Monetary Policy and Birth Rates: The Effect of Mortgage Rate Pass-Through on Fertility

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Monetary Policy and Birth Rates: The Effect of Mortgage Rate Pass-Through on Fertility Fergus Cumming and Lisa Dettling 2020-002 Please cite this paper as: Cumming, Fergus, and Lisa Dettling (2020). “Monetary Policy and Birth Rates: The Effect of Mortgage Rate Pass-Through on Fertility ,” Finance and Economics Discussion Series 2020-002. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2020.002. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Monetary Policy and Birth Rates: The Effect of Mortgage Rate Pass-Through on Fertility∗ Fergus Cumming† Lisa Dettling‡ December 12, 2019 Abstract This paper examines whether monetary policy pass-through to mortgage interest rates affects household fertility decisions. Using administrative data on mortgages and births in the UK, our empirical strategy exploits variation in the timing of when families were eligible for a rate adjustment, coupled with the large reductions in the monetary policy rate that occurred during the Great Recession. We estimate that each 1 percentage point drop in the policy rate increased birth rates by 2 percent. In aggregate, this pass-through of accommodative monetary policy to mortgage rates was sufficiently large to outweigh the headwinds of the Great Recession and prevent a “baby bust” in the UK, in contrast to the US. Our results provide new evidence on the nature of monetary policy transmission to households and suggest a new mechanism via which mortgage contract structures can affect both aggregate demand and supply. JEL classification: E43, E52, J13. Keywords: Mortgages, Monetary Policy, Birth Rates, Fertility, Natality, Interest Rates. ∗WethankSaleemBahaj,NeilBhutta,AngusFoulis,SarenaGoodman,JoanneHsu,JohnSabelhaus,Fergal ShortallandseminarparticipantsattheEuropeanCentralBank,theFederalReserveBoard,andtheUniversity of Essex for helpful comments and suggestions. The analysis and conclusions set forth are those of the authors and do not indicate concurrence with the Bank of England, the Board of Governors, or the Research Staff or committeesateitherinstitution. TheSecureResearchServiceagreesthatthefiguresanddescriptionsofresults in this paper may be published but the ONS does not necessarily endorse any interpretation of their statistical data. All statistical results remain Crown Copyright but are governed by the Open Government License. This work uses research datasets which may not exactly reproduce National Statistics aggregates. All errors remain our own. †fergus.cumming@bankofengland.co.uk, Bank of England. ‡lisa.dettling@frb.gov, Federal Reserve Board of Governors. 1

1 Introduction Economists and policy-makers have long debated the question of how and to what degree monetary policy affects the real economy. The role of household decision-making is a central part of this debate. It has been recognized for some time that one of the most direct ways households are affected by monetary policy is through mortgage interest rates. Indeed, a number of recent papers have documented that when mortgage interest rates and scheduled mortgage payments fall, families increase their spending on durables and other consumer goods (Di Maggio et al., 2017; Flodén et al., 2017; La Cava et al., 2016). But we have a limited understanding of some of the broader effects monetary policy has on the real economy, despite standard economic models suggesting a plethora of household decisions might be affected by a decline in committed expenditures. We start to fill this gap by examining whether monetary policy influences one of the most important decisions a family will make: whether or not to have a baby. In this paper, we employ administrative data covering the universe of births and mortgage originations in the UK to explore how the dramatic fall in policy interest rates in the Great Recession influenced household fertility decisions. We exploit unique institutional features of the UK mortgage market for identification. At the onset of the Great Recession, around half of UK families of child-bearing age were mortgaged home-owners. Of those, about half had mortgages that were directly tied to the policy rate. The other half had mortgages on an initial fixed-rate period, which would reset to an adjustable rate sometime over the next few years.1 High pre-payment penalties made early refinance virtually non-existent. Thus, when the Bank of England lowered its policy rate 4.5 percentage points during 2008 and 2009, about a quarter of families’ mortgage payments fell immediately, another quarter of families’ payments would 1As we describe in detail in Section 3, the timing of when the initial fixed periods would reset was determined by both the date of origination and the length of initial period families chose (typical options ranged from two tofiveyears). Mortgageswithanadjustableandfixedratecarriedroughlysimilarinterestratesatorigination, andtheparticularmortgagecontractfamilieschosewaslargelyanartefactofwhichhappenedtobemarginally cheaper on the day of origination. This price was determined by the slope of the yield curve, not borrower characteristics. 2

