables used for the earnings forecasts PEXP Potential labour experience; defined as age - years of schooling - 6 SM.FIRM Dummy variable. SM_FIRM = 1 if a person works in a small firm (less than 200 employees) LG_FIRM Dummy variable. LG_FIRM = 1 if a person works in a large firm (more than 2 000 employees) AGFF Binary variable indicating whether an individual works in the agrarian sector, in forestry or in fishery TRADE Dummy variable. TRADE = 1 if a person is employed in the trade sector AGEG50, -55, -60 Dummy variables for the age groups 50 to 54, 55 to 59, and 60, respectively UNM Dummy variable indicating whether an individual is unmarried M_WS Dummy variable. M_WS = 1 if someone is married and living with his / her spouse VOC_D Dummy variable pointing out whether the corresponding person holds a vocational degree TT.SCHOOL Binary variable indicating whether a person holds a degree from a trade school or technical school D85,..., D90 Dummy variables for the years 1985 to 1990. (1984 is the reference period) ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 Early Retirement in West Germany 411 Table A. 2 Descriptive statistics of explanatory variables in the earnings equation Variable Mean Standard Deviation Minimum Maximum PEXP/10 2.8019 0.9964 0.6000 4.9000 PEXP2 /100 8.8434 5.7015 0.3600 24.0100 SM.FIRM 0.2282 0.4197 0.0000 1.0000 LG_FIRM 0.3009 0.4587 0.0000 1.0000 UNI 0.0808 0.2726 0.0000 1.0000 TT.SCHOOL 0.0793 0.2702 0.0000 1.0000 VOC_D 0.0688 0.2532 0.0000 1.0000 MARRIED 0.8484 0.3587 0.0000 1.0000 ABI 0.1133 0.3170 0.0000 1.0000 HAUPTS 0.4764 0.4995 0.0000 1.0000 PUBLIC 0.2005 0.4004 0.0000 1.0000 AGFF 0.0048 0.0692 0.0000 1.0000 TRADE 0.0344 0.1822 0.0000 1.0000 FOREIGN 0.2394 0.4268 0.0000 1.0000 AGEG50 0.1755 0.3804 0.0000 1.0000 AGEG55 0.1478 0.3550 0.0000 1.0000 AGEG60 0.0848 0.2786 0.0000 1.0000 UNM 0.0962 0.2949 0.0000 1.0000 ZWS 117 (1997) 3 27: OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 412 Sikandar Siddiqui Table A.3 Cross-sectional Estimates of Earnings Equation Variable Wavel Wave 2 Wave 3 Wave 4 Wave 5 Wave 6 Wave 7 Income Equation CONST 6.7998 (72.989) 6.8628 (68.969) 6.8506 (60.795) 6.8481 (63.605) 6.8496 (56.037) 6.8566 (1.219) 6.9946 (43.078) PEXP 0.6589 (8.287) 0.6164 (7.715) 0.6879 (8.067) 0.6678 (8.301) 0.7570 (8.237) 0.7576 (0.197) 0.7155 (6.369) PEXP 2 -0.1202 (-7.933) -0.1102 (-7.318) -0.1238 (-7.907) -0.1179 (-8.179) -0.1343 (-8.353) -0.1315 (-0.208) -0.1261 (-6.812) ABI 0.0484 (0.898) 0.0319 (0.615) 0.0435 (0.626) 0.1419 (1.987) 0.0689 (1.218) 0.1101 (0.046) 0.0102 (0.199) REALS -0.0159 (-0.339) 0.0032 (0.069) -0.0516 (-1.016) 0.0048 (0.102) -0.0279 (-0.574) 0.0182 (0.009) -0.0327 (-0.631) HAUPTS -0.1483 (-3.628) -0.1590 (-3.988) -0.1936 (-4.452) -0.1491 (-3.668) -0.1668 (-3.903) -0.1395 (-0.085) -0.1636 (-3.547) TT_SCHOOL 0.1040 (1.769) 0.1161 (2.243) 0.1097 (2.113) 0.1059 (1.957) 0.0750 (1.800) 0.1073 (0.059) 0.1382 (2.980) UNI 0.2857 (5.862) 0.3160 (5.795) 0.2883 (4.540) 0.2608 (4.159) 0.2902 (5.588) 0.3329 (0.146) 0.4347 (7.654) MARRIED 0.2046 (4.730) 0.1806 (4.467) 0.1675 (3.922) 0.1897 (4.532) 0.1546 (3.368) 0.1576 (0.088) 0.1491 (3.056) FOREIGN -0.2455 (-5.572) -0.2598 (-5.815) -0.2915 (-6.026) -0.2819 (-6.965) -0.3029 (-6.785) -0.2437 (-0.141) -0.2610 (-5.881) VOC_D 0.1628 (2.629) 0.1299 (2.588) 0.0854 (1.517) 0.1107 (1.823) 0.1006 (1.493) 0.1221 (0.047) 0.1666 (2.232) PUBLIC -0.0743 (-2.302) -0.0653 (-2.019) -0.0382 (-0.996) -0.0500 (-1.301) -0.0509 (-1.249) -0.0791 (-0.050) -0.0809 (-2.200) TRADE -0.0065 (-0.078) -0.0124 (-0.157) -0.0425 (-0.509) -0.0371 (-0.384) -0.0338 (-0.389) -0.0011 (-0.000) -0.0415 (-0.678) AGFF -0.0530 (-0.351) -0.2502 (-1.821) -0.1977 (-1.035) -0.2253 (-1.446) -0.1774 (-1.029) -0.0736 (-0.008) -0.1360 (-0.643) SM.FIRM 0.0130 (0.419) 0.0040 (0.121) 0.0078 (0.244) -0.0089 (-0.268) 0.0079 (0.243) -0.0196 (-0.016) 0.0139 (0.410) LG_FERM 0.0541 (1.602) 0.0473 (1.474) 0.0767 (2.299) 0.0740 (2.397) 0.0878 (2.654) 0.0778 (0.062) 0.0928 (2.783) Selection Equation CONST 0.9986 (6.891) 0.9422 (8.064) 0.7977 (5.704) 0.7470 (5.949) 0.7067 (5.905) 0.7289 (0.143) 0.5624 (4.873) AGEG50 -0.0359 (0.297) -0.1304 (-1.191) -0.0617 (-0.597) -0.1037 (-0.876) -0.1306 (-1.076) -0.1001 (-0.022) -0.0855 (-0.896) AGEG55 -0.3595 (-3.020) -0.1969 (-1.961) -0.3083 (-3.388) -0.1225 (-1.359) -0.1006 (-1.269) -0.0651 (-0.021) -0.0330 (-0.437) AGEG60 -0.4577 (-1.955) -0.5200 (-3.646) -0.6680 (-6.277) -0.6611 (-6.743) -0.7818 (-8.810) -0.8685 (-0.244) -0.8780 (-10.898) M_WS -0.0302 (-0.194) 0.0276 (0.232) 0.2489 (1.723) 0.2118 (1.622) 0.2566 (2.148) 0.2390 (0.047) 0.2784 (2.383) UNM -0.0163 (-0.120) -0.0286 (-0.220) 0.1413 (0.828) 0.3855 (2.560) 0.2833 (1.708) 0.3415 (0.043) 0.2710 (1.718) an 0.1712 (7.702) 0.2017 (11.270) 0.2062 (10.732) 0.1757 (11.260) 0.1825 (9.694) 0.1520 (0.204) 0.2035 (11.346) "21 0.6036 (49.510) 0.6037 (38.987) 0.6091 (36.206) 0.6181 (45.009) 0.6186 (40.361) 0.6148 (1.192) 0.6359 (36.266) Mean Log- Likelihood -0.809942 -0.847198 -0.832625 -0.855222 -0.845093 -0.836828 -0.906481 ZWS 117 (1997)3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 Early Retirement in West Germany 413 Table AA Minimum Distance Estimate of the Earnings Equation Dependent variable: Log of monthly labour income in DM Income Equation Variable Coefficient t-Statistic CONST 6.886 169.734 PEXP 0.606 20.381 PEXP2 -0.107 -20.077 ABI 0.011 0.604 REALS -0.008 -0.519 HAUPTS -0.153 -10.067 TT.SCHOOL 0.102 5.846 UNI 0.378 19.786 MARRIED 0.177 11.097 FOREIGN -0.280 -16.361 VOC_D 0.133 6.537 PUBLIC -0.064 -5.311 TRADE -0.015 -0.547 AGFF -0.174 -2.778 SM_FIRM -1.6*10~4 -0.013 LG_FIRM 0.072 6.294 CH_MAL -0.075 -5.811 D85 0.003 0.125 D86 0.032 1.519 D87 0.089 4.285 D88 0.140 6.629 D89 0.212 10.266 D90 0.237 10.581 Selection equation Variable Coefficient t-statistic CONST 0.823 14.205 AGEG50 -0.068 -1.687 AGEG55 -0.088 -2.755 AGEG60 -0.074 -19.940 M_WS 0.218 4.872 UNM 0.161 3.180 FOREIGN -0.086 -2.413 CH_MAL -0.350 -11.064 D85 0.036 0.624 D86 0.117 2.019 D87 0.061 1.074 D88 0.104 1.819 D89 -0.050 -0.882 D90 0.011 0.202 Qll 0.219 39.493 Oi 21 0.582 117.508 