The period effects of crisis and recovery on life course and residential mobility of owner-occupants

Abstract

This study examines how the relationship between life course and mobility of owner-occupants is affected by periods of economic and housing market downturn and recovery. The impact of ‘period effects’ are largely unknown. Using Dutch register data, we compare the probabilities of moving in view of partnership, children and employment status for 2012–2014 and 2014–2016. We find that the downturn period is associated with lower mobility, yet the association is different for various household situations. Mobility to ownership in the crisis was particularly constrained for stable couples, employed owners and households with children. Moves to the rental sector were less period sensitive. Only owners who became unemployed were more likely to move into rental during crisis. ‘Delayed mobility’ has been found for moving in together, separation, households who had children, and job change. So, periods of crisis and recovery structure how home-owners adapt to life-course changes. Our findings imply that period effects should be accounted for in residential mobility studies.

Keywords:

• Housing crisis
• life course
• residential mobility
• period effect
• the Netherlands

A large body of literature on life-course and residential mobility has demonstrated that changes in household composition and employment are strong predictors of moving (eg, Rossi, 1955; Mulder & Hooimeijer, 1999; Findlay et al., 2015; Clark & Lisowski, 2017). As households undergo changes in their composition and employment situation, the distance between their current and desired housing conditions typically grows, leading to households seeking to reduce this distance by moving house. Whether such a move is possible is linked to a household’s finances, as well as to macro-level economic and housing market conditions (Clark & Huang, 2003; Coulter et al., 2016; Sander & Bell, 2016). While the relationship between life course and mobility has been studied in the context of housing crisis for specific groups (eg, Izuhara, 2015; Lennartz et al., 2016; Clark, 2016), there have been few analyses of how economic and housing market conditions may impinge upon the relationship between life course and residential mobility (Deurloo et al., 1994; Coulter et al., 2016). Hence, this article focuses on the ‘period effect’ of the economic and housing market downturn (and recovery) after the Great Financial Crisis (see Mayer, 2009).

For owners, periods of economic and housing market downturn may constrain mobility or force a move out of owner occupation. First, economic uncertainty may make owners reluctant to take on (additional) loans or to increase their housing expenditures (Kan, 2002; Marsh & Gibb, 2011). Next to this, an inability to finance the costs of moving in negative home equity, or the prospect to sell at a loss, could lead to owners being unwilling or unable to move house during a crisis (Chan, 2001; Ferreira et al., 2010; Van Veldhuizen et al., 2020). Consequently, owners may be less likely to move during crisis periods even when partnership, children and employment transitions take place. On the other hand, some owners may be unable to remain in their dwelling because of disruptive life-course events like job loss or divorce, potentially leading to a move out of owner-occupation and into rental housing.

To understand how mobility of owner-occupants is contingent on period, we compare how transitions or stable situations in partnership, children and employment affect moving behaviour for owner-occupants in the Netherlands during and after the housing market downturn following the Great Financial Crisis. We compare the final two years of the downturn (2012–2014) and the first two years of the recovery period (2014–2016). We examine moves within owner occupation and moves towards rental dwellings. During the downturn period, economic conditions deteriorated while house prices and number of housing transactions dropped (Scanlon and Elsinga, 2014; Van Veldhuizen et al., 2020). Yet, during these years, the number of defaults was relatively low even though four fifths of Dutch homeowners hold mortgage loans (Francke & Schilder, 2014). These general trends imply a relatively high level of immobility in the crisis period. This immobility may lead households that have experienced life-course changes, to adjust their housing at a later stage. To see whether the crisis period may lead to delayed mobility, a second set of analyses assesses mobility in subsequent periods to assess the effect of previously experienced life-course changes among immobile households.

Our analyses are relevant for understanding how life course and mobility are not only affected by constraints related to income, wealth and tenure but also by period. In a broader sense, our study may contribute to debates on how housing conditions, and particularly mortgaged ownership, can shape our lives. It has been well established that forced mobility can lead to a financial penalty, possibly with lasting material and psychological consequences (Clark, 2016; García-Lamarca & Kaika, 2016; Dwyer et al., 2016). Immobility can also have lasting repercussions. When households are unable to adjust to changing life circumstances by moving, they could face unsuitable housing conditions or increasing levels of mortgage stress (see Waldron, 2016; Waldron & Redmond, 2017).

Life course and residential mobility

Considering that human lives are stretched out over time and space, the relationship between life course and residential mobility has long been of interest (see Findlay et al., 2015; Coulter et al., 2016; Rossi, 1955). Many studies have shown that moves are associated with life-course changes or events (Rossi, 1955; Gotlib & Wheaton 1997). Changes in household size, economic and labour market situation of household members can typically lead to a growing divide between the desired and current housing conditions (‘housing disequilibrium’). A household may seek to resolve this discrepancy by making improvements to the dwelling or by moving house (Littlewood & Munro 1997).

While the life course can be shaped by a range of factors, many studies typically focus on partnership, children, and employment (Rossi, 1955; De Groot et al., 2011; Falkingham et al., 2016). By definition, moving in together and separating requires a move. With regard to which partner moves, considerations relate to income, location, family background and gender (see Feijten 2005; Blaauboer 2010). With childbirth, households typically re-consider the suitability and safety of the dwelling and its environment. In addition, access to schools and care amenities is typically re-evaluated (Boterman 2012; Kooiman 2020). In terms of employment, a change of employer may mean that the location of the dwelling is no longer suitable (Clark & Davies Withers 1999). Job change, or finding work, can also mean a change in social status and level of income, leading to new housing desires and options. Conversely, becoming unemployed may also ‘trigger’ a move when the rent or mortgage payments become unaffordable, or untenable (see Feijten 2005; Han et al. 2017). Looking at moving intentions and behaviour in the Netherlands, De Groot and colleagues (2011) found that, among people who had no stated intention to move, union formation, union dissolution, childbirth, job change, and becoming unemployed were particularly prone to lead to a move. These events can thus act as ‘triggers’.

