House Prices and Fertility in England

Transcription

1 House Prices and Fertility in England Cevat Giray Aksoy Preliminary Draft: October, 2015 Abstract This paper explores the effects of house prices on fertility using a new instrumental variable strategy, exploiting exogenous variation in house prices induced by planning restrictions. I find a sizable positive impact of the recent housing cycle on births among home owners, consistent with existing literature, but negative responses among renters. Using data from English counties between 1995 to 2007, the results produced by this study indicate that a 10,000 increase in house prices leads to a 3.8 percent increase in births among owners and a 4.4 percent decrease in births among renters. Birth rates react more strongly to changes in house prices in supply constrained counties. A further assessment of house prices and fertility nexus reveals that these effects vary by demographic subgroups. Keywords: house prices, fertility, home equity Jel Classification: J13, D10 University of London, Royal Holloway cevat.aksoy.2012@live.rhul.ac.uk 1

2 1 Introduction In 2007, the nominal average house prices calculated by the Office for National Statistics rose for 12 consecutive years. More specifically, between 1995 and 2007, average house prices have increased nearly 160% in real terms, and they have nearly tripled in nominal terms. Meanwhile, these housing market trends were accompanied by an increase in births. Between 1995 and 2007, the total fertility rate increased to a 1.88 children per woman, its highest level since 1973 and up from a 1.71 in Rising trends in fertility rates may be partly explained by the changes in the housing market. In other words, to the extent that housing tenure play a role in determining a number of children to have and whether to coincide births with home ownership, a rise in the house prices such as that observed during the last property cycle may be said to affect the fertility rates. Despite the logical appeal of this argument, to date there have been a few studies that set out to present an analysis of this relationship. Yi and Zhang (2010) extend a standard Beckerian model of fertility behavior to formulate the effect of house prices on fertility. Lovenheim and Mumford (2013) and Dettling and Kearney (2014) also offer two distinctive frameworks of analysis for the United States, the former conducted at the individual level, the latter at area level. Two stylized facts have emerged from these studies: 1) an increase in the house prices lead to an increase in births among home owners; and 2) the net effect of the house price changes on fertility is contingent upon country specific characteristics. Notwithstanding the fact that house prices have a significant impact on birth rates because of it is relative importance for the household budget constraint, a systematic understanding of how tightly regulated housing market and associated housing affordability contribute to fertility outcome is still lacking. In this paper I attempt to address both of these gaps in the literature. For various reasons, I suspect that the existing studies, briefly introduced above, 2

3 underestimate the effect of house prices on fertility. Since house prices for dwelling units are determined by a number of factors, the failure to control for demand shifters that play a part in affecting house prices may downwardly bias estimates of the house price-fertility nexus. For instance, under a plausible set of assumptions, a rise in the total number of households tend to bid up the house prices. This may be due to a number of causes; an increase in average life expectancy, an increase in net migration and changes in the family formation (i.e. cohabitation, marriage, separation rates). In a similar vein, a fall in average household size, the distance between home and work and the structure of social housing may exert price pressures on housing. Moreover, rapidly falling affordability of owner-occupied housing makes it difficult for non-homeowners to get onto the property ladder, thus causing longer term tenancy duration and large proportion of income spent on rent. In order to establish a causal relationship and be able to assess the effect of house price on the fertility, I deem necessary taking all these factors into account. A further issue associated with interpreting the estimates is the direction of causation. For example, assuming that the housing supply is fixed in the short term, if people who plan to have a child and to expand their families demand for larger houses, then the direction of causation run from birth rates to house prices. Moreover, the relationship could be driven by other unobserved county-specific variables. Hence, in addition to problems associated with reverse causality, previous inferences may also be flawed owing to omitted variables bias. To be more precise, omitted variables downwardly biases OLS estimates of the causal effect of house prices on fertility. In this paper, I estimate the effect of house prices on fertility using a county-level panel spanning from 1995 to I first present OLS estimates of how fertility responds to the increases in house prices separately for homeowners and renters. To 3

4 address omitted variables bias: (1) I control for observable basic and extensive demographic characteristics, and (2) I take full advantage of the panel data structure by including county, year and age group fixed effects as well as county, ownership and group specific linear time trends. The OLS estimation establishes a strong positive relationship between house prices and fertility among homeowners and a negative relationship among renters. The association between house prices and birth rates is stronger in supply constrained areas. Taken altogether, these findings suggest that house prices affected patterns of fertility over the real estate cycle, during which England has experienced a small scale baby boom and a rise in the mean age at first birth. 1 The findings continue holding when controlling for unemployment rate, gross weekly household labor income, fixed effects, geographic characteristics as well as main and extended demographic characteristics. Furthermore, to establish the causal effect of house prices, I introduce an identification strategy that exploits exogenous variation in house prices generated by local planning authorities. I use regulatory restriction measures as an instrumental variable (IV) for house prices. In particular, I calculate three-year moving average refusal rates for major development projects to quantify the planning restrictiveness which exhibit substantial variation across counties. While the coefficients are larger in magnitude, the IV estimates consistently confirm the OLS findings. The results presented in this paper complement recent studies that document the effects of house prices on fertility. To be more specific, the contribution of this paper to the literature is four-fold. Firstly, I provide causal evidence on the impact of house prices on fertility from a country with a highly restricted housing market, which has not been studied previously. These findings provide new evidence on how country-specific 1 According to the ONS the standardized mean age -takes into account the changing distribution of the population by age and other factors over time- of mothers increased to 27.9 years in 2011, compared with 26.1 years in