fall at some point over the next couple of years, and the other half of un-mortgaged families would never be affected. For families with an adjustable rate, interest rate pass-through was sizeable and swift, lowering their mortgage payments by over £1,000 a quarter (roughly 42 percent). Our identification strategy exploits these pre-determined, idiosyncratic differences in mortgage contract choice to estimate how the large, unexpected decline in the monetary policy rate in 2008-2009 affected families’ fertility decisions over the next three years. We implement this identification strategy by constructing birth rates at the local authority unit(LAU)-age-grouplevel, datedtothequarterofconception, andmergethatwithaquarterly measureofsimulated“exposure”tomonetarypolicyattheLAU-age-grouplevel.2 Ourexposure measurecapturestheshareoffamiliesinaLAU-age-grouponanadjustablerateineachquarter, and grows over time as fixed initial periods expire and mortgages reset to an adjustable rate.3 Weinteractthisquarterlymeasureofexposurewiththecumulativedeclineinthepolicyinterest rate up to that point, and employ a fixed effects design that controls for level differences in birth rates across groups and over time. Our main coefficient of interest thus describes how declining interest rates affect conception rates as the share of families who have reset to a lower rate increases. In this setting, identification arises from variation in the timing of resets across LAU-age groups. Our results indicate that a 1 percentage point reduction in the monetary policy rate - which decreased mortgage payments by 12 percent on average - leads to a 5 percent increase in the birth rate among families on an adjustable rate. At the mean adjustable rate share, this is equivalent to a 2 percent increase in the UK birth rate. In aggregate, our estimates imply that accommodative monetary policy in late 2008 and 2009 led to 14,500 additional babies being born in 2009, and increased birth rates by 7.5 percent over the following three years. Heterogeneity analysis on subsamples indicates the effects are larger for groups with higher 2Local authority units are similar in size to US counties. 3Our measure is simulated because it uses only mortgages that existed as of 2008Q2, and how those contracts evolved over the next three years. We do this primarily so that our measure is explicitly stripped of any endogenous home-buying or refinancing in response to changing rates. We describe this in more detail in Section 4. 3

loan-to-incomes (LTIs), higher loan-to-values (LTVs) and lower average incomes. This suggests liquidity constraints play a role in explaining the fertility response to mortgage rate changes. Our results are robust to a number of robustness checks on the data construction and empirical specification, and we confirm the implied elasticities we estimate are in line with previous estimates of how changes in current-period liquidity affect fertility choices. AggregatebirthratesgrewintheUKovertheperiodwestudy, butothercountries-notably, the US - experienced a Great Recession “baby bust”.4 Figure 1 shows time trends in quarterly birth rates in the US and UK, with the period in which unemployment grew from its trough to peak highlighted by the grey bar, and the path of monetary policy noted by the dashed black line.5 Inbothcountries, birthratesbegintofallalmostassoonastheunemploymentratebegins to rise, but in the UK that trend is reversed once the policy rate (Bank Rate) begins to drop. In the US, the downward trend continues through the recession. We conduct counterfactual simulations based on our estimates and find that in the absence of monetary policy, declining employment and house prices during this time period would have otherwise led to a decline in birthratesintheUK,consistentwithpreviousliteraturewhichhasdemonstratedthatfertilityis pro-cyclical (e.g., Schaller (2016)). In other words, the fertility stimulus effects of UK monetary policy were sufficiently large to outweigh the headwinds of the recession. We also conduct a simulation based on levels of interest rate exposure, and find that once 30 percent of mortgages have an adjustable rate, the change in birth rates over the recession becomes positive. Though only suggestive, we note that this threshold is well above the US adjustable rate share.6 Of course, other factors might confound our interpretation that the drop in rates caused a change in UK fertility decisions, particularly since our estimates are set in the context of the Great Recession. To this end, it is crucial that our analysis controls for quarter fixed effects, so that our estimates are net of any changes in economic conditions that affect all 4Thiswaswidelycoveredinthemedia,see,forexample,https://www.theatlantic.com/business/archive/2014/09/therecessions-baby-bust/380909/. 5The US series was constructed to be directly comparable to the UK series. More details on the US data can be found in Section 5.3.3. UK and US unemployment rates peaked at 8.4 and 9.5 percent, respectively. 6AccordingtotheSurveyofConsumerFinances,theshareofmortgagedfamilieswithafullyfloatingadjustable rate was 6 percent in 2007. Authors’ calculations from the US Survey of Consumer Finances. 4