The term ‘trigger’ does not imply immediacy. While moves can happen fast, the timing between life-course transitions, motivation, and an actual move is not set in a standard order or time frame (Coulter, 2013; Coulter et al., 2016). There may be a window of time in which an event and move take place (Mulder and Wagner 1993; Clark & Davies Withers, 2009). A move is not always ‘triggered’ after a change, but can be done in anticipation of life-course changes, like childbirth, or they can be co-constitutive of employment situation (see Feijten & Mulder, 2002; Clark & Davies Withers, 2007). Similarly, a move can be delayed or moving desires can be abandoned when adjustments cannot be made (Coulter, 2013; Waldron & Redmond, 2017). Whether a move can be made or has to be delayed (indefinitely), has been linked to education, income, and ethnicity and race (De Groot et al., 2011; cf. Clark and Lisowski, 2017). The ability to move after a life-course event is also determined by what kind of housing is available, accessible and affordable, as determined by housing policy regimes and housing market dynamics (Mulder & Hooimeijer, 1999; Li 2004; Kooiman, 2020).

Crisis period effects for owner occupants

The ability for households to realise a desired or triggered move is related to macro-level processes that affect both households and the housing market (Coulter et al., 2016). This suggests that periods of economic and housing market downturn (and recovery) will impinge upon the relationship between life course and residential mobility. This is referred to as a period effect: the collective experience of the structural conditions during several years (Mayer, 2009).

In periods of crisis, economic and housing market conditions may constrain the ability to move (Forrest, 1987). Economic uncertainty, low labour demand and income insecurity may make households less likely to move, particularly when it means taking on (more) debt or higher monthly costs (Kan, 2002; Marsh & Gibb, 2011). Mobility behaviour in the owner-occupied sector can be affected by financial conditions and concerns. Moves to and within the owner-occupied sector are generally subject to higher transaction costs than moves within or towards the rental sector (eg, stamp duty, notary and banking costs), and they require sufficient and stable household income. For this reason, lower income households are both less likely to enter owner-occupied housing, and have greater difficulty staying in it, particularly during a crisis (Deurloo et al., 1994; Helderman, 2007; Clark, 2016).

Falling house prices during downturns may significantly constrain mobility, most strongly for households with high loan-to-value ratios (Chan, 2001). When house prices decline, households lose accrued housing wealth. As a result, the proceeds that could arise from the sale of the house could be less than what it has previously been or even less than the outstanding mortgage loan (negative home equity), which can reduce the likelihood of a move (Ferreira et al., 2010; Steegmans & Hassink, 2018; Van Veldhuizen et al., 2020). Yet, households in negative home equity are also more likely to experience financial problems, job loss or divorce, which can result in involuntary moves (Francke & Schilder, 2014; Bricker & Bucks, 2016).

Negative home equity and falling house prices may affect mobility in different ways. A period of housing market downturn resulting in negative home equity may put households in a situation where they lack the financial capacity to cover the combined costs accompanied with moving. They are thus ‘locked-in’ (see Bricker & Bucks, 2016; Chan, 2001; Ferreira et al., 2010). The decision to move house, however, is not only a matter of rational choice, meaning that mobility behaviour is not just driven by utility and costs.

In their model of housing consumption, Marsh and Gibb (2011) argue that mobility behaviour may be structured by a range of behavioural factors, a number of which may be of particular relevance in periods of economic crisis and lower housing prices. These are loss aversion, anchoring and adjustment bias, and reference to social groups. First, households may be loss averse, meaning that they may be financially able, but not willing to make a residential move because of their reluctance to sell at a loss. A number of studies, comparing financial constraints and loss aversion, suggest that loss aversion is the most dominant mechanism through which mobility behaviour of owner occupants is hindered in times of crisis (Engelhardt, 2003; Steegmans & Hassink, 2018). Second, and related to loss aversion, anchoring and adjustment bias refer to a tendency in people to overvalue existing consumption conditions in relation to alternatives, particularly when there is uncertainty (Marsh & Gibb, 2011). Owners may continue to appreciate the value of their dwelling based on its worth pre-crisis, or they are reluctant to move unless they are confident that a new dwelling will be equally valuable. This implies that during uncertain times, owners are more hesitant to move to another owner-occupied dwelling. Lastly, decisions are informed by social reference. Households relate their housing situation to their social status, and will make decisions accordingly (eg, Boterman, 2012). More important here, ‘homeownership’ has long been seen as an indicator of social status (Van Gent, 2010; Robertson, 2017). This could mean that owners avoid moving out of ownership, despite changes in household composition or finances.

So, housing consumption behaviour is not fully rational, yet owners are still pressed to consider the financial ramifications of a move during a period of crisis, particularly when they have a mortgage loan. The literature on housing and the ‘financialisation of everyday life’ view such considerations as an internalization of a housing/financial regime (Aalbers, 2008; García-Lamarca & Kaika, 2016): decades of homeownership policies have ‘taught’ owners to view their house as an object of real estate investment rather than as a home. While such discourse is influential, owners are able to reflect on their position and are generally reluctant to take on mortgage debt, but they may see no other choice given their options and their responsibilities to family and care (Pellandini-Simányi & Banai, 2021), ie, life course.

