Modelling old‐age retirement: An adaptive multi‐outcome LAD‐lasso regression approach

Using unique administrative register data, we investigate old-age retirement under the statutory pension scheme in Finland. The analysis is based on multi-outcome modelling of pensions and working lives together with a range explanatory variables. An adaptive LAD-lasso regression method applied to obtain estimates earnings socioeconomic factors affecting decide which these variables should be included our model. proposed statistical technique produces robust less biased coefficient context skewed outcome distributions an excess number zeros some results underline importance late life course employment final amount reveal differences outcomes across groups. We conclude that promising could usefully employed studying various topics industry. Nous examinons les retraites prévues par le régime légal de finlandais en nous basant sur données uniques provenant des registres administratifs. Cette analyse se fonde une modélisation à résultats multiples et carrières professionnelles, ainsi que toute série explicatives. La méthode adaptative régression du lasso (avec écart type plus faible) été employée afin d’obtenir estimations revenus facteurs socio-économiques qui affectent retraites, décider inclure dans notre modèle. statistique proposée produit solides moins biaisées, contexte d’une répartition inégale d’un nombre excessif zéros pour certaines Les obtenus soulignent l’importance l’emploi lors dernières années la vie professionnelle calculer montant pension. Ils révèlent ailleurs différences entre groupes au niveau concluons constitue prometteuse, pourra servir étudier différentes thématiques liées aux pensions. A partir datos registros administrativos únicos, estudia jubilación el régimen obligatorio pensiones Finlandia. El análisis basa modelización con múltiples resultados las y vida laboral, así como un conjunto explicativas. Se aplica método regresión LAD-LASSO adaptativo para obtener estimaciones los ingresos factores socioeconómicos afectan jubilación, determinar cuáles esas debe introducirse modelo. técnica estadística propuesta genera sólidas menos sesgadas del coeficiente regresión, contexto distribuciones desequilibradas número excesivo ceros algunas Los subrayan importancia empleo últimos años laboral calcular cuantía pensiones, revelan diferencias percibidas por distintos grupos socioeconómicos. conclusión es una prometedora, puede resultar utilidad estudiar varios aspectos sector pensiones. Anhand einzigartiger Registerdaten wurde die Altersrente obligatorischen Rentensystems Finnland untersucht. Die Analyse beruht auf einer Modellierung der Renten anhand Arbeitsbiografien und verschiedener unabhängiger Variablen. Eine LAD-lasso-Regression angewendet, um abzuschätzen, wie sich Verdienst sozioökonomische Faktoren auswirken, zu ermitteln, welche dieser Variablen das Modell aufgenommen werden sollte. Das vorgeschlagene statistische Verfahren ergab bei asymmetrischen Ergebnisverteilungen robuste weniger verzerrte Schätzungen Regressionskoeffizienten sehr viele Nullen einigen erklärenden Ergebnisse zeigen, wichtig Einkünfte eine Erwerbstätigkeit im späteren Berufsleben für endgültige Rentenhöhe sind, sie machen deutlich, dass Unterschiede den Rentenhöhen sozioökonomischer Gruppen gibt. Schlussfolgerung lautet, ein vielversprechendes statistisches ist, gut Untersuchung Aspekte Rentenwirtschaft eignet. Используя уникальные данные административного регистра, мы исследуем выход на пенсию по старости в рамках государственной пенсионной схемы Финляндии. Анализ основан многорезультатном моделировании пенсий и трудового стажа вместе с рядом пояснительных переменных. Для получения предварительных данных доходам социально-экономическим факторам, влияющим старости, для принятия решения о том, какие из переменных следует включить нашу модель, был применён адаптивный многорезультатный регрессионный метод LAD-lasso. Предлагаемый статистический даёт надёжные менее смещённые показатели коэффициента регрессии контексте асимметричных распределений результатов избыточного количества нулей некоторых Полученные результаты подчёркивают, насколько сильно влияют источники заработка занятость почтенном возрасте окончательный размер пенсии, а также выявляют различия результатах пенсионного обеспечения разных социально-экономических группах. Мы пришли к выводу, что является многообещающим статистическим методом, который можно пользой применять при изучении различных вопросов отрасли. 我们使用独特的行政登记数据, 调查了芬兰法定养老金计划下的老年退休情况。分析基于养老金和工作生活的多结果建模以及一系列的解释变量。使用自适应的多结果LAD-lasso回归方法获得了影响老年退休的收入和社会经济因素估值, 并确定将其中哪些变量纳入模型。在结果分布偏斜和一些解释变量中零值过多的情况下, 拟议的统计技术产生了稳健且偏差较小的回归系数估值。结果强调了晚年收入和就业对最终养老金数额的重要性, 并揭示了不同社会经济群体的养老金结果差异。我们得出结论, 自适应LAD-lasso回归是一种前景光明的统计技术, 可以有效地用于研究养老金行业的各种主题。 