Pension systems were developed in Europe from the second half of the 19th century and the beginning of the 20th century. In most Continental European countries, the so-called Bismarck model was adopted, conceived as a mandatory protection system financed through contributions from both the employee and the employer (Fenge & Von Weizsäcker, 2010). In Northern Europe and the Anglo-Saxon countries, the so-called Beveridge model was developed, based on the universal recognition of a vital minimum financed from general budgets (Cigno, 2008; Ebbinghaus, 2021).

In 1977, the Institute of Social Security of the Catholic University of Louvain organised the "September Colloquiums" which concluded with the determination of what has become known as the "Louvain Code." This code outlined the architecture of the European social security model, consisting of three pillars: the first pillar constituted by public social security, the second by supplementary social security for employment and the third by individual instruments. Implementation across European countries has not been uniform, with significant differences in both the replacement rate and the relevance of each of the three pillars (Montanari et al., 2008).

A significant challenge facing developed societies, particularly European ones, is the ageing of their populations, due to declining birth rates and a sustained increase in life expectancy. This demographic shift impacts pensions (Rey-Ares et al., 2018) and the labour market (Gurvich & Ivanova, 2018).

Given the foreseeable difficulty in financing public pensions, countries have actively pursued reform policies, at varying speeds and extents. This has primarily involved parametric reforms that make the existing pay-as-you-go system less generous and the establishment of a funded contribution system to complement or partially replaces the pay-as-you-go system, enhancing the replacement rate of future retirees (Arcanjo, 2019; Bonenkamp et al., 2017; Börsch-Supan et al., 2005)

Spain is currently at a pivotal moment in the reform of its pension system, covering all three pillars. Legislative developments align with some recommendations of the latest report on the evaluation and reform of the Toledo Pact (Congreso de los Diputados, 2020). The reform of the second pillar, through the recently approved Law 12/2022 regulating the promotion of occupational pension plans (Spain, 2022), aims to foster the development of supplementary social provision at company level. As the second pillar is voluntary, the success of this initiative depends on the success of the social dialogue and the transfer of its results to sectoral collective bargaining agreements.

Amidst a consistent increase in overall public pension expenditure (Harper, 2014), and parametric reforms of the first pillar aiming to reduce the initial pension amount (for the Spanish case see Díaz-Saavedra, 2020), promoting occupational pension plans could serve as an alternative for families to save in the long term, supplementing Social Security public pension in retirement (Mussio & Garabato, 2009). The extended scope of this initiative may increase the population's awareness of retirement savings products. Since not all companies promote occupational pension schemes, many workers will not have the option of joining this second pillar instrument, instead they will have to rely on individual private pension plans as retirement savings alternatives. However, in contrast to other countries such as the Netherlands, Switzerland and the United Kingdom, individual private pension plans are underdeveloped in Spain (OECD, 2023).

The debate on the complementary or substitutable nature of the second and third pillars is ongoing, both in the specialised literature (Higgins, 1985; Schoeplein, 1970) and at the social and political level. In Spain, the current reform, encouraging the promotion of occupational pension schemes and establishing a common legislation for occupational and individual pension schemes, may be detrimental to the development of the third pillar (Schoeplein, 1970). However, according to Gómez-García & Villanueva (2022), this effect is positive, suggesting that the development of occupational pension schemes would also stimulate the development of individual pension schemes. Nevertheless, a comprehensive debate has emerged within the financial-insurance sector on this issue.

This paper analyses, quantifies and determines the relationship between the second and third pillars using data from the Spanish Survey of Household Finances 2020 (Banco de España, 2022b). The primary objective is to examine whether the development of the second pillar influences the development of the third pillar. This analysis is conducted at both the household and individual levels, focusing on the reference person. Additionally, we explore whether the economic profile of households has any effect on this relationship between the second and third pillars, utilizing the employment status variable of the reference person and several household-related variables.

One of the key findings of this paper underscores the importance of distinguishing the household member participating in the occupational pension plan when evaluating the impact of the second pillar on the third pillar through the reference person. If the reference person is the participant, the effect is negligible. However, if it is another household member, there is a 10.1% increase in the likelihood that the reference person possesses a third pillar instrument. In households with substantial real assets and anticipating lower future income (evidenced by high disposable income, low indebtedness, and the reference person nearing retirement), this percentage escalates to 46%.

When confronted with the question of whether the second and third pillars in Spain are complementary or substitutes, our study allows us to conclude that, at a global level, these two pillars complement each other. Policies aimed at promoting the development of the occupational pillar also have a positive influence on the number of contracts for instruments of the pillar focused on individual savings.