5 factors have influenced fertility outcomes by showing the importance of housing market regulations. Secondly, I deal with the endogeneity of house prices in a robust manner and carefully disentangle the impact of house prices from other factors, namely, (i) share of social housing, and, (ii) physical supply constraints (share of developable land and topography-elevation data). Thirdly, while highlighting the long-run impact of restrictions through its influence on fertility patterns, my research goes beyond the role of house prices to uncover the effects of housing affordability. Lastly, I provide an extensive quantitative interpretation of the results. In this suggested framework, fertility rates are predicted in counterfactual scenarios in which house prices are hypothetically set at different levels. This broader focus on the real estate market provides a distinct addition to the literature and this research will further our knowledge about an increasingly important and understudied socio-economic problem. The remaining part of the paper proceeds as follows: The next section discusses the conceptual framework and related literature. The third section introduces the empirical strategy and data description. The fourth section presents the results and the fifth section concludes. 2 Conceptual Framework and Related Literature Ever since Malthus (1798) discussion of An Essay on the Principle of Population, researchers have focused on the observed differences in birth rates across and within countries. These differences are mostly explained by changes in household wealth through variation in labor market conditions and in social security benefits. While there are a number of other factors that may affect fertility outcome, the movements in house prices offer an appropriate measure to examine the effects of household wealth changes on fertility. The main advantage of this measure is that changes in house 5

6 prices disqualify the opportunity cost of time argument and minimize the biases associated with cross-sectional data (Lovenheim and Mumford, 2013). To date, a number of studies have paid particular attention to the effects of labor market outcomes on fertility (Kravdal 2002, Adsera 2005, Luedicke et al and Schaller 2012). Financial markets and social security have also attracted considerable interest (Cigno and Rosati 1992, Van Groezen et al. 2003, Boldrin et al. 2005). 2 Housing market, in contrast, remains mostly uncharted. Until recently, the housing market has never been explicitly included into the analysis of the life cycle fertility models. Becker (1960) models couples as utility-maximizing agents deciding on the number of children and on child-related expenditures, yet did not formally consider house prices in his theoretical framework. The potential effect of housing market has also been ignored in the quality and quantity approach (Becker 1960; Becker and Lewis 1973; Willis 1973). It implies that an increase in income might have depressing effect on fertility conditional on income elasticity for the number of children is substantially less than that for quality of children. The timing of fertility approach (Mincer 1963; Becker 1960) attributes the negative relationship between income and fertility to the higher cost of parental time experienced by higher income families, either because of increased market wage rates or because higher household income raises the value of parental time in non-market activities. In contrast to the implications of the life-cycle fertility models, Lovenheim and Mumford (2013) and Dettling and Kearney (2014) show that house price is an important determinant of the fertility outcome. More specifically, Lovenheim and Mumford (2013) find that increase in housing wealth among homeowners leads to an increase in the probability of having a child. For renters, however, there is no evidence of an effect of rising house prices on the fertility. Dettling and Kearney (2014) also find the 2 See Sobotka, Vegard, and Philipov (2011) for a review of the earlier literature. 6

7 positive impact of the house price increases on fertility among homeowners and the opposite is true for renters. They explain their results in terms of an income effect : under the assumption that children are normal goods, an increase in household wealth (through increases in house prices) generates income effect leading to an increase in the number of births. On the other hand, increase in house prices may negatively effect the household wealth of renters and it is expected to generate price effects leading to a fall in birth rates. This conclusion is based mainly on the evidence of Lino (2007) who show that housing cost constitutes the greatest portion of the annual cost of raising a child. These findings reveal the importance of the real estate market in the decision of transition to parenthood or having another child. However, a further analysis in England is necessary. It is mainly because: (1) House values in England overvalued, prices amongst the highest in the world, (2) Prices exhibit extreme volatility, real house prices in England as a whole are substantially more volatile than in the most volatile metro areas in the United States (Hilber and Vermeulen, 2014), (3) In terms of size, houses are substantially smaller than the United States as well as than the continental European countries, (4) house prices have been risen faster than than the Eurozone countries. All in all, a special attention should be devoted to England as it exhibits different housing market characteristics than other countries. 3 Data and Empirical Strategy The main empirical methodology of this paper is to use a county-level panel data to test for an association between county level fertility rates and county level house prices. The balanced-panel covers the period for the 131 counties and two age groups. 3 Given that the main empirical tests rely on the aggregate (county level) relationship 3 note on counties. 7

8 between fertility rates and house prices, isolating the effect due to a behavioral response to the house prices (that is, sorting into counties with rising or falling house prices) need careful consideration of other factors that vary systematically with property cycle and that affect birth rates. The main factors influencing the relationship between the house prices and fertility can be divided into seven components: (1) location with better labor market opportunities, (2) household income, (3) changes in family formation and household size (separation rate, marriage rate, lone-mothers, one person household), (4) net inward migration, (5) social housing market, (6) cost of borrowing and credit markets, (7) taxation and transaction fees. The first two factors have a key impact both on housing tenure and fertility outcome, a fact widely discussed in both branches of literature. The factors listed in the latter five categories are also likely to effect the empirical relationship. The changes in family formation and an increase in net inward migration mainly increase the demand for additional housing units and tend to bid up house prices. They can also potentially affect the birth rates. In areas with large social housing market house prices/rents may be lower as potential buyers/tenants have relatively cheaper options. This is especially true for credit-constrained households. On the borrowing side, if interest rates remain low, for a given levels of income and house prices, fall in true cost of borrowing leads to an increase in housing demand. Furthermore, the credit market liberalization and new securitization methods (i.e. fall in default risk for lenders) result in greater credit availability. Muellbauer and Murphy (2008) confirm this view and show that financial deregulation has a positive demand-shift effect on house prices. Lastly, high transaction fees and taxation may encourage renting if these costs represent a significant portion of the total cost of home ownership. 4 4 In England, stamp duty serves a transaction tax for housing. It is paid by the buyer and levied progressively, ranging between 1% and 4%. 8