Tensions between life course and housing finance in times of crisis

A crisis period may bring several tensions between the housing necessities related to life course – eg, moving in together, divorce, childbirth, job change- and the ability and willingness to move during times of downturn. Simply put, changes in the life course generally lead to a move, while a crisis period has been related to immobility. Of course, a period of crisis may also mean that people have no choice but to move. Mobility associated with disruptive events such as unemployment and loss of income is more likely in crisis periods (Clark, 2016). These moves are also more often out of ownership, and into rental housing. This implies that the likelihood to move to rental housing is higher during a period of crisis, particularly in response to disruptive life course events.

The following analyses will assess whether there is a crisis period effect on mobility behaviour of owners moving to ownership housing and to rental housing. More specifically, it will look at which changes (and stable situations) in partnership, children and employment are sensitive to period effects. Secondly, because a move may take place later than it normally would (see Clark & Davies Withers, 2009), subsequent analyses will also look at owners who haven’t moved and assess whether they are more likely to move once recovery sets in. These analyses will take into consideration changes that would have ‘triggered’ a move during the crisis but didn’t.

Dutch context

After the Great Financial Crisis, many countries experienced an economic fallout, yet its impact on household finances, house prices and mobility within and towards the owner-occupied sector varied widely (Lennartz et al., 2016; Van der Heijden et al., 2011). On the household level, labour rights and the presence of welfare state arrangements may soften or compensate a loss of work and income (see Dewilde & Ronald, 2017). In the Netherlands, unemployment benefits provide employees who lose their job up to 75% of their former income for a limited period depending on employment years.¹

Owner-occupation has been actively promoted by the Dutch state from the mid-20^th century (Wind et al., 2017). While 42.6% of all housing was owner occupied in 1985, this share had increased to 57.3% in 2012 (Aalbers et al., 2020). Like in other countries, the expansion of the owner-occupied sector has been underwritten by the promise of financial gains, economic security and social stability (Van Gent, 2010). The growth of the owner-occupied sector in the Netherlands has gone hand in hand with a similar rapid growth in mortgage debt as homeownership expansion has been supported by comparatively strong liberalisation of mortgage markets and extensive fiscal benefits (Aalbers et al., 2020; Wind et al., 2017). Dynamic housing systems, such as in the Netherlands, with lenient mortgage credit regulations have been more susceptible to falling house prices and decreasing housing transactions (Van der Heijden et al., 2011). In 2015, 81.2% of Dutch owner-occupants held a mortgage (CBS, 2018).

Prior to the crisis, it was common practice for lenders to grant loans with loan-to-value ratios of over 100% (Scanlon & Elsinga, 2014; Van Veldhuizen et al., 2020; Francke and Schilder, 2014). In addition, many interest-only loans were granted to Dutch households. These loans were particularly attractive because of generous mortgage interest tax deductions. Because of the lenient mortgage lending practices, even before the housing market crisis over 12% of owner-occupants were in negative home equity (CBS, 2015).

Mortgages in the Netherlands are typically full-recourse loans, meaning that in case of default, the lender can seize assets beyond the collateral. This contrasts to, for instance, the US, where non-recourse mortgage loans are more common. For this reason, ‘strategic defaults’ for households in negative home equity are unlikely in the Netherlands (Francke & Schilder, 2014; Van Veldhuizen et al., 2020). A part of mortgages, however, are covered by a national residential mortgage insurance (Nationale Hypotheek Garantie (NHG)) which can conditionally cover losses when households with negative home equity are forced to sell as a result of unemployment, divorce, or the loss of a spouse. Maximum transaction cost caps are set yearly by NHG and peaked at a maximum of 350.000 euros in the housing crisis period. In 2008, around one-third of all transactions were financed with NHG (SWEW, 2009).

In sum, most of owning households hold a mortgage loan, and at comparatively high loan-to-values. In the Dutch context becoming unemployed may not lead to distress right away for employees with established rights. In case of forced selling at a loss, owners of less expensive housing may under certain conditions be covered by the national mortgage insurance scheme NHG. These arrangements imply that there will be less forced mobility during a crisis than other contexts.

Data and methods

To investigate how life-course transitions affect moving behaviour during and after times of housing market downturn, we made use of longitudinal register data from the System of Social-statistical Datasets (SSD) of Statistics Netherlands. This database includes geocoded information for all individuals registered as inhabitants of the Netherlands.

For the analyses, we have drawn a 20% combined random sample of the 2012–2014 and 2014–2016 period, and selected all owner-occupants aged 25 years and older on January 1st 2012 and January 1st 2014.² For owning couples, we have randomly selected one of the partners. Owner-occupants who have been identified in both periods, have been randomly assigned to one of the two periods.

We compared these two period groups in their life-course positions and moving behaviours over a two-year period. The first period captures the final two years of the housing market downturn, and the second period captures the start of the subsequent recovery period.³ Our main dependent variable is moving house between January 1st in year t₀ and year t₂. While our analysis focusses on owner-occupants at t₀, the tenure of the moving destination also includes rental housing. When moving house, owner-occupants are most likely to move to another owner-occupied dwelling (Clark, 2016). Moves out of owner-occupation and into rental have been associated with specific disruptive life-course transitions such as job loss and separation (Helderman, 2007; Lersch & Vidal, 2014; Han et al., 2017). Moves to institutional or parental housing are excluded. The independent variables related to life course are discussed below.

Our analysis has two parts. First, we gauge how life-course status or transitions impact on residential mobility in the final years of downturn in the Dutch housing market (2012–2014) and the subsequent recovery period (2014–2016). We present the estimates of a multinomial logistic regression model that predicts the likelihood of staying, moving to owner occupancy and moving to rental housing in the two periods. The focus will be on the estimates of three life-course variables in the two periods.