families, regardless of their mortgage type (or lack thereof). We also control for local house prices and unemployment rates to allow for spatial differences in economic conditions. In some specifications, we further employ two-way interactions between the group and time fixed effects, allowing, for example, different LAUs or age-groups to be on flexible fertility trends. Essentially, our strategy leverages the fact that all families in particular age-group and LAU (e.g., with adjustable rate, fixed rate, or no mortgage at all) are similarly exposed to changing economic conditions, but only families who transition onto an adjustable rate experience a change in their mortgage payment. Another potential threat to identification is the possibility that pre-determined exposure rates were related to families’ future fertility intentions. Survey evidence from the summer of 2008 indicates that just 10 percent of households expected rates to fall at all in the coming year (Bank of England, 2018). Thus, it seems very unlikely that families would have been motivated by any kind of strategic, anticipatory behavior. Still, it could be that couples planning to have children were more likely to choose a particular contract, or move at a particular time, causing a spurious correlation between reset timing and fertility decisions. To allow for this possibility, our analysis includes a host of controls for families’ mortgage contract terms, including whether the contract originated with an adjustable or fixed rate and tenure. We also control for a host of characteristics that might be related to group-level fertility intentions, such as home ownership rates, educational attainment, household income and housing LTVs. We also present estimates from an alternate event study-style analysis, which uncovers no evidence of differential fertility pre-trends in the three years prior to monetary easing across groups of families of differing levels of exposure. An important question that arises from our analysis is whether the effects we estimate represent a shift in the timing of births, or in the number of children ever born to a woman (completed fertility). Given the recentness of the natural experiment we study, we cannot empirically answer this question.7 However, there is some suggestive evidence in favor of at 7We would need to wait for all women who were of child-bearing age during the Great Recession to complete their child-bearing years in order to conduct such an analysis. 5

least some change in completed fertility. First, survey evidence on interest rate expectations, combined with the ultimate length of the low interest rate environment, suggests the shock we study was perceived to be permanent. Standard models suggest this should lead to a change in completed fertility (Hotz et al., 1997). Second, time series evidence does not reveal a decline in birth rates among older women following the monetary easing period we study, which might be expected if women were bringing forward their child-bearing plans (see Figure 8). Third, there is evidence that decisions that begin as changes in timing ultimately lead to changes in completed fertility (Sommer, 2016). Still, changes in timing that lead to cyclical fluctuations in birth rates can have important effects on the economy. For example, fluctuations in cohort and class sizes have meaningful effects on children’s educational attainment, human capital investment and future labor market outcomes (e.g., Angrist and Lavy (1999); Stapleton and Young (1988); Korenman and Neumark (2000)). Our paper provides new evidence on the nature of monetary policy transmission to the real economy. Children are expensive, so a change in birth rates plausibly has spillover effects on consumer spending. In addition to food, clothing, and other daily necessities, many consumer durables purchases (such as a larger vehicle) are prompted by the addition of a child.8 Indeed, estimates indicate that the average cost of raising a child during their first year in the UK is almost £11,000.9 If we make the strong assumption that families previously saved this money, a simple back-of-the-envelope suggests that the additional 14,500 babies implied by our point estimates could have led to up to £130 million in additional spending in 2009 alone. And although this is surely an upper bound, it does not factor in the costs of raising children beyond their first year of life (the total cost of raising a child is around £230,000), nor any additional births that occurred due to the low interest rate environment in the years that followed 2009. On the whole, we view our results as shedding new light on a mechanism underlying previous 8Using data from the UK Living Cost and Food Survey, we find that mortgaged families with children under 2 are 4.1 points more likely to have recently purchased a vehicle than similar families without children (see Appendix Table A2). 9Estimatebasedonthe2014CenterforBusinessandEconomicResearchandLiverpoolVictoriaCostofaChild Survey (Liverpool Victoria, 2014). 6