9 Failure to take these factors into account may cause bias in the estimated coefficients. For instance, assuming that single person household is positively associated with house prices and negatively associated with birth rates, oversight of this causal element from the estimation would downwardly bias the coefficients. In a similar manner, pro-cyclical movements in lending behavior could induce upward bias in the point estimates. To address such omitted variables issue, I employ extensive control variables for county level observable characteristics and take advantage of the panel nature of my data to account for unobservable factors. Similar to Dettling and Kearney (2014) the fully saturated model specification that I estimate is given by the equation: log(birth cgt ) = β 0 +β 1 HP c(t 1) +β 2 HP c(t 1) Own cg(t 1) +β 3 Own cg(t 1) +γx cg(t 1) + θ c + φ t + α g + η c(t 1) + η c(t 1) Own + η c(t 1) Age + ε cgt where c, g, and t index counties, age groups and years. Age groups defined by two bands and The data set consists of a balanced panel of 131 counties, two age groups, each observed over a period of 13 years (1995 to 2007). Birth cgt is the log of age specific fertility rates for a given county, age group and year. HP c(t-1) is the house price index and shows how an increase in house prices affects the relationship between house prices and fertility among renters. HP c(t-1) *Own cg(t-1), house price index is interacted with a baseline measure of ownership rates, and indicates how an increase in house prices affects the fertility among home owners. I employ the Land Registry s house price index for the purpose of this study. The main advantage of this index is that it is based on actual housing market transactions -whether these are made with a mortgage or with cash- and it disentangles changes in house prices from changes in the type or quality of properties being sold. Essentially, the index uses repeat sales method which takes the differences in prices between the first time sale and the last time sale of a property and averages these changes in prices 9

10 to produce the house price index. To address depreciation or appreciation of properties over time, the index also allows some changes in prices being due to random variation. Although the land registry index incorporates the largest sample, it has a shortcoming as it excludes all new properties. 5 Despite that, it is still a more suitable measure for the aim of this paper compared to the other indices which are based on hedonic regressions. Because: 1-) the repeat sales method minimizes the concerns about unobserved heterogeneity in house prices via following same property over time, 2-) it covers all cash and mortgage transactions, 3-) since it is based on actual property transactions, it eliminates bias that arises from mortgage valuation approvals. Because some houses that receive a loan offer appear in mortgage based data even if they are not sold (Chandler and Disney, 2014). I construct mean age-group home ownership rates at the county level. It is based on the UK Labor Force Survey (LFS) and in order to minimize endogeneity concerns I use the baseline measure of the home ownership rates. To be specific, it is a proportion of home owners in each age group-county cell in X cg(t-1) is a vector of control variables and it includes: (1) labor market controls, (2) main and extended demographic controls and (3) additional housing market controls. First, to account for pro-cyclical variation in labor market outcomes, I include unemployment rate and the average gross household weekly income. 6 Second, to adjust for the effect of demographic structure on fertility rates I directly control for time varying demographic characteristics. Each variable (level of education, ethnicity, country of birth, marital status, country of residence 12 months ago, the pres- 5 ONS, Nationwide and Halifax incorporates smaller samples that are only based on mortgage data, i.e. excludes cash transactions. 6 To be clear, unemployment rate is constructed based on the International Labour Organization (ILO) standard definition of basic economic activity. Gross household income is based on labour earnings and calculated using gross weekly pay of heads and spouses in their main job and second job. I then averaged this over each county-age group-year cell. 10

11 ence of children -any child <5 & any child >5-, working for the same firm 12 months ago, average travel to work time) calculated as a proportion of individuals in age groupcounty-year cell. While I would like to have a direct measure for net inward migration, these data are unavailable. Hence, I use country of birth and country of residence 12 months ago to proxy for variation in immigration. Moreover, given that proximity to work, travel-to-work expenses and access to amenities may empirically link the house prices and fertility, I also control for the share of individuals who were working for the same firm 12 months ago and average travel to work time (less than 20 minutes, minutes, more than 60 minutes). Third, I include the share of developable land in counties in all specifications. Scarcity of open land for future development is likely to exert price pressures on property market. If share of developable land leads to higher house prices via imposing long-term supply constraint, omitting this variable from estimating equation would upwardly bias the coefficients of primary interest. In addition, I also control for length of residency to account for premia or discount related factors of housing tenure. Finally, county fixed effect, θ c, and age group fixed effect, α g, minimize all variation in birth rates caused by factors that vary across counties and age groups and are constant over time. The year fixed effect, φ t, eliminates the time variant macroeconomic shocks that lead changes in fertility rates shared by all counties over time. For example, given that total house demand is largely credit rationed, relatively low and stable inflation rate builds up housing demand or, alternatively, low nominal and real interest rates significantly affect the cost of borrowing and make it more attractive to invest in housing. Because of the possibility that people likely to sort into counties over time for different reasons, I also control for county-specific linear time trends in fertility, η c(t 1). In more highly parameterized specifications, I allow linear trends in birth 11

12 rates to vary across age groups or ownership rates. Because some of changes in house price-fertility nexus related to other factors (i.e. the effect of school quality, increasing/decreasing house prices) this strategy should isolate the effects from any sorting behavior and be able to generate conservative estimates of the effects of house prices on fertility. However, there is a potential reverse causality issue inherent in the above estimation as well as omitted variables bias. Such endogeneity may arise for instance if people who plan to have a child demand for larger houses. This may eventually lead to the direction of causation run from birth rates to house prices. Birth rates and house prices may also be jointly affected by unobserved factors. Furthermore, preferred house price index may cause a attenuation bias due to measurement error. An alternative strategy that addresses these issues would be to use an instrumental variable that determine county level house prices yet are unrelated to fertility rates. To establish a casual interpretation, I first wanted to determine a variable that would capture the restrictiveness of the regulatory system and, hence, I employ the refusal rates for major residential projects as an instrumental variable. The identification assumption is straightforward: conditional on locations are not perfect substitutes for each other and households are not perfectly mobile, an increase in housing demand should lead to a higher house prices in counties where local planning authorities are more restrictive. The refusal rates defined as the proportion of housing projects with 10 or more dwellings that was rejected by an local planning authority. Previous research has shown that restrictiveness of the regulatory system affect the house prices to a greater extent during boom periods (insert the reference!!!). Moreover, price pressure on housing market is often attributed to the England s peculiar, inflexible and outdated planning system, including the restrictions on land supply. 7 Hilbert 7 The UK s current land planning system has its origins in the 1947, Town and Country Planning Act. 12