Second, we are interested in whether housing market downturns can lead to delayed mobility responses. As discussed, owners may not have been willing or have not been able to move during the crisis despite a change in their household or employment situation. They may seek to move in a subsequent recovery period. Therefore, we analyse whether there may be heightened mobility during the recovery period among owners that have not moved in preceding years while having experienced life-course transitions during their immobility. Therefore, we again look at moving behaviour in the 2012–2014 and 2014–2016 periods (t₀-t₂) but here we have selected owners who had not moved in the previous two-year period (t_-2-t₀), to assess how their life-course transitions in the prior period may have contributed to moving house in the immediate subsequent two-year period. Our expectation is that the 2014–2016 group will have a higher likelihood to move after a transition in the preceding crisis period. We present the estimates of a logistic regression model that predicts the likelihood of staying and moving in the two periods. Again, the focus will be on the estimates of three life-course variables in the two periods.

Because we investigate whether moving behaviour associated with life course is contingent on the period, we follow the recommendations by Mize (2019). This means adding an interaction term between the three life-course variables and a period dummy in the models. Also, rather than presenting the models’ coefficients and significance values, our interpretation is based on the size and significance of margin estimates. More specifically, we assess whether a period difference for the three main variables exists by presenting average marginal effects (AMEs), and we also determine whether these period differences are equal between categories (‘contrast’, or ‘second difference’). The significance of AMEs and contrasts are based on Wald tests. The analyses below will present margin estimates, AMEs and contrasts for life course variables. The full models and the descriptive statistics can be found in the online supplementary files. All analyses were done using Stata 15.

Independent variables

To measure demographic life-course positions and transitions, we use two household variables: partnership and children. Both are measured between January 1st of year t₀ and year t₂. The categorical variable for an owner’s household composition consists of ‘single stable’, ‘couple stable’, ‘new couple’, ‘new single’ and ‘other’. The ‘other’ category includes all other household configurations, for example an individual living together with a couple, whether at t₀, t₂ or both. Owners who are in a couple at both t₀ and t₂ but with a different partner, as well as owners whose partner has passed away during the two-year period are also categorised as ‘other’. A separate variable classifies the household as having no child(ren), a stable number of children or having a first child, fewer children or additional child(ren) during the two-year period. For the second part of the analysis, we add the data point t_-2. Here, we use the same categorization to capture household transitions in the t_-2 to t₀ period, and use t₂ to determine whether the household has remained stable compared to the t₀ position or has undergone another change.

Employment position is determined by the primary source of household income over the two-year period. This can be stable employed, unemployed (benefits), pension, new unemployed (from employed to benefits), new employed (from benefits to employed), job change (changing employers) or ‘other’. The latter category encompasses a range of dominant income sources such as student bursaries and capital income or no income at all. Household income classes are classified on the basis of standardized household income percentiles in year one in three categories: 0–40, 40–80 and 80–100. Household (non-housing) wealth is included as the sum of all debts and wealth holdings, excluding the primary residence and mortgage debt.

Whether an individual’s household is in negative home equity is determined by the household’s loan-to-value position: the value of the mortgage divided by the value of the home according to the tax records (‘WOZ value’).⁴ Owners who have been in negative home equity (loan to value over 100%) for at least one full year at t₀ are labelled as being in negative home equity. Please note that loan-to-value ratios of over 150% are unlikely to be (solely) market-induced.⁵ To capture the home equity position for the second part of the analysis we use the same classification procedure, but for the position at the start of the preceding two-year period.

Lastly, we control for gender, migration background (migrants and their children), years since last move and educational level. These variables, as well as non-housing wealth, are based on the situation in t₀.

Results: period effects

Our first analyses compare residential mobility behaviour among owner-occupants during the final years of the housing market downturn and in the immediately following two-year period. In our sample, we see that owner-occupants mostly moved to the same tenure: 2.9% to owner occupied and 1.9% to rental between 2012 and 2014. The share of movers to owner occupied dwellings had more than doubled in the post-crisis period (4.1% in 2014–2016). The share of movers to rental housing had increased by far less (2.0% in 2014–2016). Further descriptive statistics for our sample of owner-occupants in the 2012–2014 and 2014–2016 period and the models can be found in the online supplementary files. There are slightly more cases from the 2014–2016 period (50.3%) compared to the 2012–2014 period (49.7%) as a result of a growing research population (owner-occupants 25 years and over).

Our model holds no surprises with regard to the non-life-course variables. In short, they indicate that native-Dutch, younger, wealthier and higher-educated owners were relatively more likely to move to another owner-occupied dwelling compared to staying. A move to rental housing was relatively more likely than staying for owners in old age, with lower income and lower wealth. Women were also more likely to move than men (to both tenures). Negative equity meant that owners are less likely to move to a new owner-occupied dwelling. Higher negative home equity categories are associated with further decreases. Conversely, very high LTVs were positively associated with a move to the rental sector (see online supplementary files).

Table 1 shows the predicted probabilities for the three main life course variables for each outcome (no move, move to another owner-occupied dwelling, move to rental dwelling) per period. The table indicates the average marginal effects (AMEs) for all variable categories between the two periods. Found AMEs may be a reflection of a general period effect. To assess particularity of household status and change, the ‘contrast’ column shows which AMEs are significantly different from others. So, it indicates which households transitions were affected differently by the crisis and the recovery period with respect to residential mobility. Figure 1 charts the significant AMEs with confidence intervals.