This paper makes a valuable contribution to both the scientific and social debates. It delves deeply into the relationship between savings in third and second pillar instruments in the specific case of Spain, providing objective data for reflection on the application of measures for the development of supplementary social provision.

This paper is organised as follows. In the next section, we review previously published papers dealing with the factors that influence household savings decisions, with special emphasis on those related to retirement. In the third section we define the sources of information and discuss the statistical methodology used (logit and cluster analysis). In the fourth section we establish three logit models, where the dependent variable is whether the reference person is a member of an individual private pension plan and/or has a voluntary retirement insurance, these are used to test the hypotheses. We divide the general population into 3 homogeneous clusters based on economic variables. Following that, we present the empirical results, distinguishing between those applied to the general population and those resulting in each of the clusters. Finally, the paper concludes with a discussion of the results and a summary of the main findings.

Household savings decisions are influenced by two types of factors: those linked to the global economic situation (macroeconomic and political factors), and on the other hand, those arising from the household’s internal structure. The latter includes factors related to each of the members of the household (age, gender, marital status, education, occupational sector, employment status, among others) and factors considering the family unit as a whole (level of risk aversion, real income and assets, debts or the number of household members). Households' savings behaviour is determined by interpretation and prioritisation of these factors.

Macroeconomic factors play an important role in household saving decisions. Saving is influenced by a complex interaction of macroeconomic variables, encompassing economic growth, inflation, consumer confidence, interest rates and wealth effects. The impact of economic growth on saving remains a debated topic, with studies suggesting higher household savings due to higher incomes, precautionary motives, and favorable interest rates, decrease in savings due to higher spending on public services or potential economic imbalances if savings are excessively high. For an in-depth discussion, refer to Staal (2023) or Aizenman et al. (2019). Inflation, influenced by its present value and future performance expectations, can alter household savings behaviour (Howard, 1978).

Political factors can also significantly affect economic agents' decisions, directly impacting household saving behaviour through changes in taxation, social security (Bloom et al., 2007; Callen & Thimann, 1997 or Niculescu-Aron & Mihăescu, 2012) , retirement benefits and savings incentives. Political stability and effective governance contribute to economic stability, influencing household savings. Political decisions, including those promoted by governments, can influence long-term targeted savings instruments, such as IRAs in the US (Holden & Schrass, 2021), German Rürup-Rente or Riester-Rente (Metzger, 2017) or Spanish simplified employment plans. Changes in eligibility requirements, contribution limits or the structure of these programmes may influence the retirement saving behaviour of households.

Even within the same macroeconomic scenario and policy context, household savings decisions vary due to future income expectations. The life-cycle model, initially emphasising preparing for old age, has been extended to include various saving motives, such as precautionary saving (Ando & Modigliani, 1963; Carroll, 1997; Friedman, 1957 or Gourinchas & Parker, 2002) housing-related motives (Hayashi et al., 1988) and inheritance (Hurd, 1987). Life-cycle theory allows for the analysis of retirement savings decisions by providing a model in which individuals anticipate their financial needs over their lifetime, considering factors such as employment status, occupational sector, disposable income, assets fungibility (Fall et al., 2001 or Han et al., 2009), age, in particular, the time remaining until retirement (Kaplan et al., 2014). Ekerdt (2009), revealing that the closer to retirement, the more motivated workers were to plan their pension savings. Life-cycle theory helps to explain how people plan for a stable standard of living throughout their lives and make informed savings, consumption and investment decisions.

Retirement savings decisions are also influenced by family structure as defined by marital status, number of members (Tamborini & Purcell, 2016), the gender of family members (Lee et al., 2018) and level of education (van Rooij et al., 2011), as well as other subjective factors such as perceived health status (Ricketts et al., 2013) and the degree of risk aversion (Hariharan et al., 2000; Hilgert et al., 2003). Marital stability correlates with higher savings, while divorce and widowhood are associated with lower financial well-being (Szinovacz & Deviney, 2000; Wilmoth & Koso, 2002). Other studies have shown similarities and differences in retirement preparedness between men and women in their 30s and 50s (Hassan & Lawrence, 2007a, 2007b). Education is another key factor affecting the likelihood of having a savings plan and the ability to make informed financial decisions. According to Cupák et al. (2019) retirement savings in supplementary pension plans are positively associated with financial literacy after controlling for a set of relevant socio-economic variables. For Barth et al. (2020) financial literacy and education acquired in the workplace (pension plans developed at the company level), can positively impact retirement planning and saving behaviour. Research suggests that less risk-averse individuals tend to lean towards savings mechanisms with a high-risk component (Jacobs-Lawson & Hershey, 2005), influencing retirement savings decisions (Sunden & Surette, 1998).