estimates of mortgage rate pass-through to aggregate consumer spending (Di Maggio et al., 2017; Flodén et al., 2017; La Cava et al., 2016). But in contrast to that work, the outcome we study implies not only a short-term change in spending patterns, but also suggests monetary policy can have effects on aggregate demand that last a lifetime. Our work is related to a growing literature on the importance of contractual rigidities in the mortgage market for monetary policy transmission during economic downturns (Rubio, 2011; Calza et al., 2013; Auclert, 2019; Cumming, 2018). These papers have shown that monetary policy is more effective as an automatic stabilizer in recessions when mortgage contracts have fewer rigidities and rate pass-through is quicker. We provide direct estimates of how one such rigidity - the share of fixed-rate mortgages - affects the demand for children (which, like the demand for consumer goods, falls in recessions).10 For example, our estimates imply that monetary policy has three times the stimulus effect on birth rates when 75 percent of families have an adjustable rate compared to when 25 percent of families have an adjustable rate. To the extent that birth rates have spill-over effects on consumption, this confirms the notion that a higher prevalence of adjustable-rate mortgages can increase the spending response to expansionary monetary policy (Auclert, 2019; Guren et al., 2018; Rubio, 2011). Though our paper uncovers a direct effect through the mortgage market, there are many ways that monetary policy indirectly affects birth rates. There is ample evidence that monetary policy affects house prices (Ahearne et al., 2005), which have been shown to positively affect fertility decisions (Dettling and Kearney, 2014; Lovenheim and Mumford, 2013). There is also evidence that lower mortgage rates positively affect home-buying (Bhutta et al., 2017) and household formation (Martins and Villanueva, 2009), which are often viewed as pre-cursors to child-bearing. And accomodative monetary policy increases labor demand and employment (Cumming, 2018), which also positively affects birth rates (Schaller, 2016). By design, our empirical strategy isolates the direct effect through the mortgage market so that our estimates are net of any of these general equilibrium effects. But once we consider all of the ways that 10See, for example, Schaller (2016); Dettling and Kearney (2014). 7

monetary policy might affect birth rates in general equilibrium, it becomes clear that our estimates are likely a lower bound on the total effect of monetary policy on birth rates. Our results have important implications for monetary policymakers, and speak to a mechanism via which monetary policy can affect the supply side of the economy. Parenthood affects labor supply decisions (Angrist and Evans, 1998), which directly affects measurement of the unemployment rate, and could have knock-on effects for household income and productivity. Changes in birth rates also feed into population growth and dependency ratios, which affect the transmission of monetary policy (Berg et al., 2019). And there is evidence that changes in dependency ratios can alter the natural rate of interest (Lisack et al., 2017; Rachel and Smith, 2017), which suggests the assumed exogeneity of the natural interest rate might need re-examining. Overall, our results highlight several otherwise unexplored ways that monetary policy can influence aggregate demand and supply in the short and medium run. 2 Conceptual Framework In this short section, we briefly outline how a change in mortgage payments might affect household fertility decisions. Beginning with the seminal work of Becker (1960), economists have modeled fertility behavior within the neoclassical choice-theoretic framework. In the simplest static models, children are treated like durable goods and parents are consumers who choose the quantity of children that maximizes lifetime utility subject to the price of children and their budget constraint. In this framework, a reduction in a family’s scheduled mortgage payments can be thought of as either: (1) a reduction in committed expenditures which increases disposable income; or, if families view housing as an input in child-rearing, (2) a reduction in the price of child-rearing. In either case, theory predicts that reducing scheduled mortgage payments will increase the demand for children, increasing the number of children born.11 Con- 11Becker and Lewis (1973) proposed that parents have preferences for both the quantity and quality of their children, and face a trade-off between the two. Thus, any increase in the demand for children might plausibly lead to an increase in the number of children, or in investment per child. Our paper will ignore the latter channelandfocusonwhetherthequantity ofchildrenrespondstochangesinmortgageinterestrates. Though 8