Figure 1. Average marginal effects of period (2012–2014 versus 2014–2016) on residential mobility outcomes in relation to partnership, children and employment status, based on model 2. Only significant AMEs are shown with 95% confidence intervals (N = 1,394,636).

Source: SSD, Statistics Netherlands, own calculations.

Table 1. Probability of a move by life course (partnership, children and employment) and period: marginal effects of period and differences in period effects across three household variables (N = 1,394,636).

		period	period	Period difference
	NOT MOVED	2012–2014	2014–2016	(AME of period)	Contrast
	all	0.952	0.938	−0.014***
a	single	0.966	0.958	−0.008***	b, c, d, e
b	couple	0.970	0.956	−0.014***	a, c, d
c	new couple	0.707	0.659	−0.048***	a, b, d, e
d	new single	0.664	0.631	−0.033***	a, b, c
e	other	0.721	0.700	−0.021**	a, c
a	no children	0.956	0.944	−0.012***	b, c, d
b	children (stable)	0.966	0.952	−0.014***	a, c, d
c	first child	0.874	0.844	−0.030***	a, b, e
d	extra child(ren)	0.924	0.891	−0.033***	a, b, e
e	fewer children	0.915	0.902	−0.014***	c, d
a	employed	0.953	0.936	−0.016***	b, c, d, e, f, g
b	on benefits	0.928	0.918	−0.010**	a, e, f, g
c	pensioned	0.966	0.957	−0.009***	a, e, f, g
d	unemployed > employed	0.923	0.920	−0.003	a, f
e	employed > unemployed	0.893	0.899	0.006	a, b, c, f
f	change of employer	0.929	0.905	−0.023***	a, b, c, d, e, g
g	other	0.938	0.937	−0.000	a, b, c, f
	MOVED TO	Period	Period	Period difference
	OWNER-OCCUPIED	2012–2014	2014–2016	(AME of period)	Contrast
	all	0.028	0.042	0.013***
a	single	0.015	0.022	0.007***	b, c, d, e
b	couple	0.021	0.034	0.013***	a, c, d, e
c	new couple	0.216	0.261	0.045***	a, b
d	new single	0.094	0.132	0.037***	a, b
e	other	0.159	0.193	0.034***	a, b
a	no children	0.024	0.036	0.011***	b, c, d, e
b	children (stable)	0.023	0.035	0.013***	a, c, d, e
c	first child	0.092	0.119	0.027***	a, b, e
d	extra child(ren)	0.057	0.090	0.033***	a, b, e
e	fewer children	0.039	0.056	0.017***	a, b, c, d
a	employed	0.032	0.047	0.015***	b, c, d, e, f, g
b	on benefits	0.017	0.027	0.010***	a, d, f, g
c	pensioned	0.014	0.022	0.008***	a, d, f, g
d	unemployed > employed	0.034	0.028	−0.006	a, b, c, e, f
e	employed > unemployed	0.026	0.033	0.008**	a, d, f
f	change of employer	0.044	0.065	0.021***	a, b, c, d, e, g
g	other	0.038	0.039	0.001	a, b, c, f
	MOVED TO RENTAL	Period	Period	Period difference
		2012–2014	2014–2016	(AME of period)	Contrast
	all	0.019	0.020	0.001***	n/a
a	single	0.019	0.020	0.001	e
b	couple	0.008	0.009	0.001***	e
c	new couple	0.077	0.081	0.004	e
d	new single	0.242	0.237	−0.005	–
e	other	0.120	0.107	−0.013**	a, b, c
a	no children	0.020	0.021	0.001**	e
b	children (stable)	0.012	0.013	0.002***	e
c	first child	0.034	0.037	0.003	e
d	extra child(ren)	0.019	0.020	0.000	e
e	fewer children	0.046	0.043	−0.003**	a, b, c, d
a	employed	0.016	0.017	0.001***	e, g
b	on benefits	0.055	0.055	0.000	e
c	pensioned	0.020	0.020	0.000	e
d	unemployed > employed	0.044	0.052	0.009	e, g
e	employed > unemployed	0.081	0.067	−0.014***	a, b, c, d, f, g
f	change of employer	0.028	0.030	0.002	e, g
g	other	0.025	0.023	−0.001	a, d, e, f

Notes: Contrasts column reports which period differences are significantly different across household statuses for each of the three variables (second differences) at p<.05. AMEs can differ from difference of presented period estimates due to rounding off to three decimal points. AME significance: *p<.05; **p<.01; ***p<.001.

Source: SSD, Statistics Netherlands, own calculations.

The AMEs can be read as the estimated ‘period effects’. Yet, some categories have higher probabilities for moving than others, particularly for the partnership variable. Here, stable categories (single and couple) are less likely to move than those that change. This is understandable given that for a change in partnership at least one person needs to move. To better appreciate the period differences for these categories, we also present a relative measure for the significant AMEs for the mobility outcomes. This is the AME divided by the estimate for the crisis period (2012–2014). These ‘relative AMEs’ are shown in Figure 2.

Figure 2. Relative average marginal effects of period (AME divided by estimate for period 2012–2014) on two mobility outcomes in relation to partnership, children and employment status, based on model 2. Only significant AMEs are shown (N = 1,394,636).

Source: SSD, Statistics Netherlands, own calculations.

Overall mobility

The estimates show several interesting period effects. The probabilities for immobility are clearly different between the two periods. Outside of the socio-economic ‘other’ category, only the relationship between finding work, ie, going from unemployment to employment, and moving house did not appear to be subject to period effects. For all other categories a period effect was found. As expected, immobility was more likely during the crisis period.

