How does real estate income affect retirement age?: Lessons from national survey of tax and benefit^*

The realm of retirement research is vast and encompasses numerous studies; while it’s impossible to provide an exhaustive list, the majority of these studies tend to focus on analyzing the factors influencing actual retirement age. Notably, as the most recent domestic study, Kim et al. (2021) unveiled the significance of factors such as physical and biological aging, employment stability, and gender in the retirement decision-making process. Common themes emerge from domestic research regarding the determinants of retirement. For instance, deteriorating health status tends to correlate with earlier retirement (Kim et al., 2015), retirement timing often hinges on job security (Park, 2003), and men typically retire later than women (Kim et al., 2015). Furthermore, due to the distinctive dynamics of family relationships in Korea, it’s argued that parents tend to delay retirement when they have dependent children (Kim et al., 2015; Kim et al., 2021).

Conversely, there are conflicting findings regarding income levels. Individuals with lower incomes often prolong their working years to make ends meet, while even higher-income individuals with specialized skills or expertise may delay retirement due to the high opportunity cost associated with leaving the workforce (Kim et al., 2015). An outlier in Korean retirement research is Son (2010), who focused on real estate as a determinant of retirement. This study indicated that individuals with greater real estate holdings tend to have higher post-retirement satisfaction, but it did not establish a clear link between real estate assets and retirement age. The absence of a significant relationship between real estate assets and retirement age in Son’s study (2010) may be attributed to the fact that, as previously discussed, retirement timing in Korea is predominantly influenced by involuntary factors rather than voluntary decision-making. Furthermore, Son’s study (2010) has limitations due to its reliance on cross-sectional analysis rather than panel analysis. This limitation arises from the use of only the first-year data from the Korean Longitudinal Study of Aging (KLoSA), making it impossible to ascertain whether retirement decisions were influenced by real estate assets or if real estate assets were acquired or disposed of after retirement.

Korea has been experiencing a diversification in the distribution of retirement ages, a departure from the past. This shift is primarily driven by the prolonged life expectancy and improved physical health among the population. As life expectancy increases and overall physical status improves in Korea, the retirement landscape is becoming increasingly diverse. Gradual retirement, partial retirement, and post-retirement re-employment have become more prevalent, as noted by Cho (2014). Indeed, according to Statistics Korea (2022), a survey conducted among individuals aged 55 to 64 revealed that while the average age at which they retired from their primary occupation was 49.3, a remarkable 69.9% continued working after their initial retirement. Even individuals in their 70s and 80s display a strong desire to postpone retirement. In Korea, which grapples with one of the world’s highest rates of elderly poverty, many elderly citizens are compelled to continue working for their livelihoods rather than pursuing work as a preference.

Studies conducted in various other countries similarly identify involuntary factors like age, health, and economic crises as significant determinants of retirement (Anderson et al., 1986; Disney and Tanner, 1999; Loughran et al., 2001). However, in these countries, individual autonomous preferences often play a more pivotal role in retirement decision-making compared to Korea. Notably, many studies pertaining to retirement decisions in these countries have placed a strong emphasis on analyzing the impact of pension systems on retirement choices. Beginning with Feldstein’s groundbreaking study (1974), a consistent pattern emerges in subsequent research: as public pension accumulations increase, the likelihood of early retirement also rises. Given the profound influence of pensions on retirement, there is research suggesting that when governments raise the pension eligibility age to stabilize pension finances, it tends to lead to a delay in the retirement age (Gustman and Steinmeier, 2005). Historically, in the United States and Europe, the retirement age often coincided with the pension eligibility age. However, in recent times, governments have increased the pension eligibility age significantly, reaching into the late 60s and even 70s, in efforts to bolster fiscal stability. Consequently, this has decoupled the retirement age from the pension eligibility age, empowering individuals to make voluntary decisions about when to retire before reaching the pension eligibility age (Pilipiec et al., 2021).

Li et al. (1996) uncovered a nuanced aspect of retirement planning: a non-linear relationship between actual cash flows and subjective assessments of the adequacy of old-age resources. Interestingly, even when individuals experienced relatively modest cash (income) flows, they often perceived their retirement income as sufficient. Conversely, those with more substantial cash flows sometimes felt their retirement income remained inadequate. As a result, the timing of retirement might be influenced more by subjective judgments about the adequacy of cash flows rather than the objective cash flow amounts themselves (Li et al., 1996). Furthermore, it’s worth noting that the utility derived from work itself, independent of objective or subjective assessments of income, assets, or pensions, could lead individuals to lean toward postponing their retirement (Pilipiec et al., 2021).

Some studies have focused on individuals’ anticipated or desired retirement age rather than their actual retirement age. In this regard, Montalto et al. (2000) discovered that as individuals age, a preference for delaying retirement tends to emerge. Intriguingly, they also observed that individuals with higher current income levels are more inclined to postpone their retirement.

A prevalent trend observed in both Eastern and Western societies in recent years is the inclination to postpone retirement. In Western countries, this tendency is amplified by the availability of better health conditions that enable individuals to continue working well into their advanced years (Hess, 2017; Wheaton and Crimmins, 2012).