sistent with the predictions of these models, there is growing empirical evidence using credible, quasi-experimental methods that changes in income and prices have tangible effects on currentperiod fertility (see, for example, Black et al. (2013); Kearney and Wilson (2018); Autor et al. (2018) on income effects and Milligan (2005); Cohen et al. (2013) on price effects). In dynamic, life-cycle models of fertility, families not only choose their desired family size, but also time their child-bearing throughout their life-course (see Hotz, Klerman and Willis (1997) for a review). In these models, transitory changes in prices over the life-cycle can affect the optimal timing of child-bearing without affecting the total number of children that will be born to a woman. If capital markets are imperfect, and families cannot borrow or save, transitory changes in income also affect the timing of child-bearing. Thus, whether a change in income or prices affects completed fertility (a so-called quantum effect) or the timing of fertility (aso-calledtempo effect)dependsonwhethertheshockisviewedtobepermanentortransitory, as well as the extent to which families are liquidity or borrowing constrained.12 Interest rates are generally understood to be cyclical, thus, we might model a change in mortgage payments due to interest rates as a transitory shock. Still, what matters in these models is what households believe about the nature of the shock. To that end, it is useful to examine survey evidence on interest rate expectations to better gauge families’ beliefs. In the summer of 2009, 70 percent of households expected rates to either stay the same or rise only a little.13 This suggests most families perceived the changes to their mortgage payments to be at least somewhat permanent. And this expectation of permanence proved to be fairly accurate: rates have remained low for the 10 years since 2009, ultimately encompassing a large portion of many women’s child-bearing years. These patterns would suggest the change in mortgage we do not study this in our paper, it is clear that parents might also adjust the quality of their children in response to a change in mortgage interest rates, for example, by investing more in their children’s education. We leave the investigation of such a possibility to future work. 12In these models, monetary policy can affect the optimal timing of child-bearing by affecting intertemporal allocation via the real rate of interest. We will ignore that dimension of monetary policy in our analysis for two reasons. First, because the variation in the real rate in the Great Recession was substantially less than the nominal rate, and second, because changes in the real rate would affect all families and not just mortgage-holders. Our empirical strategy will abstract away from those effects by design. 13SourceistheInflationAttitudesSurvey,whichisavailableathttps://www.bankofengland.co.uk/statistics/researchdatasets 9