13 and Vermeulen (2010) show that regulatory constraints imposed by the planning authorities mainly explain the high house prices in England, especially in the southern regions. Indeed, in most counties house prices are mainly driven by two factors: 1-) to a greater extent planning constraints, 2-) to a lesser extent physical constraints. In other words, these constraints caused a persistent mismatch between housing units desired and housing units supplied. 8 This well-documented fact also bolsters my confidence in assigning the refusal rates as an appropriate instrumental variable. One potential concern is that refusal rates may behave pro-cyclically during the housing boom period. That is, more profitable economic environment may encourage developers to apply for higher number of projects, leading to higher refusal rates in some counties. In this case, my instrument would be correlated with the second-stage residuals. Although including year fixed effects in the estimating equation should address this issue, I further investigate the validity of my instrument by: 1-) using the baseline average refusal rate in 1994, 2-) using 3- year moving average refusal rate between 1982 and It is worth noting that the results of these alternative measures were similar to the presented IV results. Moreover, Kleibergen-Paap F-statistics suggest that both measures reveal sufficiently strong identification and neither send any warning signals of endogeneity. Data on this project are compiled from various sources. Number of births by area of usual residence of mother are complied from Birth Statistics Data of the Office for National Statistics. The Age Specific Fertility Rates (20-29, 30-44) are constructed by dividing the number of births by the corresponding female population in given a county-age group and year in which annual data for female population are from the censuses. Main and extended demographic controls, unemployment and household 8 According to European Union Housing Statistics report, the ratio of vacant dwellings to total dwellings stock was well below the EU average, around 4% in England and 11% in EU. 13

14 labor income are drawn from the UK Labor Force Survey. The housing market data are gathered from various sources, including Department for Communities and Local Government and the Land Registry. I dropped some counties because of missing data and for the remaining counties I compiled all the data based on the 2001 local planning authority boundaries. 9 Table 1 and Table 2 present means and standard deviations for control variables. The descriptive statistics for age band are displayed in first two columns and for age band are displayed in last two columns. The descriptive statistics for aggregate level housing market variables are displayed in Table 3. As illustrated in Table 1 and Table 2, age groups differ from each other in a number of characteristics. For instance, the fertility rate is higher for the age band 20-29, around 93 births per 1000 women. The same cohort also earn less on average, more likely to be unemployed and renter, less likely to be non-uk Born, highly educated, married, reside in a same property and belong to a ethnic/racial minority. Age band is on average earn more, less likely to be unemployed, renter, non- UK Born and belong to a ethnic/racial minority, more likely to be highly educated, married and to reside in a same property. On the other hand, with regards to the partnership and presence of children characteristics, there are no differences between the two age groups. 10 In terms of housing market, a first look at the data show that house prices vary greatly across counties. Over the sample period, mean house price was 172,001 with a standard deviation of 103,024. Taking affordability into account and averaged over the sample period, house prices were substantially higher and were more volatile in the South East than the rest of the country. I obtained the refusal rate of major residential 9 Notes on dropped counties. 10 Age group-specific housing tenure measures are time invariant and held constant in 1996 values. 14

15 projects data from the Department for Communities and Local Government. Appendix Figure A1 illustrates the average refusal rate by county, averaged between 1995 and Refusal rates over the sample period were clearly highest in the South East and lowest in the North of England. 4 Results 4.1 Ordinary Least Squares Estimation Table 4 presents the results from the OLS estimation where the dependent variable is the log of the age-specific fertility rate. Column 1 reports the estimation with all fixed effects included (county, year and age group); column 2 adds basic demographic characteristics (education, ethnicity, country of birth and partnership status) as well as unemployment and wage measures; column 3 adds extended demographic characteristics (length of residency, country of residence 12 months ago, the presence of children, working for the same firm 12 months ago, travel to work time, share of social housing); column 4 adds county-specific linear time trends; column 5 adds ownership-county specific linear time trends and column 6 adds demographic group-county specific linear time trends. The HousePrice*OwnershipRate coefficient in the first column yields a positive and statistically significant estimate of.0322 and the House Price coefficient yields a negative and statistically significant estimate of After adding more controls and time trends, the results consistently and significantly show that for home owners an increase in house prices is positively associated with fertility and for renters an increase house prices is negatively associated with fertility. The magnitude of the relationship indicates that a 10,000 increase in the house prices cause a 3.22 percent rise in the 15

16 fertility rates among home owners. For renters, the same amount of increase in the house prices cause a 2.1 percent fall in the fertility rate. Inclusion of county-specific trends increases the coefficient both for homeowners and for renters. This implies that the county-specific fertility trends driven by the omitted fertility related factors tend to move in the opposite direction of the trends in house prices over the sample period. To assess whether different demographic groups are more likely to move in response to change in house prices, I include ownership-county specific linear time trends and age group-county specific linear time trends in columns 5 and 6. The coefficients of the main interest remain similar to the ones in columns 1 to 4, suggesting that demographic groups do not systematically sort into different counties owing to other factors (i.e. school quality, upward/downward trending house prices). According to the estimate from the fully saturated specification in column 6, a 10,000 increase in the house prices is associated with a 2.52 percent increase in birth rates among home-owners and with a 1.24 percent decrease in birth rates among renters. At mean ownership rate a 10,000 increase in the house prices is associated with 1.37 increase in birth rates in England. This finding suggests that fertility moves pro-cyclically and the positive effect of house prices on fertility is considerably larger than those found for the U.S. (Dettling and Kearney, 2014). The main reasons for this finding are two-fold: On the ownership side, house prices almost tripled between 1995 and 2007, leading to a substantial increase in equity extraction. On the renting side, unlike the U.S., the rental cost also followed the increases in house prices. 11 Consequently, a rise in home-owners wealth and renters living cost have led to an increase in birth rates at aggregate level. In order to mitigate biases may arise from omitted-variables and sorting patterns, the rest of the analysis is carried out based on the specification in column 6, in which I control for observable demographic and geo- 11 Please see appendix figures A3 and A5. 16