Moves to rental

While significant, there is only a small difference in the prediction of moving towards the rental sector between the two periods, and only a few categories show significant average marginal effects (AME). Two groups show relatively large negative AMEs. First, owners who became unemployed were more likely to move to rental housing during the crisis than those who became unemployed during recovery. Second, the relatively small ‘other’ category for the partnership variable shows a negative AME. Interestingly, contrary to expectations, moves to rental housing in relation to partnership transitions, notably divorces and separation (‘new single’), were not affected by crisis period in a similar way. Their AMEs are not significant. They were significantly affected by period in their probabilities of immobility and moving to owner occupation though. Lastly, some AMEs for the rental moves are positive. Couples, households with children, households without children, and employed owners were somewhat more likely to move to rental housing in the recovery period. These groups also show a much higher probability to move to ownership after the crisis than during (see below). The relatively moderate increase for moving to rental housing here may reflect a broader surge in mobility after a period of crisis-induced immobility for these groups.

Moves to owner-occupation

Contrary to moves towards rental housing, there are more substantial and significant period differences in moving to another owner-occupied dwelling. The AMEs are all positive and they mirror the negative AMEs for the ‘not moving’ outcome; for most categories, the estimated increase in immobility in the crisis period is similar – within half a percentage point - to the estimated increase in moves to owner occupancy in the recovery period. The few exceptions have been discussed in relation to the rental outcome. So, moving within the owner-occupied sector appears to be more period sensitive, yet in varying degrees for different households.

For the partnership variable, couples show relatively high period increases in probability for moving to another owner-occupied dwelling (see Figure 2). This stable category seems to be most affected by crisis and recovery. Our analysis indicates that often the non-owning partner moved in with a home-owning partner.⁶ This was more so the case during the crisis. When moving in together, owners were much more likely to move into a new owner-occupied dwelling together during the recovery period. In case of separation, the person or persons moving out were more likely to move to owner occupation during the recovery period. As moves to rental did not seem to be affected by period for ‘new singles’, it is unsure what happened during the crisis. It could be that couples remained living together -or at least stay registered as such- during the crisis to avoid losses or because there were no good housing alternatives. This would imply delayed mobility (see below).

All categories of the ‘children variable’ show increases in the probability of moving to an owner-occupied dwelling in the recovery years. The absolute and relative increases are highest for family households who had more children (‘extra children’). The AME for households with a first child is also high, but somewhat lower in relative terms, as mobility rates are generally higher for households during family formation. These two categories contrast with households with children (stable situation), with no children (stable situation), and households who had fewer children. Figure 2 does indicate that households with children (stable) have a relatively high AME, like stable couples. So, it seems that any households with children were more likely to move within owner occupancy during a recovery period than in a crisis period, compared to households without children and households that saw their children move out. Perhaps these households also delayed their move until after the crisis.

The estimates for the employment variable show that particularly owners in stable employment and in stable employment but with a different employer, and to a lesser extent owners on benefits and pensions, were more likely to move to another owner-occupied dwelling in the recovery period. We do not find significant differences for formerly unemployed owners who found work or for the ‘other’ group. Owners who lost employment (see also above), were more likely to move to another owner-occupied dwelling, but less so in relative terms (Figure 2).

Results: delayed mobility

On the whole, fewer owners moved house during the downturn period. This could indicate that many households may have delayed their moves in times of crisis. Particularly, households that had experienced life-course changes may have been seeking to adjust their housing situation during recovery. Indeed, Table 2 confirms the sharp increase in mobility after the crisis. Yet, the share for moving was not that different for owners who had not moved in the preceding two-year period. This group consisted of both a large segment of established owners who may have not had the desire to move and a share of owners who may have been unable to move despite a wish to change their living arrangements (ie, moving in together or separate), or despite changes in household size or employment status.

Table 2. Mobility of owners in two periods (2012–2014 and 2014–2016).

	Share moved 2012–2014	Share moved 2014–2016	Relative growth of moving owners	Growth of group size
All Owners	4.6%	6.0%	29.0%	0.3%
Owners who did not move in preceding two years	4.5%	5.8%	30.3%	1.3%

Source: SSD, Statistics Netherlands, own calculations.

To further understand period effects on mobility and how ‘delayed mobility’ may help to explain the sharp increase in mobility after the downturn, this section looks at which life-course transitions that did not ‘trigger’ a move in a two-year crisis period, are relevant explanatory variables to predict residential moves in the subsequent two-year crisis or recovery period. The following analyses focus on owner-occupants who did not move during the two years leading up to the 2012–2014 and 2014–2016 periods but may have experienced life-course changes that are generally associated with mobility. If households experiencing life-course transitions during the years of downturn had been waiting out the crisis to adjust their housing situation to their household situation, we would expect a positive period effect in the recovery period.

As we are interested to see how events in the preceding two-year period may affect moving behaviour in the subsequent period, we include a similar set of life-course variables as in the analyses above, yet they cover the period between t_-2 and t₀. In addition, we add a stable or changed distinction for each to reflect any changes between t₀ and t₂. The models include the same background variables as above at t₀, and a dummy for whether the household had been in negative home equity at t_-2.

Table 3 presents the predicted margins, AMEs and contrasts for the life-course variables. Singles and couples who remain stable in the four-year period show large period differences which are comparable to the predictions of the first model above. The similarity makes sense as most owners do not move twice in a period of four years. The probabilities show a small period difference for couples that stayed together but then separated (couples > change). So, we find some evidence that couples were forced to or chose to stay together, or even just stay registered together, as long the downturn was lasting. Similarly, moving in together after living alone in t_-2-t₀ (single > change) shows a relatively small increase compared to overall increases in mobility.