باستخدام بيانات السجل الإداري الموحد، نُحقق في تقاعد الشيخوخة بموجب نظام التقاعد القانوني فنلندا. ويعتمد التحليل على النمذجة متعددة النتائج للمعاشات التقاعدية والحياة الشغيلة بالإضافة الى مجموعة من المتغيرات التفسيرية. ويطبق نهج الانحدار التكيفي المتعدد للحصول تقديرات لمداخيل الاشخاص والعوامل الاجتماعية والاقتصادية التي تؤثر سن الشيخوخة. ولتحديد أي هذه ينبغي تضمينها نموذجنا. تنتج التقنية الإحصائية المقترحة معامل انحدار ثابتة وأقل تحيزًا سياق توزيعات دخل مشوهة وعددًا زائدًا الأصفار بعض وتؤكد أهمية الدخل المتأخر مسار الحياة والتوظيف المبلغ النهائي للتقاعد، وتكشف عن الفروق نتائج المعاشات عبر الفئات والاقتصادية. ونستنتج أن هو أسلوب إحصائي واعد يمكن استخدامه بشكل مفيد دراسة مواضيع مختلفة مجال التقاعدية. dados exclusivos do registro administrativo, investigamos aposentadoria idade no regime pensão estatutária na Finlândia. análise baseia-se simulação vários aposentadorias e vidas ativas, em com uma gama variáveis Um adaptável regressão LAD Lasso é aplicado obter estimativas ganhos fatores socioeconômicos exercem efeitos sobre decidir quais dessas deveriam ser incluídas nosso estatística proposta produz tendenciosas distribuições renda distorcidas excessivo algumas Os enfatizam importância dos salários emprego curso recente da o valor revelam diferenças nos socioeconômicos. Concluímos adaptativa promissora poderia útil estudo tópicos setor previdenciário . In this article, aim preceding leading retirement. Like many other European countries, Finland has flexible policy, age ranging from ages 63 68 (see Kuitto Helmdag, 2021). This means individuals approaching have consider whether retire early at cost reduced or remain work until pensionable beyond claim full increased Under current rules target for cohorts born 1954 earlier years. For younger it progressively higher. 2020, cohort retiring was 1956; their years 6 months. Just over half (52 per cent) persons who retired 2020 were 1956, remainder either late. Our focus here study earnings-related 1 evolved much same way as elsewhere Europe. overriding been prolong postpone retirement, good progress made respect. Finland, statistics 2019 show effective 60.2 figure, however, by major schemes are disability scheme. Moreover, there still occupational arrangements force, especially public sector, lower ages. more accurate measure describe median age, 63.3 Finish Centre Pensions, 2020b). length closely related accrual ends Statistics indicate had 31.7 years, 35.6 2020a). Kuivalainen al. (2018) studied using similar data those used but descriptive account focuses question how longer careers overall level security. purposes here, use framework developed Whitehouse (2009). Two six general objectives towards risks identified relevant examination: i) coverage system, pensions, ii) adequacy benefits, key concern study. accrued simultaneously wages strategy yields valuable information different affect outcomes, is, Associated contributing factor effects often simple linear methods (see, example, Riedel, Hofer Wögerbauer, 2015) logistic methods, yield risk measures such odds ratio (OR) relative reduction (RRR). Another appropriate addressing topic duration Cox’s model Aranki Macchiarelli, 2013). also technique, namely least absolute deviations (LAD) method. considered particularly suitable analysing long-tailed distribution outcome. 2 To knowledge, not previously income, both may distributions. choice known even presence Together variable selection Tibshirani, 1996), indicates kept can left out. penalized estimation performs shrinking coefficients negligible zero leaves important intact. typically combined ordinary squares (OLS), limited capacity cope extreme values very large numbers augmented shrinkage technique. reliable situation where (zero stage course). joint income coupled examination high-quality aimed gaining better understanding system groups people. backbone Finnish comprehensive coverage. Statutory (first pillar) provision consists defined benefit pension, aims preserve individual’s pre-retirement reasonable degree. It covers practically all types upper limit Residence-based (including national guarantee pension) ensure minimum standard security provide protection against poverty. non-contributory nowadays plays smaller role, although remains source basic Earnings-related gainful employment, including self-employed. include private after reforms 1990s 2005 virtually identical benefits conditions receipt. chosen universal because they affected want analyse. role second pillar employer-specific third individual insurance modest vast majority pensioners Barr, outlined Figure 1. first than countries. First cover, practice, employees. occupation (second pillar), common play system. Third omitted analysis. Source: Authors. age. Since (cohorts 1948 after), driven up rising expectancy. 