17 graphical variables, labor market variables and exploit the panel aspects of the data by aforementioned fixed effects and linear time trends. The analysis continues with a stratification of the regressions with county level housing supply characteristics so as to gain further insight into the housing market basis of this result. The first two columns of Table 6 suggests that the fertility is the most responsive to house price changes in more supply constrained counties where ownership rate above the 75th percent. In these counties, a 10,000 increase in the house prices leads to a 4.03 percent increase in fertility among home owners and 2.64 percent decrease in fertility among renters. It is because supply shortage leads to higher house prices, generating higher equity for home owners and higher cost for renters. In the case of low ownership rate, coefficients attenuated toward zero but still significant at conventional levels. Altogether, the results in this table suggest that people react differently when they are exposed to house price shocks in supply constrained areas. The next table, proceeds to further examine the relationship by demographic characteristics. The differences across demographic groups are highlighted in Table 7 in which I expect to find that younger, non-native, less educated and ethnic groups will be more affected by increases in the house prices. Overall, the results are in line with this notion. For example, a 10,000 increase in house prices leads to a 4.01 percent decrease in births for the non-white group. For home owners, the same amount of increase is associated with 1.33 percent in birth rates. Aged and renter cohort seemed to hit harder than aged and renter cohort. This may be because this cohort witnessed substantially higher rises in house prices, implying less affordability. On the other hand, the older cohort seemed to benefit from this long-standing real house price growth. In other groups, the results maintain the expected sign of direction while being highly significant in most cases. This finding could be because there is lower levels 17

18 of access to loan and personal finance among disadvantaged groups. The following section of the paper is concerned with endogeneity of house prices and instrumental variable estimation results are presented. 4.2 IV Estimation In this section, I present the IV estimates of the house price-fertility relationship using county level refusal rates for major development projects as an instrument for the county level house prices. As previously discussed in Section 3, if the empirical estimation omits fertility related factors that are correlated with house prices and that are not accounted for by the linear trends and the fixed effects, the OLS estimates will yield biased parameter estimates. In addition, if higher fertility rates cause rises in house prices, the OLS results will mislead inferences regarding the effects of the house prices on fertility. Before discussing the instrumental variable estimates, an assessment of the power of the instrument at the first-stage is needed. Appendix Table A1 presents the results from first-stage regressions of house prices on the refusal rate variable. The table presents the coefficients for the instrument similar to the Table 3, adding successively more controls in the models. In all specifications, the refusal rate is positively associated with the house prices and the effect is statistically significant at one percent level. As such, the instrument seems to have a particularly stronger effect as more control and trends included. The results for first stage F-test also show that the first-stage relationships are fairly strong (above 10) and the Kleibergen Paap F statistic for weak identification is greater than the Stock and Yogo values, implying that I reject the null that the instrument is weak. Table 5 presents the IV estimates in which I replicate the OLS specifications from 18

19 Table 4. Confirming the OLS results, the IV coefficients indicate that an increase in house prices leads to an increase in birth rates among home owners, whereas the opposite is true among renters. In each model, the IV estimates are larger than the OLS estimates. Finding higher coefficients suggest that these estimates also reflect the local spillovers in housing market. In other words, among others, they capture: 1-) the main effect due to an increase in house prices, and 2-) expectations about future price growth. Table 6 present results based on the housing supply constraints (the counties with low -25th Percentile- and high -75th Percentile- level of ownership rates). In general, the pattern remain similar to the OLS findings. The Column 3 presents results for the counties with low ownership rate in which the IV results are almost three times larger than the OLS estimates and are significant at five percent level. For the counties with high ownership rate, in column 4, the IV results are also substantially larger and significant at one percent. The IV results for the demographic characteristics are also strong. However, nonwhite and degree level categories fail to reject the over-identification restriction. Therefore, the OLS estimates are clearly preferable to IV estimates for these categories. The estimates are highly significant in other characteristics and the effects of house price increases are even stronger than those from the OLS regression for both home-owners and renters. In particular, the coefficients on the non-native and younger specifications exceed the OLS estimate considerably, indicating the importance of credit constraint. The following part of the results proceed to explore whether the responses remain robust to alternative housing price measures. 19

20 4.3 Robustness Checks The results presented thus far demonstrate the fact that house price is an important determinant of the fertility outcome. Even though the effects vary with estimation features of the model, an increase in house prices clearly has a positive impact on birth rates among home owners and has a negative impact on birth rates among renters. Nevertheless, I conduct additional analyses in order to detect whether the main findings remain stable to the different measures of the house prices, namely, median house prices, lower quartile house prices and lagged house prices. In fact, the results hold through columns 1 to 4. For the median and lower quartile house prices, the magnitude of the estimation smaller than the main OLS and IV specifications. Between columns 3 to 5, I replace the house price variable with additional lags. Although the effects get smaller at higher lags, the results are still in line with the main findings of this paper. While in the main specification I control for length of stay to account for premia or discount related factors of housing tenure, I now use this measures in an attempt to further check the robustness of my results. More specifically, I investigate the relationship by three different measures: length of stay less than 2 years, between 2 to 5 years and more than 5 years. Table 9 presents the results from this exercise. Overall, the coefficients are similar to the ones in the main models and become larger after 2 years of residence. In sum, the analysis conducted in this section indicates that the effects found in the main specifications are robust to different measures. 5 Conclusion The results presented in this paper show that house prices significantly affect birth rates. The stronger effects of house prices appear in the models where counties exhibit high 20

21 level of home ownership rate as well as the models of the demographic characteristics where cohorts tend to be credit constraints. Moreover, the findings do not depend on the estimation methodology used, even though I do find relatively larger coefficients when I instrument for county level house prices. I also find similar results when I use different measures of house prices. In both OLS and IV results, I find a strong evidence that the median and lower quartile house prices significantly affect birth rates. This finding, again, is reproduced when I instrument for house prices. For the lagged house prices, the results indicate that house prices have a significant impact on fertility. However, the findings attenuate towards zero after the third lag. These findings for different length of lags cast an important behavioral interpretation of the fertility responses to housing market trends. That is, people seem to take housing market changes in previous years into account. In the introduction section, I mentioned previous works conducted in the US as well as the recent housing market and fertility trends in the UK. In this respect, the findings of this paper mainly show that: 1-) Tightly regulated housing market has larger effects on fertility, 2-) recent trends in fertility can be partly explained by the increases in house prices. Finally, the expected variation in the house prices and fertility relation by age, educational attainment, marital status and country of birth is also documented. Although the study has successfully demonstrated the aforementioned findings, it is however limited by the use of total birth rates, and the findings cannot be transferable to birth orders. Future research should therefore concentrate on the investigation of birth orders while considering demographic subgroup characteristics. 21