Table 3. Probability of a move in t₀-t₂ by life course (partnership, children and employment) in t_-2-t₀ and t₀-t₂ and period of owners who did not move in the preceding two years: marginal effects of period and differences in period effects across three household variables (N = 1,281,961).

			t0-t2 period	t0-t2 period	Period difference
	MOVED		2012–2014	2014–2016	(AME of period)	Contrast
	all		0,047	0,061	0,014***
	t-2 - t0	t0 - t2
a	single	stable	0,031	0,038	0,007***	all others
b		change in hh	0,328	0,379	0,051***	a, c, d, g, i
c	couple	stable	0,029	0,042	0,014***	a, b, d, e, f, h
d		change in hh	0,356	0,388	0,031***	a, b, c, g, i
e	new couple	stable	0,110	0,147	0,036***	a, c, g, i
f		change in hh	0,135	0,183	0,048**	a, c, g
g	new single	stable	0,068	0,085	0,017***	a, b, d, e, f, h
h		change in hh	0,260	0,313	0,054***	a, c, g, i
i	other	–	0,054	0,073	0,020***	a, b, e, f, h
	t-2 - t0	t0 - t2
a	no children	stable	0,045	0,056	0,012***	b, d, e, f, g, h
b		change in child status	0,107	0,136	0,029***	a, c, d, e, f, i, j
c	children (stable)	stable	0,036	0,048	0,013***	b, e, f, g, h
d		change in child status	0,061	0,074	0,013***	b, e, f, g, h
e	first child	stable	0,088	0,128	0,041***	all others
f		change in child status	0,090	0,131	0,042***	all others
g	additional child(ren)	stable	0,059	0,091	0,032***	a, c, d, e, f, h, i, j
h		change in child status	0,081	0,110	0,029***	a, c, d, e, f, h, i, j
i	fewer children	stable	0,039	0,051	0,012***	b, e, f, g, h, j
j		change in child status	0,052	0,066	0,014***	b, e, f, g, h, j
	t-2 - t0	t0 - t2
a	employed	stable	0,046	0,063	0,016***	b, e, f, j, k, m
b		change in status	0,068	0,074	0,006***	a, k, l
c	on benefits	stable	0,045	0,056	0,010***	f, k
d		change in status	0,052	0,061	0,009***	k
e	pensioned	stable	0,034	0,042	0,008***	a, f, k, l
f		change in status	0,087	0,082	−0,005***	a, c, e, g, i, k, l, m
g	unemployed > employed	stable	0,049	0,063	0,014***	f
h		change in status	0,073	0,091	0,018***	none
i	employed > unemployed	stable	0,072	0,087	0,015***	f
j		change in status	0,064	0,062	−0,001***	a, k, l
k	change of employer	stable	0,068	0,095	0,026***	a, b, c, d, e, f, j, l, m
l		change in status	0,080	0,095	0,015***	b, e, f, k, l
m	other	–	0,043	0,054	0,011***	a, f, k

Notes: Contrasts column reports which period differences are significantly different across household statuses for each of the three variables (second differences) at p<.05. AMEs can differ from difference of presented period estimates due to rounding off to three decimal points. AME significance: *p<.05; **p<.01; ***p<.001.

Source: SSD, Statistics Netherlands, own calculations.

In contrast, the period differences for partnership are most pronounced for the ‘new couple’ category (new couple > stable). Here, we also see some evidence for ‘delayed mobility’. Instead of moving into a new dwelling, these owners stayed in their dwelling while their partner moved in. These new-coupled owners were more likely to adapt their housing situation in a subsequent post-crisis period than in the crisis period. The estimates and contrasts indicate that these period effects were greater both in absolute and relative terms.

There are large period effects for owners who had their first child in the t_-2-t₀ period with regards to mobility in the subsequent two-year period. Also, owners who had additional children in the preceding two-year period are more likely to move in the post-crisis period. These AMEs differ in relative and absolute terms, suggesting crisis-period induced delays.

Employed owners (who remain employed in both periods) show a relatively large period difference. This is difference is even higher for those who remained in work and changed employers in the t_-2-t₀ period. A change of employers may have signified a change of income or location of work, yet an adjusting move is not necessarily made in times of crisis. We find no notable period differences in mobility for owners who had gone in or out of unemployment in the preceding two years. The effect of employment loss and gain on residential mobility during a downturn period may be more direct - we have found evidence for an association between loss of work and moving to rental during the crisis above-, or could also work over a longer period of time.

Conclusion

Our study sought to illuminate how the relationship between life-course and mobility of owner occupants may be affected by periods of downturn and recovery. The ‘period effect’ has received comparably little attention (Coulter et al., 2016). In general, a period of downturn constrains owners’ residential mobility. Yet, difference between periods varies for different household and employment statuses and changes. Our analysis showed a strong period effect between the final years of the housing market downturn and the subsequent recovery period. Mobility within the owner-occupied sector was substantially lower during the crisis period and higher in the recovery period. The crisis period seemed to have particularly affected the behaviour of working family households: stable couples, households with (additional) children, and employed owners, including individuals who had changed jobs. The finding that these households were affected in their mobility, implies that their desire to move has been delayed due to economic uncertainty and loss aversion (see Engelhardt, 2003; Marsh & Gibb, 2011; Steegmans & Hassink, 2018), or perhaps that opportunities offered in the recovery period inspired them to move house. In any case, further research into the period effect on moving desires is necessary to fully explain this outcome (see Coulter, 2013; Coulter et al., 2016; Waldron & Redmond, 2017). These period effects were mostly found for moves to owner-occupied dwellings. The relationship between life course and moving into rental was only moderately affected by period. Only owners who lost their job were clearly showing a higher probability of moving into rental during a crisis period. Interestingly, separation and divorce, changes that typically ‘trigger’ an urgent move do not show any period differences in moves towards rental housing. Regardless of period though, the Dutch rental sector has been the main alternative for people leaving their partner.