payments we observe may have been perceived as closer to a permanent shock. Before moving on, we note that for exposition we will discuss fertility choice in our paper as though it is a simple decision. Of course, in reality it is a complex decision with a stochastic outcome, and families’ child-bearing goals will not always be exactly realized. Thus, any monetary policy pass-through to birth rates that we observe is likely to be a muted reflection of a family’s latent fertility preferences. 3 Background on UK Mortgage Market Prior to the onset of the Great Recession, the vast majority of UK mortgages featured a short initial period with either an adjustable rate or fixed rate. The interest rate on the former was tied one-for-one to Bank Rate, while the interest rate on the latter was typically fixed for 2-5 years. Only 3 percent of families obtained initial fixed-rate periods of 10 years or more.14 Following the end of the initial fixed or adjustable period, all UK mortgages revert to the Standard Variable Rate (SVR), which is an adjustable rate. Henceforth, we will refer to all mortgages on an adjustable rate (either initial period or having reset to the SVR) as an ARM.15 All UK mortgages carried high pre-payment penalties during the initial period. Typically, these fees ranged between 2 and 5 percent of the loan balance outstanding (where the fees increased with time left in the initial period). Thus, it was rarely beneficial to refinance into a new mortgage rate before the end of the initial period (even if rates fell substantially).16 After the end of the initial period when the mortgage reverted to the SVR, these fees were lifted. And because the SVR was typically several hundred basis points above prevailing refinancing offer rates (see Figure 2), the majority of mortgagors found it beneficial to refinance their contracts 14Authors’ calculations based on the data used in this paper. 15Conceptually, UK mortgage contracts are similar to the short-run hybrid mortgage product in the US that is commonly referred to as an adjustable rate mortgage. As is the case in the US, UK mortgages typically feature a 20 or 30 year term. However, in contrast to the US, virtually no UK mortgages will include a fixed ratefortheentire20or30yearterm. Thus, themostcommonmortgageproductintheUS-the30yearfixed rate mortgage- is virtually non-existent in the UK. 16This contrasts with most fixed rate mortgages in the US, where it is possible to pre-pay (and thus, refinance) during the fixed rate period without penalty. 10

upon reset when they reverted to the SVR. As a result, most mortgagers in the UK refinanced into a new mortgage precisely at the time each initial period expired (Cloyne et al., 2019). In practice, this implies that most families with a mortgage had obtained that mortgage in the past two to three years, and tenure in the mortgage was often disconnected from the year of home purchase. Becauseoftheseshortinitialperiodsandthestableeconomicenvironmentintheearly2000s, the choice between an adjustable or fixed initial period was viewed to have little effect on the lifetime cost of the loan. Many families switched between an fixed and adjustable contracts over time as they refinanced: between 2005 and 2008, around 40 percent of households chose the opposite contract when they refinanced, providing prima facie evidence that contract choice was largely unrelated to fixed family preferences. Rather, Cumming (2018) shows that relative prices at the time the borrower happened to originate or be eligible to refinance a mortgage are highly predictive of product choice. These price differences were induced by the slope of the yield curve, rather than individual borrower characteristics, and varied over time (see Figure 2). Between October 2008 and March 2009, Figure 2 shows that Bank Rate fell from 5 percent to 0.5 percent. Families with an ARM at this time experienced a substantial, unexpected decline in their monthly mortgage payment. Table 1 displays information on the distribution of the change in quarterly ARM payments in our data.17 The average family with an ARM experienced a decline in quarterly payments of £1130, representing 42 percent of their initial mortgage payment, or 7.5 percent of their gross income. There is also considerable variation in the data in the size of payment decline, ranging from £447 at the 25th percentile to £1228 at the 75th percentile. Across age-groups, older families experienced the largest average drop in terms of pounds sterling (£1208), and families aged 16-24 experienced the largest drop relative to their initial income (8.5 percent). 17Note that families who were on initial adjustable period saw their rate fall one-for-one with Bank Rate. Families on the SVR (having reset from either an adjustable or fixed rate) experienced a smaller decline in their rate. Table 1 displays the average payment drop pooling both types of families. Families on adjustable initial periods are 33 percent of the sample. 11