22 References [1] Adsera, Alicia. "Vanishing children: From high unemployment to low fertility in developed countries." American Economic Review (2005): [2] Becker, Gary S. "An economic analysis of fertility." In Demographic and economic change in developed countries, pp Columbia University Press, [3] Becker, Gary S., and H. Gregg Lewis. "On the Interaction between the Quantity and Quality of Children. Journal of Political Economy. 81: 2, pp." S (1973). [4] Boldrin, Michele, Mariacristina De Nardi, and Larry E. Jones. Fertility and social security. No. w National Bureau of Economic Research, [5] Chandler, Daniel, and Richard Disney. "The Housing Market in the United Kingdom: Effects of House Price Volatility on Households." Fiscal Studies 35, no. 3 (2014): [6] Cigno, Alessandro, and Furio C. Rosati. "The effects of financial markets and social security on saving and fertility behaviour in Italy." Journal of Population Economics 5, no. 4 (1992): [7] Dettling, Lisa J., and Melissa S. Kearney. "House prices and birth rates: The impact of the real estate market on the decision to have a baby." Journal of Public Economics 110 (2014): [8] Hilber, Christian AL, and Wouter Vermeulen. "The impact of supply constraints on house prices in England." The Economic Journal (2014). [9] Hunt, Ben. "How Should Policymakers Respond to a Decline in House Prices?." United Kingdom: Selected Issues (2005). [10] Kravdal, Øystein. "The impact of individual and aggregate unemployment on fertility in Norway." Demographic Research 6, no. 10 (2002): [11] Kuenzel, Robert, and B. Bjornbak. "The UK housing market: Anatomy of a house price boom." ECOFIN Country Focus 5, no. 11 (2008). [12] Lino, Mark. "Expenditures on children by families." Family Economics and Nutrition Review 14, no. 2 (2002): 3. [13] Lovenheim, Michael F., and Kevin J. Mumford. "Do family wealth shocks affect fertility choices? Evidence from the housing market." Review of Economics and Statistics 95, no. 2 (2013):

23 [14] Malthus, Thomas. "An ESSAY ON THE PRINCIPLE OF POPULATION, AS IT AFFECTS THE FUTURE IMPROVEMENT OF SOCIETY WITH RE- MARKS ON THE SPECULATIONS OF MR. GODWIN, M. CONDORCET, AND OTHER WRITERS. LONDON, PRINTED FOR J. JOHNSON, IN ST. PAUL S CHURCH-YARD, 1798." St. Paul s Church-yard, London (1798). [15] Mincer, Jacob. "Market prices, opportunity costs, and income effects." Measurement in economics (1963): [16] Muellbauer, John, and Anthony Murphy. "Housing markets and the economy: the assessment." Oxford review of economic policy 24, no. 1 (2008): [17] Ozcan, Berkay, Karl Ulrich Mayer, and Joerg Luedicke. "The impact of unemployment on the transition to parenthood." Demographic Research 23, no. 29 (2010): [18] Schaller, Jessamyn. "Booms, busts, and fertility: Testing the Becker model using gender-specific labor demand." University of Arizona, unpublished manuscript (2012). [19] Sobotka, Tomáš, Vegard Skirbekk, and Dimiter Philipov. "Economic recession and fertility in the developed world." Population and development review 37, no. 2 (2011): [20] Van Groezen, Bas, Theo Leers, and Lex Meijdam. "Social security and endogenous fertility: pensions and child allowances as Siamese twins." Journal of public economics 87, no. 2 (2003): [21] Yi, Junjian, and Junsen Zhang. "The effect of house price on fertility: evidence from Hong Kong." Economic Inquiry 48, no. 3 (2010): [22] Willis, Robert J. "A new approach to the economic theory of fertility behavior." The Journal of Political Economy (1973): S14-S64. 23

24 Table 1: Summary Statistics Age Age Variable Mean Std. Dev. Mean Std. Dev. Fertility Rate (1000s) Avg. Gross Weekly Earnings Unemployment Rate UK Born Non UK Born White Others Degree Level/Higher Education A Level Below A Level Married/Civil Partnership Single Divorced/Widowed N Notes: The table provides age-specific demographic characteristics for the 131 counties used in the analysis. All means displayed are within cell means. Fertility rates are constructed by dividing the number of births by the corresponding female population using mid-year population estimates based on censuses, in which female ages range between Source for county level birth data is UK Office for National Statistics. The UK Labour Force Survey is being used to construct county-year-age group specific unemployment rates, gross weekly wages and demographic characteristics. Unemployment rates refer to the percentage of economically active people who are unemployed by ILO standards. The gross weekly wages are calculated by dividing self-reported gross annual pay by the number of weeks worked in the same calendar year and are CPI adjusted to 2005 pounds. 24

25 Table 2: Summary Statistics Age Age Variable Mean Std. Dev. Mean Std. Dev. Housing Tenure Owned Outright Bought with Mortgage Rented Length of Residency < 2 years years > 5 years Country of residence 12 months ago England/Wales Others The presence of children Any child <5 years old Any child 5-16 years old Working for the same firm 12 months ago Same Different Travel to work time <20 minutes minutes >60 minutes N Notes: The table provides age-specific demographic characteristics for the 131 counties used in the analysis. All means displayed are within cell means. The UK Labour Force Survey is being used to construct county-year-age group specific demographic characteristics. 25

26 Table 3: Summary Statistics Variable Mean Std. Dev. Source Description House Price 172, ,024 Land Registry Reports from the individual house price records of all residential property sales in England. Median House Price 141,214 86,447 Land Registry The median property price is determined by ranking (midpoint) all property prices within a county. Lower Quartile House Price 106,758 55,281 Land Registry The lower quartile property price is determined by ranking all property prices in ascending order in a county. The lowest 25 per cent of prices are below the lower quartile ; the highest 75 per cent are above the lower quartile Private Sector Weekly Rents DCLG Local Authority Housing Statistics Public Sector Weekly Rents DCLG Local Authority Housing Statistics No of Private Registered Dwellings 38,495 66,279 The Tenant Services Authority s Regularity Statistical Return (RSR) No of Local Authority Registered Dwellings 44,331 63,039 The Housing Strategy Statistical Appendix (HSSA) The rent data are only taken from the larger Private Registered Providers (PRPs) completing the relevant form. The rent data prior to have been collected on various different collection scales, onwards data were based on standardised 52 week collection calculated by DCLG from figures provided by local authorities. The data based on counts of self-contained units only. The data based on counts of housing that is owned and managed by Local Authorities. Notes: This table provides aggregate level housing market variables, their means and standard deviations for the 131 counties used in the analysis over the sample period, Above variables are aggregated up to the county level from the district level. All nominal values are CPI adjusted to 2005 pounds. 26