In the literature focusing on the Spanish case, Domínguez & Turner (2022) analyse the evolution of supplementary pension systems in the period 1994 to 2019. Díaz-Giménez & Díaz-Saavedra (2017) study the effects of two previous pension reforms. Antón et al.(2014) examines the impact of tax relief on retirement savings instruments, while Moreno-Herrero et al.(2017) explores the variables that influence the uptake of private pension plans.

To the best of our knowledge, the relationship between savings in third pillar instruments and second pillar instruments has not undergone comprehensive scrutiny either in general or in the specific case of Spain, from both theoretical and empirical perspectives. Our research specifically investigates this relationship between the second and third pillars using EFF2020. We study the connection between the second and third pillars within Spanish households where the reference person is between 18 and 65 years of age.

The study utilizes EFF2020 to create two databases: DB1, households with at least two members where the reference person and at least one other individual are between 18 and 65 years old, and DB2, households excluded from DB1 where the reference person is between 18 and 65 years old. Two main research hypotheses guide our study:

(H1) The development of the second pillar has an influence on the development of the third pillar, taking into account only the fact of having the product.

(H2) Concerning the reference person, being a member of an occupational pension plan has an influence on being a member of an individual private pension plan and/or having a voluntary retirement insurance.

To test these two hypotheses in the two databases, we propose three models, the fundamental characteristics of which are defined in the Table 2.

In addition, we study whether the economic profile of households has any effect on the relationship between the second and third pillar, determining the validity hypotheses H1 and H2 on the economic profile group in which the test is conducted.

In this study, EFF2020 has been used as a source of data.

In order to test the first hypotheses, we propose 3 logit models whose basic characteristics are shown in Table 2.

In the three models, the dependent variable is that the reference person is a member of an individual private pension plan and/or has a voluntary retirement insurance (henceforth referred to as "individual private pension plan" or "third pillar products"). The independent variables are classified into three groups. The first group includes a series of dichotomous variables, one of which is essential for testing the desired hypothesis. The other two groups encompass control variables, which in principle are also expected to influence the dependent variable, related to either the reference person (second group) or the household (third group). Table 7 details the explanatory variables included in the three models and Table 8 to 10 present descriptive statistics.

Table 4 illustrates the percentage of households with a reference person being a member of an individual private pension plan. The data reveals that 22.78% (20.83%) of households in DB1(DB2) possessed an individual private pension plan. This indicates that among the 10,205,214 (3,674,103) households, only 2,325,270 (765,395) had at least an individual private pension plan or a voluntary retirement insurance.

In DB1 and DB2, health and age exhibit similar effects, where better the health and increase age positively impacting the percentage. As expected, the percentage is zero for ages 18-24. The analysis of educational levels indicates that having an education level lower than high school negatively affects the percentage, as observed with singles and the unemployed. The occupational sector demonstrates significant discriminatory power, where the percentage of managers is three times higher than that of unskilled workers (skilled craftsmen and skilled workers in manufacturing, construction and mining) in DB1 (DB2). Gender also plays a discriminatory role, with 26% of men having an individual private pension plan compared to 19% of women in DB1, while in DB2 the percentages are reversed (17% of men and 24% of women). Among the household-related variables, in DB1 and DB2, the only one that clearly has a discriminatory effect on the percentage of households whose reference person is a member of an individual private pension plan is risk aversion, where risk-averse households have a 10 point difference compared to risky households. The evolution of the household's future income significantly distinguishes between those expecting lower future incomes and those who do not, with the number of household members only having a discriminative effect in DB2 for households with four members or more due to their composition.

The economic clusters are constructed based on per capita income of the household, ratio of expenditure to total household income, ratio of debt to income, ratio of debt payments to income, risk aversion, evolution of future household income, real assets of the household and employment situation of the reference person. The justification of the variables chosen for the construction of the clusters is based on previous works referred to in the literature review section (see for example Fall et al., 2001; Han et al., 2009; Hilgert et al., 2003 or Staal, 2023). We initially obtained four clusters, but one was discarded due to its small size (only three families). The analysis continues with three clusters. The number of surveys and the number of households they represent is detailed in Table 3. The two variables that actually determine the configuration of the clusters are real assets and evolution of future household income1. The importance of the latter variable is also shown in Table 12, where 99% of households in Cluster 1 expect lower future incomes, 99% of households in Cluster 2 expect to maintain their income level, and the same percentage of households in Cluster 3 anticipate higher future income. Real assets are also important to the definition of the clusters, with mean (and median) values higher in Cluster 1 than in Cluster 2, and higher in Cluster 2 than in Cluster 3.