For families with fixed initial periods, high pre-payment fees (combined with falling house pricesandtightcreditconditions)meantthatitwascost-prohibitive(orimpossible)torefinance into a lower rate. Rather, those families had to wait for their fixed term to expire before they would automatically reset to the SVR. For some families, this was a couple of days, but for others it was up to 5 years. The timing of the reset depended on a combination of when they had last refinanced (or originated) their current mortgage and the fixed term length they had chosen. Consider, for example, two families who refinanced into a fixed rate mortgage in October 2006, where the first chose a 2 year initial period and the second chose a 3 year initial period. The thought experiment motivating our empirical approach is to compare how the 4.5 percentage point decline in Bank Rate affected the first family (whose fixed term ended in October 2008), with the second family (whose fixed term would not end until October 2009). Although both families opted for a fixed rate in October 2006, the first family received a large reduction in their monthly mortgage payment between October 2008 and March 2009, while the other family did not. And a family who had chosen an adjustable initial period in October 2006 would have received a large payment reduction even sooner, beginning in September 2008. This sort of pre-determined, idiosyncratic variation in mortgage contract terms is the natural experiment we exploit in this paper. 4 Data and Empirical Approach 4.1 Data We obtained birth data from the Office for National Statistics (ONS). These data are based on UK Vital Statistics andrepresentacountofvirtuallyallbirthsinEnglandandWales.18 Werefer to our analysis as covering the United Kingdom for exposition only. The data includes the exact date of birth and conception and the mother and father’s (when present) age and location of residence. The data do not include information on home ownership or mortgage characteristics, 18The data do not include Scotland and Northern Ireland so we will not examine those areas in our analysis. 12

so in order to match to the relevant independent variables of interest we aggregate births by LAU, age-group and quarter of conception. We use the date of conception since that is the time period in which a family is making the decision to get pregnant, and thus, the economic conditions at that time would be the most relevant in the decision-making process. WeconstructquarterlyconceptionratesattheLAUage-groupandquarterlevelbymatching theaggregatedcell-levelbirthstoannualfemalepopulationcounts, alsoobtainedfromtheONS. The age-groups we use are 16-24, 25-35, and 35-44 because those are the groups available across all of our datasets. The time period we study are the three years (12 quarters) from the third quarter of 2008 through the second quarter of 2011. We begin in the third quarter of 2008 because it is just before the 4.5 percentage point drop in the policy rate occurred. We end our analysis in 2011 because by that point more than 80 percent of the fixed terms originated before the third quarter of 2008 had already expired, which means we no longer have a clean source of identification as families would have been able to choose a new mortgage. Figure 3 provides more details on the timeline of policy, our data and our analysis. Themortgagedataisderivedfromadministrativedataontheuniverseofnewandrefinanced mortgages issued by UK lenders from 2005 through 2009, which is collected by the Financial Conduct Authority and distributed in the Product Sales Database (PSD). Because households in the UK rarely obtain fixed rate terms longer than five years and there is an active refinancing market, weestimatethatbyJune2008thiscountoforiginationsrepresentsatleast80percentof the stock of all mortgages outstanding at that time.19 The mortgage data includes information on the home’s location (local authority), purchase price, loan value, initial period length, term, whether there is a fixed or adjustable rate, whether the mortgage was a purchase or refinance, whether the purchase was a first home purchase, and the borrowers’ gross annual income and 19There are no good estimates for the UK of the stock of all mortgages during this time period. However, data from 2015 on the stock suggest an estimate of 80-85 percent. Since there are almost no long term fixed contracts issued in the UK, the missing mortgages would be all be on the SVR. So unless these missing mortgages are unevenly distributed across the UK, it should have little impact on the relative proportions on an adjustable rate. In the robustness checks in section 5, we implement a procedure to impute missing mortgages where we age backwards the stock from 2015 that we do not observe refinancing from 2005-2015, as in Cumming (2018). 13