X2{d.f. = 152) = 164.491, p-value = 0.2309 ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36 414 Sikandar Siddiqui References Amemiya, T. (1985): Advanced Econometrics. Oxford (Blackwell). Berkovec, J./Stern, S. (1991): Job Exit Behavior of Older Men. Econometrica, 59, 189-210. Börsch-Supan, A. (1992): Population Aging, Social Security, and Early Retirement. Journal of Institutional and Theoretical Economics, 148,533-557. Börsch-Supan, A. / Schmidt, P. (1993): Early Retirement in East and West Germany. Paper presented at the „Conference of Users of the German Socio-Economic Panel: Using Panel Data to Answer Policy Questions", Berlin, July 7-8. Börsch-Supan, A./Hajivassiliou, V./Kotlikoff, L./Morris, J. (1990): Health, Children, and Elderly Living Arrangements: A Multiperiod-Multinomial Probit Model with Unobserved Heterogeneity and Autocorrelated Errors. NBER Working Paper No. 3343, Cambridge/Massachusetts. Bundesminister für Arbeit und Sozialordnung (ed., 1991): Übersicht über die soziale Sicherheit. Bonn. - (ed., 1993): AFG. Das Arbeitsförderungsgesetz. Bonn. Burtless, G./Moffitt, R. (1984): The Effect of Social Security Benefits on the Labor Supply of the Aged. P 135 - 171 in: Aaron, H. and G. Burtless (ed.): Retirement and Economic Behavior. Washington D.C. Diamond, P. A./Hausman, J. A. (1984): The Retirement and Unemployment Behaviour of Older Men. In: Aaron, H. and G. Burtless (ed.): Retirement and Economic Behavior. Washington D. C. Döring, D. (1980): Das System der gesetzlichen Rentenversicherung: Eine sozialpolitische Einführung. Frankfurt / New York (Campus). Genosko, J. (1985): Arbeitsangebot und Alterssicherung. Regensburg (Transfer-Ver- lag). Gouriéroux, C. /Monfort, A. (1989): Statistique et Modèles Econométriques. (Two volumes). Paris (Economica) Hurd, M. D JBoskin, M. J. (1984): The Effect of Social Security on Retirement in the Early 1970's. Quarterly Journal of Economics, 99, p. 767 - 790. Laisney, F./Lechner, M./van Soest, A./Wagenhals, G. (1993): A Life-Cycle Labour Supply Model with Taxes Estimated on German Panel Data: The Case of Parallel Preferences. The Economic and Social Review, 24, p. 335 - 368. Mincer, J. (1974): Schooling, Experience, and Earnings. New York. Mitchell, O. S ./Fields, G. S. (1984): The Economics of Retirement Behavior. Journal of Labor Economics, 2, p. 84 - 103. OECD (1988): Ageing Populations: The Social Policy Implications. Paris. Petersen, H.-G. (1986): Theorie und Praxis der Alterssicherung: Stand, Ansatzpunkte für Reformen und ihre Auswirkungen in der Bundesrepublik Deutschland. Gießen. Pudney, S. (1989): Modelling Individual Choice: The Econometrics of Corners, Kinks and Holes. Oxford (Blackwell). Pohlmeier, W. (1993): Einkommen, Unsicherheit und Ruhestandsentscheidung. Eine Optionswertanalyse der Erwerbsbeteiligung älterer Arbeitnehmer. Habilitationsschrift; Fakultät für Volkswirtschaftslehre und Statistik, Universität Mannheim. ZWS 117 (1997) 3 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.117.3.391 | Generated on 2023-04-04 12:23:36