Descriptive statistics for the rest of variables in each cluster can be found in Table 11, Table 12 and Table 13. Among these variables, age and the ratios of total expenditure, debt and debt payments to income indicate that households in Cluster 3 have much higher ratios to income compared to the other two clusters and also feature a younger reference person (only 18.72% are over 55 years old, in contrast to Clusters 1 and 2 with percentages of 43.47% and 36.29% respectively).

.

The percentage of households whose reference person is a member of an individual private pension plan in the different clusters is included in Table 4. In general, it is observed that this percentage is lower in Cluster 3, for all variables and databases. This is reasonable given that the reference person of Cluster 3 households is younger and these households have less income and less disposable income.

In this section, we first present the main results of the three logit models defined in Table 2, followed by the results of the logit models applied to the different economic groups.

The data reveal that approximately one in five households in Spain has an individual private pension plan. This ratio drops to one in six when focusing only on those households with low real assets, higher expected future income, characterised by low disposable income, and whose reference person is far from retirement.

Our study emphasizes the importance of distinguishing the household member participating in the occupational pension plan when assessing the influence, through the reference person, of the second pillar on the third pillar. If the reference person is the participant, the effect is nil; however, if it is another household member, the probability that the reference person has a third pillar instrument increases by 10.1%. This percentage rises to 46% in households with high real assets, lower expected future income, high disposable income, low indebtedness, and a reference person nearing retirement). These results, though not entirely comparable due to different surveys, models, and variables, contradict those presented by Moreno-Herrero et al. (2017), who worked at the household level without distinguishing the participant in the pension plan.

The results of this study show that the significance and effect of the variables related to the household and the reference person included in the logit analysis are highly sensitive to the household composition in terms of number of members and their ages. In households with two or more members aged between 18 and 65, the probability of the reference person having an individual private pension plan is unaffected by employment status, age, health status, risk aversion and real household assets. However, educational level and occupational sector have a significant impact, with a university education increasing the probability by 15%, while being single reduces the probability by 3%. In households where the only person aged between 18 and 65 is the reference person, gender is no longer significant, and educational level now heavily favours high school). Marital status (being widowed or divorced) decreases the probability by 28% and 16%, respectively.

Some results regarding the relationship between household economic characteristics and the existence of a substitute or complementary effect between the second and third pillars lack sufficient explanation. The data aims to determine if there are mandatory occupational pension plans within the productive sector where the individual in question is employed. The EFF2020 provides information on the occupational sector, which doesn't align with the National Classification of Economic Activities 2009 (Spain, 2007). The latter is closely linked to the established structures of social dialogue in Spain (Boj et al., 2022).

The personal socio-demographic variables used in this paper focus on the reference person, who is the one who answered the survey and that normally coincides with the person handling household economic matters (Banco de España, 2022a), making comparisons with other papers challenging (e.g. Moreno-Herrero et al., 2017). The Bank of Spain defines the head of household as the reference person designated by the household for the purposes of answering the survey if he is male, or the partner of the reference person, if the reference person is female, providing the partner resides in the household (Banco de España, 2022b). The problems arising from this anachronistic definition of the head of household will be analysed in subsequent papers.

The relationship between contributions to second and third pillar instruments will be explored when EFF data for the 2023, the year after the latest regulatory changes (Spain, 2020, 2022) becomes available.

The results highlight gender disparities in individual private pension plans, showing that women who live with other adults aged 18-65 have fewer plans than their male counterparts, while women who do not live with other adults aged 18-65 have more plans. Table 5 and Table 6 confirm these influences on the likelihood of having an individual private pension plan. On the other hand, it appears from the data that marital status and age also have an influence on the likelihood of having an individual private pension plan. Further analysis is needed to understand the cross-over effect between gender, marital status and age, shedding light on saving patterns in third pillar products with homogeneous groups.

In resolving the dilemma of whether the second and third pillars in Spain are complementary or substitutes, this study concludes that, globally, these two pillars complement each other. Policies promoting occupational pillar development positively influence contracts for individual savings instruments. These findings provide objective data that can help the social partners, the social welfare industry and legislators to contemplate measures for the advancement of complementary social welfare.