The finding that stable households were more affected than owners who experienced a change in partnership or children confirms that residential mobility is a highly demographic process (Clark & Huang, 2003). Yet, households undergoing changes in their demographic or employment situation were still affected by crisis period; they were also less likely to move. The ‘delayed mobility’ analyses show that owners who formed a new couple (ie, became registered in the same household) were more likely to stay put during the crisis period and move once recovery set in, or move in with one partner during the crisis and move again after. It also showed some evidence for a delay for divorce and separation. Couples were somewhat more likely to separate after the crisis. Although it could be that partners were already living apart but were still registered as a household. Our analyses also indicated that other life-course changes were also more likely to lead to a delayed post-crisis move compared to during the crisis. These include owners who saw a change in number of children during the downturn or who changed jobs. So, when these life-course events did not ‘trigger’ a move during the crisis, they tended to do so after.

In sum, the relationship between life course and residential mobility is contingent on period, at least for most owner-occupants that move within the sector. While the Dutch context may be more conducive to immobility, the finding that the relationship varied between periods and across various household categories implies that such the period of study should be accounted for, or at least reflected upon, in life course and residential mobility research.

Our findings also problematize the relationship between mortgaged ownership and life-course related mobility. While our predictions do not show dramatic increases in (forced and voluntary) moves out of ownership, life-course-related mobility is shown to be affected by crisis period. While social costs of housing market crisis are usually understood through statistics on defaults and forced mobility, many households weather it out. Among these, there may be many ‘unrevealed casualties’: owners who may no longer find their housing situation suitable, who want to start a new life with someone, or who want to urgently end their relationship (see Feijten & Van Ham, 2007), but continue to make monthly payments at a cost to their household budgets, quality of life, and their emotional and psychological wellbeing (see Waldron, 2016; Dwyer et al., 2016). These owners are effectively leading ‘mortgaged life courses’ (cf. García-Lamarca & Kaika, 2016). Our study provides incentives for further qualitative or biographical studies and more focused quantitative analyses to understand how financialised housing systems are structuring our lives (eg, Izuhara, 2015).

We have shed a new light on the dynamics between period, housing, and life course. Yet, our study has several caveats that may provide an impetus for further research. First, our assertions are based on owner-occupants. We did not compare owners to tenants, and living in rental housing will likely have its own life course and mobility dynamics in times of crisis. A comparison, or a similar set-up as here that takes into account rent levels and arrears may further illuminate how housing tenure can structure lives during times of crisis.

Second, we were able to reliably compare high-quality register data for all years between 2012 and 2016, with some data from 2010. As the downturn already started in 2008, we may have missed the effects of transitions in the early phase of the crisis. In addition, future research may gauge the timing between transitions and moves over longer periods to establish the full extent and durability of crisis effects and postponements (see Falkingham et al., 2016).

Third, our study focused on the Netherlands. This case is particularly interesting because of its highly financialized owner-occupied sector and strong recourse to the borrower. Mortgages debts are relatively high, meaning that owners are more sensitive to go into negative home equity by decreasing house prices, particularly for recent entrants. Yet, our outcomes were likely skewed towards immobility rather than (forced) mobility. Similar analyses in countries where welfare arrangements, mortgage insurance and labour market regulation offer less protection and compensation in times of crisis may show different results. Generally, (changes in) lending practices and mortgage market regulations may also play an important role in the degree to which financial considerations impact mobility. Likewise, the availability of alternative tenures may also factor into mobility during a crisis. Also, our findings point to increased mobility after the downturn, but this is also not a given everywhere as low construction rates during the crisis may depress opportunities (see Dwyer et al., 2016). In sum, housing market fluctuations matter for how, where and when households match their housing to their (changing) needs.

Acknowledgements

We would like to thank Sako Musterd, Cody Hochstenbach, and Amber Howard as well as the anonymous referees for their valuable comments and suggestions. The usual disclaimers apply.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

The position of Rik Damhuis was co-financed by Corpovenista (a research cooperative programme of Dutch housing associations), Woonbedrijf Eindhoven and the municipalities of Amsterdam and Almere.

Notes on contributors

Rik Damhuis

Rik Damhuis worked as a Researcher at the Department of Geography, Planning and International Development Studies at the University of Amsterdam. He is currently employed in the private sector.

Wouter van Gent

Wouter van Gent is a Senior Lecturer at the Department of Geography, Planning and International Development Studies at the University of Amsterdam.

Notes

1 Before 2016, the maximum was 38 months.

2 All analyses have consistent data sources and definitions for each year (January 1^st 2010, 2012, 2014, 2016). To establish owner occupancy, we have used data based on population registers (current residence) and dwelling ownership.

3 On a quarterly basis, pre-crisis house prices peak in Q3 2008. The lowest quarterly price index readings are Q2 and Q4 2013. From Q1 2014 onwards house price start to rise again consistently (Kadaster 2021).

4 Our data accounts for premiums paid into the tax-exempt savings accounts and insurance policies issued jointly with mortgages (‘(bank)spaar- en beleggingshypotheken’).

5 Some owners may be overleveraged, but their high mobility indicates that high ratios are also due to timing lags in ownership and loan registration surrounding a move. Also, some households may hold additional loans for a bought dwelling that is under construction (and therefore missing in the housing value dataset).

6 The estimates can go below 50% as partners can move in from outside of owner-occupation. Only home-owners at t0 are part of our dataset.