27 Table 4: Effects of House Prices on Fertility - Alternative OLS Specifications (1) (2) (3) (4) (5) (6) OLS OLS OLS OLS OLS OLS House Price *** *** *** *** *** *** (0.0029) (0.0030) (0.0022) (0.0028) (0.0046) (0.0045) HousePrice*OwnershipRate *** *** *** *** *** ** (0.0050) (0.0052) (0.0029) (0.0031) (0.0080) (0.0098) Unemployment Rate * (0.0010) (0.0008) (0.0008) (0.0007) (0.0008) Gross Weekly Wages * (0.0001) (0.0000) (0.0008) (0.0002) (0.0007) Year Fixed Effects County Fixed Effects Demographic Controls Ext. Demographic Controls County Linear Trends No No No No No No Ownership Specific Trends No No No No Group Specific Trends No No No No No N R Robust standard errors in parentheses and clustered at age group-county level. Dependent variable: log fertility rate by year, age group and county. Demographic controls include: marital status (single, married, divorced/widowed), ethnicity (white, others), education (degree level/higher education -HE qualification below degree level-), A-levels, below A levels, country of birth (UK Born & Non UK Born). Extended demographic controls include: the presence of children (any child <5 & any child 5), country of residence 12 months ago, working for the same firm 12 months ago, travel to work time, the length of residency (year <2, year 2-5, year>5), share of social housing. Mean ownership rates are calculated in the UK LFS by year, county and age group. House prices (10,000s), unemployment rates and gross weekly wages are matched by county and year of conception, specifically year, (t- 1). * p < 0.1, ** p < 0.05, *** p <

28 Table 5: Effect of Unemployment on Fertility - Alternative IV Specifications (1) (2) (3) (4) (5) (6) IV IV IV IV IV IV House Price *** *** *** *** *** *** (0.0130) (0.0097) (0.0089) (0.0153) (0.0178) (0.0191) HousePrice* OwnershipRate *** *** *** *** *** ** (0.0146) (0.0132) (0.0147) (0.0187) (0.0232) (0.0273) Unemployment Rate * (0.0011) (0.0008) (0.0008) (0.0007) (0.0008) Gross Weekly Wages (0.0001) (0.0001) (0.0002) (0.0003) (0.0001) Year Fixed Effects County Fixed Effects Demographic Controls No Ext. Demographic Controls County Linear Trends No No No No No Ownership Specific Trends No No No No Group Specific Trends No No No No No N R Robust standard errors in parentheses and clustered at age group-county level. Dependent variable: log fertility rate by year, age group and county. Demographic controls include: marital status (single, married, divorced/widowed), ethnicity (white, others), education (degree level/higher education -HE qualification below degree level-), A-levels, below A levels, country of birth (UK Born & Non UK Born). Extended demographic controls include: the presence of children (any child <5 & any child 5), country of residence 12 months ago, working for the same firm 12 months ago, travel to work time, the length of residency (year <2, year 2-5, year>5), share of social housing. Mean ownership rates are calculated in the UK LFS by year, county and age group. House prices (10,000s), unemployment rates and gross weekly wages are matched by county and year of conception, specifically year, (t-1). * p < 0.1, ** p < 0.05, *** p <

29 Table 6: Effects of House Prices on Fertility by County Level Supply Constraints (1) (2) (3) (4) OLS High IV Low Ownership Rate Ownership Rate (75 th Percentile) (25 th Percentile) OLS Low Ownership Rate (25 th Percentile) IV High Ownership Rate (75 th Percentile) House Price ** ** ** *** (0.0051) (0.0159) (0.0512) (0.0707) HousePrice*OwnershipRate *** * ** *** (0.0102) (0.0190) (0.0561) (0.0436) County Fixed Effects Year Fixed Effects Demographic Controls Ext. Demographic Controls County Linear Trends Ownership Specific Trends Group Specific Trends N R Robust standard errors in parentheses and clustered at age group-county level. Dependent variable: log fertility rate by year, age group and county. Demographic controls include: marital status (single, married, divorced/widowed), ethnicity (white, others), education (degree level/higher education -HE qualification below degree level-), A-levels, below A levels, country of birth (UK Born & Non UK Born). Extended demographic controls include: the presence of children (any child <5 & any child 5), country of residence 12 months ago, working for the same firm 12 months ago, travel to work time, the length of residency (year <2, year 2-5, year>5), share of social housing. Mean ownership rates are calculated in the UK LFS by year, county and age group. House prices (10,000s), unemployment rates and gross weekly wages are matched by county and year of conception, specifically year, (t- 1). * p < 0.1, ** p < 0.05, *** p <