In Western countries, the timing of retirement is primarily influenced by the sufficiency of pension income, as highlighted by Montalto et al. (2000). Conversely, in Korea, where real estate assets are predominant rather than pensions, this study conducts an empirical analysis of retirement decision-making with a focus on real estate income. Additionally, taking into account the distinctive feature of the Korean context, where retirement timelines are often determined involuntarily, this research examines individuals’ expected or preferred retirement age rather than their actual retirement age.

<Eq. 1> can be employed to assess the influence of real estate income on the retirement age.

In <Eq. 1>, RetireAge_it represents the anticipated retirement age of individual i at time t as self-assessed by the individual. RealEstate_it corresponds to the real estate income of individuals. The vector X_it encompasses explanatory variables that may impact an individual’s expected retirement age, excluding real estate income. α denotes the regression coefficient, and β signifies the vector of regression coefficients. The primary objective of this study is to estimate the value of α. If α is greater than zero (α>0), it indicates that retirement age tends to be postponed with an increase in real estate income. Conversely, if α is less than zero (α<0), it suggests that an increase in real estate income is associated with an earlier retirement age. Finally, u_it represents the error term in the equation.

However, within the framework of <Eq. 1>, there is the potential for a correlation to exist between the primary explanatory variables and the error term. To address this challenge, one potential solution is to employ a fixed-effects model (Cameron and Trivedi, 2005; Greene, 2003). The fixed-effects model mitigates the potential correlation which gives rise to endogeneity issues, through a subtraction process, as illustrated in <Eq. 2> below.

where $\overline{{\text{RetireAge}}_i}={\displaystyle \sum _{t=1}^T{\text{ Retire Age}}_{it}}$, $\overline{{\text{RealEstate}}_i}={\displaystyle \sum _{t=1}^T{\text{RealEstate}}_{it}},\overline{X_i}={\displaystyle \sum _{t=1}^T{X}_{it}}$, and

To determine the most suitable modeling approach, whether fixed or random effects, when there is no discernible correlation between the primary explanatory variables and the error term, a random effects model is often preferred (Cameron and Trivedi, 2005; Greene, 2003). The appropriateness of the chosen model can be confirmed through the Hausman test (Hausman, 1978). In all the analyses conducted within this study, the Hausman test consistently affirmed the suitability of the fixed effects model. However, this study opted to present the analysis results for both models. The rationale behind this decision lies in <Eq. 2>, where all time-invariant variables, such as gender, are eliminated from the fixed effects model. Particularly, a random effects model becomes indispensable when analyzing the impact of variables that are not entirely time-invariant but exhibit relatively low volatility over time, such as residence and education level.

In <Eq. 1>, apart from the endogeneity concern, issues related to causality or reverse causality may also arise. For instance, choices pertaining to asset allocation and savings behavior can differ based on an individual’s intended retirement age (Lusardi, 2006). To address this complex interplay, we conducted an empirical analysis using panel data, tracking each individual’s trajectory and observing how their expected retirement age evolves when their real estate income undergoes changes. Additionally, we analyzed both unbalanced and balanced panels to ensure the robustness of our findings. Furthermore, while the initial analysis encompassed all household members, subsequent investigations focused specifically on the household head and their spouse. Finally, to validate the robustness of our empirical analysis, we conducted additional assessments. This involved replacing the primary explanatory variable in <Eq. 1>, real estate income, with a dummy variable indicating the presence or absence of real estate income, as well as the ratio of real estate income to total household income.

For our empirical analysis, we utilized data spanning from 2016 to 2021, sourced from the National Survey of Tax and Benefits (NaSTaB). The NaSTaB offers comprehensive real estate income and personal information for each member within households, making it a rich resource for our research³⁾.

<Table 1> displays the variables employed in the empirical analysis conducted and provides their respective definitions. In line with our earlier discussion, the primary explanatory variables used in <Eq. 1> or <Eq. 2> encompassed the presence or absence of real estate income, real estate income itself, and the ratio of real estate income to household income. As the dependent variable, we opted to use expected retirement age, which can also be interpreted as the preferred retirement age.

The set of other control variables encompassed age, work type, marital status, health status, education level, gender, and residence. Given the prominence of real estate income as the primary variable and the tendency in Korea for individuals to enter the labor market relatively late in life, our analysis focused exclusively on individuals aged 30 to 64. Furthermore, considering that the key variable of interest is expected retirement age, we excluded individuals who had already retired or were not part of the workforce during the analysis period spanning from 2016 to 2021. Among the control variables, age, work type, marital status, and health status are variables that could be integrated into the fixed effect model. Meanwhile, gender, region, and education level were treated as time-invariant variables. Although education level and region are not strictly time-invariant from a theoretical standpoint, they were included solely in the random effects model. This decision was based on the fact that our analysis encompassed individuals aged 30 or older, and there were relatively few instances of residential region changes during the analysis period. For each categorical variable, we established a reference group comprising individuals in their 60s, self-employed, single, with less than a high school education, females, and residents of Seoul.