3 essence, each its own Remaining offsets effect expectancy Vidal-Meliá, Boado-Penas Settergren, 2009). According Lassila, Määttänen Valkonen (2014), indeed most tool coordinating timing (future) expenditures. up-to-date overview requires closer schemes. structure 2. average monthly 1,762 euros (EUR) (Finish 2021); includes benefits. Increasing people depend alone, without claiming benefit. histories continues grow, translating into need support old Regardless, meagre People persistently low fragmented courses must (national paid 580,000 (or 18.9 cent new retirees). relatively 4 114,000 3.5 broad sense, housing support, scope ties choices Given interested impact security, strictly focused associated factors. uses dataset drawn registers. background later stages shed light two dependent variables: (amount life. analysed simultaneously, strategy. cover history several apply (least deviations) part family. see potentially contribute chose order able perform estimate retirement-related among set cases larger observations. However, opposite true. Drawn registers comprised population (N=41,729). 10 random sample (n=4,197). life, Socioeconomic serve predictors gender, last employer, group, benefit, education, during latter contains missing values. From substantial point view, consist (in 2020) 18 end year (31 December 2019). Note survivors' SSA ISSA, 2018). useful comment additional collected. First, partial allowances; cash rehabilitation benefit; (varhennettu vanhuuseläke); 5 years-of-service (työuraeläke). Early being phased out, while currently beneficiaries. Second, small part-time Third, grouped three classes. reference class close cohort-specific age; classes retirees retirees. popular option early, around 61. Fewer deferring group person’s recent status period 2013−17. gives us reasonably about Level education highest completed 2018. analysis, classified follows: Gender (sex) categorized male (ref=reference group) female. categories farmers; self-employed; upper-level employees administrative, managerial professional occupations; lower-level clerical manual workers (ref); pensioners; long-term unemployed, students otherwise classified. There categories: (ref), Timing Last before employer self-employed farmers. primary secondary post-secondary non-tertiary short-cycle tertiary bachelor’s equivalent master’s summary shown Table These binary class-level practice categorical encoded dummy coding; words, coding compares (reference) level. variable, {0,1} standardization present natural, real-life interpretations. situations seen ordinal measure, classification used. undertook experiments Helmert Hayes Montoya, 2017). difficulties interpreting desire retain consistency persuaded instead. say context, measures. addition variables, continuous predictors: wage 2000 2019, 20 prior annual valid proxy course. high-dimensional problems, parameters estimated higher sample. cases, (OLS) does work. Simultaneous (Tibshirani, 1996; Li Sillanpää, 2012) approach sparse problems. well-known drawback. Even though improve accuracy (i.e. reduce variance estimates), introduce (typically downward) bias estimates. alleviate problem, suggested (Zou, 2006) so selected subject penalty unselected attract heavier penalty. If outlying observations, increase robustness simplified list properties typical well modelling, univariate Wang, Jiang, 2007) Möttönen 2015; Li, 2015). Multi-outcome (Yuan Lin, object function shrinkage-inducing suffer downward bias. this, Arslan (2012) (2021). motivation zeroes further found presented Wang Leng (2008). How, then, choose value tuning parameter λ? main determining plausible model, then Akaike’s criterion (AIC) Bayesian (BIC), example. prediction accuracy, cross-validation (CV) choice. 5-fold choosing al., showing amounts A.2 (Appendix) exhibits long tail consists, first, whose cut short permanent disability; second, unemployed inactive spells. Those poor health disabled rarely recover return labour market Overall, centres 35.9 (median 39.3 years), figure Appendix shows sub-groups population. men’s 1.5 women. Self-employed farmers regular Long-term unemployment, indicated career status, lasting scar development, (21.4 years). left, observations false measurements true modelled correctly. A.2, mean EUR 1,976/month, line figures official statistics. Pension vary significantly. Higher University graduates received twice education. high position (upper-level employees) contrast, (e.g. pensioners, workers) levels systemic problems inherent view described Salonen, Koskinen Nummi (2020). Note: asinh scale log-transformation. essential determinants observe gender between analyses meaningful research questions. sizes largest educational (40.0 cent). Occupational evenly spread groups: (31.9 cent), (20.9 (19.5 (59.5 opt forced early. provides detailed look plots systematic concentration outcomes. natural explanations specifies ceiling extraordinarily Having said that, combination fragmentary history, absences market, translate shortfalls As expected, men consistent lives. university oftentimes slightly shorter lives, transition market. clear inadequate despite careers. findings revealed workin