30 Table 7: Effects of House Prices on Fertility by Demographic Characteristics (1) (2) (3) (4) (5) (6) (7) (8) OLS White Non White UK Born Not UK Born Age Age Degree Level <Degree Level House Price *** *** *** ** ** ** ** *** (0.0052) (0.0153) (0.0040) (0.0131) (0.0099) ( ) (0.0025) (0.0020) HousePrices*OwnRate *** * *** ** * ** *** *** (0.0079) (0.0134) (0.0066) (0.0284) (0.0272) ( ) (0.0064) (0.0040) (1) (2) (3) (4) (5) (6) (7) (8) IV White Non White UK Born Not UK Born Age Age Degree Level <Degree Lev House Price *** ** *** *** *** ** *** (0.0271) (0.0472) (0.0264) (0.0509) (0.0359) (0.0390) (0.0310) (0.0227) HousePrices*OwnRate *** ** ** *** * *** *** *** (0.0516) (0.0353) (0.0921) (0.0336) (0.0921) (0.0636) (0.0523) (0.0268) County Linear Trends Ownership Spec. Trends Group Specific Trends N R F Statistic (First Stage) Standard errors in parentheses and clustered at age group-county level. All regressions include county fixed effects, year fixed effects, demographic and extended demographic controls. Dependent variable: log fertility rates by year, age group and county. Demographic controls include: marital status (single, married, divorced, widowed), ethnicity (white, others), education (degree level/higher education -HE qualification below degree level-), A-levels, below A levels, country of birth (UK Born & Non UK Born), and the presence of children (any child <5 & any child 5). Extended demographic controls include: country of residence 12 months ago, working for the same firm 12 months ago, travel to work time, the length of residency (year <2, year 2-5, year>5) and share of social housing. Mean ownership rates are calculated in the UK LFS by year, county and age. House prices (10,000s), unemployment rates and gross weekly wages are matched by county and year of conception, specifically year, (t-1). * p < 0.1, ** p < 0.05, *** p <

31 Table 8: Effects of House Prices on Fertility by Alternative Measures (1) (2) (3) (4) (5) OLS Median House Prices Lower Quartile House Prices Average House Prices ct-2 Average House Prices ct-3 Average House Prices ct-4 House Price *** *** ** * (0.0052) (0.0153) (0.0026) (0.0079) (0.0083) HP*OwnRateNoLoan *** *** *** ** * (0.0009) (0.0034) (0.0029) (0.0038) (0.0036) IV Median House Prices Lower Quartile House Prices (1) (2) (3) (4) (5) Average House Prices Average House Prices ct-2 ct-3 Average House Prices ct-4 House Price ** ** *** * * (0.0192) (0.0358) (0.0124) (0.0262) (0.0352) HP*OwnRateNoLoan * * *** ** * (0.0031) (0.0124) (0.0216) (0.0312) (0.0390) County Linear Trends Ownership Specific Trnd. Group Specific Trends N R F Statistic (First Stage) Standard errors in parentheses and clustered at age group-county level. All regressions include county fixed effects, year fixed effects, demographic controls and extended demographic controls. Dependent variable: log fertility rates by year, age group and county. Demographic controls include: marital status (single, married, divorced, widowed), ethnicity (white, others), education (degree level/higher education (HE qualification below degree level), A-levels, below A levels, country of birth (UK Born & Non UK Born), the presence of children (any child <5 & any child 5). Extended demographic controls include: country of residence 12 months ago, working for the same firm 12 months ago, travel to work time. Fraction of cell for length of residence (year <2, year 2-5, year>5) is matched by county, year and age group and calculated for both owners and tenants. Mean ownership rates are calculated in LFS by year, county and age. House prices (10,000s), unemployment rates and gross weekly wages are matched by county and year of conception, specifically year, (t-1). * p < 0.1, ** p < 0.05, *** p <

32 Table 9: Effects of House Prices on Fertility by Length of Stay (1) (2) (3) OLS Less than 2 years Between 2-5 years More than 5 years House Price *** *** ** (0.0038) (0.0031) (0.0091) HousePrice*OwnershipRate *** *** * (0.0081) (0.0054) (0.0012) (1) (2) (3) IV Less than 2 years Between 2-5 years More than 5 years House Price ** * *** (0.0464) (0.0481) (0.0576) HousePrice*OwnershipRate *** ** ** (0.0313) (0.0502) (0.0308) County Linear Trends Ownership Specific Trends Group Specific Trends N R F Statistic (First Stage) Standard errors in parentheses and clustered at age group-county level. All regressions include county fixed effects, year fixed effects, demographic controls and extended demographic controls. Dependent variable: log fertility rates by year, age group and county. Demographic controls include: marital status (single, married, divorced, widowed), ethnicity (white, others), education (degree level/higher education (HE qualification below degree level), A-levels, below A levels, country of birth (UK Born & Non UK Born), the presence of children (any child <5 & any child 5). Extended demographic controls include: country of residence 12 months ago, working for the same firm 12 months ago, travel to work time. Fraction of cell for length of residence (year <2, year 2-5, year>5) is matched by county, year and age group and calculated for both owners and tenants. Mean ownership rates are calculated in LFS by year, county and age. House prices (10,000s), unemployment rates and gross weekly wages are matched by county and year of conception, specifically year, (t-1). * p < 0.1, ** p < 0.05, *** p <

33 Appendix Table A1: First Stage Estimates Alternative Specifications (1) (2) (3) (4) (5) (6) FS FS FS FS FS FS 3-Year MA Refusal Rate *** *** *** *** *** *** (0.0038) (0.0037) (0.0061) (0.0064) (0.0092) (0.0097) County Fixed Effects Year Fixed Effects Demographic Controls No Ext. Demographic Controls No No County Linear Trends No No No Ownership Specific Trends No No No No Group Specific Trends No No No No No N R 2 F Statistic (First Stage) Robust standard errors in parentheses and clustered at age group-county level. Dependent variable: mean house prices by year and county. Demographic controls include: marital status (single, married, divorced/widowed), ethnicity (white, others), education (degree level/higher education -HE qualification below degree level-), A-levels, below A levels, country of birth (UK Born & Non UK Born). Extended demographic controls include: the presence of children (any child <5 & any child 5), country of residence 12 months ago, working for the same firm 12 months ago, travel to work time, the length of residency (year <2, year 2-5, year>5), share of social housing. Mean ownership rates are calculated in the UK LFS by year, county and age group. House prices (10,000s), unemployment rates and gross weekly wages are matched by county and year of conception, specifically year, (t-1). * p < 0.1, ** p < 0.05, *** p <

34 Appendix Figure A1: Average Refusal Rates in England 34

35 Appendix Figure A2: Time Series Correlation - House Prices and Fertility Appendix Figure A3: Time Series Correlation - House Prices and Mean Weekly Rents 35

36 Appendix Figure A4: Share of Housing Wealth Appendix Figure A5: Time Series Correlation Equity Extraction and House Price Index 36