In the unbalanced panel analysis involving household members, we utilized a dataset comprising 31,357 samples. For the unbalanced panel analysis focused on household heads and their spouses, we worked with 29,007 samples. In the balanced panel analysis involving household members, we utilized a sample size of 16,070 (with 3,214 samples in each year), while for the balanced panel analysis focused on household heads and spouses, we worked with 15,130 samples (with 3,026 samples in each year).

<Table 2> provides descriptive statistics for the variables utilized in the empirical analysis. We conducted a t-test to assess the mean differences between groups with and without real estate income.

The average expected retirement age across the entire dataset was 65.95 years. However, individuals with real estate income had a slightly higher expected retirement age of 66.84 years compared to those without real estate income (65.91). Out of the total sample of 31,357 individuals, only 1,329 (4.23%) reported having real estate income. When considering only the household head, this percentage increases to 4.80%. For those with real estate income, their annual earnings from real estate averaged 15.13 million won, equivalent to a monthly average of 1.26 million won. Real estate income constituted 17.05% of their total income.

People with real estate income had an average annual household income of 92.87 million won, which was 1.36 times higher than the 68.53 million won annual income of those without real estate income. Notably, the income disparity between these two groups amounted to 24.34 million won, exceeding the difference in real estate income (15.13 million won). This observation underscores the strong correlation between real estate income and household income, suggesting a virtuous cycle where a high household income serves as an investment to generate real estate income, subsequently increasing household income further (Kim, 2021).

The average age of individuals with real estate income was 52.2 years, notably higher than the average age of those without real estate income, which stood at 47.96 years. Interestingly, among individuals with real estate income, wage workers comprised 59%, whereas among those without real estate income, the percentage of wage workers was notably higher at 71%. Conversely, the proportion of self-employed individuals among those with real estate income was 41%, significantly exceeding the 29% among those without real estate income.

Among individuals with real estate income, 88% were married, a higher percentage compared to those without real estate income, where 80% were married. Subjective health status, assessed on a scale of 1 to 5, was slightly lower for individuals with real estate income (3.80) than for those without real estate income (3.86). While it’s common for education levels to decline with age, individuals with real estate income exhibited a higher level of education despite being older than those without real estate income.

<Table 3> displays the results of the analysis concerning the variation in expected retirement age based on the presence or absence of real estate income. Model 1 and Model 2 represent the outcomes obtained from the random effect model and fixed effect model, respectively. The fixed effect model reveals that individuals who initially lacked real estate income extended their expected retirement age by 2.25 years following the acquisition of real estate income.

The analysis revealed that as income levels rise, individuals tend to extend their expected retirement age. Interestingly, there is an association between increasing age and a higher expected retirement age. In other words, when individuals were younger, they expressed a desire to retire early, but as they grew older, they became more inclined to work longer.

Marital status and the presence of a spouse were not found to have a significant impact on retirement age. However, there was a notable correlation between individuals intending to work longer and an improvement in their health status. The estimated coefficient indicates that for each unit of improvement in health status, assessed on a scale of 1 to 5, the desired retirement age increases by 0.28 years. These findings align with a study by Swedas et al. (2017), which emphasized that health plays a pivotal role in income-earning activities, with poorer health being associated with earlier retirement.

The results from the random effect model indicate that individuals who are male and possess a higher level of education are inclined to retire later. In comparison to residents of Seoul, individuals residing in Busan, Daejeon, Cheongbuk, Jeonnam, and Gyeongbuk exhibit a tendency to retire at a later age. Conversely, those living in Daegu, Incheon, Ulsan, and Gangwon demonstrate a preference for earlier retirement. However, there was no discernible difference in the expected retirement age of residents in Gwangju, Gyeonggi, Chungnam, Jeonbuk, Gyeongnam, Jeju, and Sejong, when compared to residents in Seoul.

<Table 4> displays the results of the analysis investigating the variation in expected retirement age as real estate income increases. The analysis revealed that as real estate income increased, the expected retirement age also increased. According to the estimated coefficient from the fixed effect model, the expected retirement age increased by 0.072 years for every 1% increase in real estate income. Therefore, if real estate income increases by 10%, the expected retirement age can be interpreted as increasing by 0.72 years⁴⁾. Furthermore, if real estate income doubles (increases by 100%), the expected retirement age can be interpreted as increasing by 7.2 years. Since the estimated coefficients of other variables align with those in <Table 3> in terms of sign, magnitude, and statistical significance, a detailed interpretation is not provided.

<Table 5> displays the results of an analysis examining how the expected retirement age changes with variations in the proportion of real estate income in household income. In the fixed effect model, the estimated coefficient indicates that the expected retirement age increases by 0.22 years for every 1% increase in the proportion of real estate income. In practical terms, if real estate income accounts for 10% of household income, the expected retirement age would increase by 2.2 years. It’s worth noting that the estimated coefficients for other variables exhibit consistency with those found in <Table 3> and <Table 4> in terms of direction, magnitude, and statistical significance. Therefore, a detailed interpretation is not provided here.