age.20 We describe this data in more detail in the Appendix. To measure mortgage interest rate exposure we simulate the share of families on an adjustable rate at the LAU by age-group by quarter level. We use a simulated measure because home-buyingor refinancingdecisions duringthe periodin whichinterestrates were fallingcould be endogenous to fertility choice. Therefore, we construct our measure of exposure using baseline characteristics of mortgages that existed as of June 2008. In particular, we use information on the origination date and initial period length to estimate the expiration date of the mortgages that had fixed terms. We use this to construct the share of mortgagers on an adjustable rate in each LAU, age-group and quarter (which we define as simply 1 minus the imputed share still on a fixed rate). Because the mortgage data does not include information on those without a mortgage and we are interested in the exposure rate among all households in a LAU-age group, we supplement the mortgage data with data on home ownership rates. In particular, we interact the mortgager adjustable-rate share with LAU-age group home ownership rates, which we obtained from the 2001 UK Census.21 To be precise, we estimate Equation 1 for each local authority l, age-group a and quarter t: fixedrate owners l,a,t l,a,2001 exposure = (1− )×( ) (1) l,a,t mortgagers households l,a,t l,a,2001 In practice, there is considerable variation in this simulated measure of exposure across local authorities and over time within the UK. Figure 4 shows this geographic dispersion in ARM shares across England and Wales as of June 2008. ARM shares ranged from 15 percent to 43 percent across LAUs, and there are areas of relatively high and low ARM shares throughout the 20The mortgage data has information on the “lead borrowers” age, but not sex. In our main specification we assume that is the mother’s age and match accordingly, but in robustness checks we will instead assume that is the father’s age, when it is available. 21We use 2001 data because it is the most recent year available, but even if we had contemporaneous data we mightpreferatleastsomelagsincerecenthome-buyingcouldbeendogenoustofertilityintentions. Ourmeasurewillignorefamilieswhoareoutrightowners,however,thisisaverysmallshareofownersintheage-groups we study. For example, the UK housing Survey shows that in 2009 outright owners were 1.1 percent of 16-24 yearolds,2.7percentof25-34yearoldsand7.5percentof35-44yearolds. See: https://assets.publishing. service.gov.uk/government/uploads/system/uploads/attachment_data/file/6695/1750765.pdf. 14

country, even, for example, within London. There is also considerable variation in ARM shares over time (top panel, Table 2): on average 21 percent of households were on an adjustable rate in the third quarter of 2008 (with a standard deviation of 8.7 percent across LAUs), and this increases to 34 percent a year later (with a standard deviation of 12.5 percent across LAUs) and 43 percent by 2010 (with a standard deviation of 14.9). 4.2 Empirical Specification Our main analysis will consist of a series of ordinary least squares regression models of the following form: ln(birthrate ) = β exposure ×∆bankrate l,a,t+3 1 l,a,t 2008Q2,t +β E +β X +γ +α +θ +(cid:15) (2) 2 l,t 3 l,a a l t l,a,t Where the level of analysis is the local authority (l) age-group (a) quarter (t) cell. In these specificationswematchallrelevanttime-varyingindependentvariablesofinteresttothequarter of conception (notated here for exposition as three quarters after the measurement of economic variables, although we use the actual date of conception in our data construction). Note that LAUs in which any cell has fewer than ten observations in any of our datasets are dropped. Because there are 12 quarters of data, three age groups, and 343 local authorities, our analysis sample will have up to 12,348 observations. All regressions are weighted by the total number of births in each cell, and standard errors are adjusted for clustering at the LAU level. Thecoefficientofinterestβ capturestheinteractionbetweenexposure and∆bankrate , 1 lat 2008Q2,t which describes how an increase in the share of households on an adjustable rate affects the relationship between the cumulative decline in Bank Rate since the second quarter of 2008 and the log of birth rates three quarters later.22 In other words, β captures the extent of monetary 1 policy-pass through, since the interaction between exposure and ∆bankrate will be lat 2008Q2,t 22The cumulative decline in Bank Rate is defined as the average decline over the quarter, where the average is based on the daily value of Bank Rate. 15

larger when there are larger changes in Bank Rate and more families are exposed to those changes. If monetary policy pass-through affects birth rates, we would expect β to be positive. 1 The vectors γ , α , θ represent a full set of age-group, quarter and local-authority fixed a l t effects to account for fixed differences across local authorities, time and age-groups in birth rates and mortgage rate exposure. Because ∆bankrate only varies over time, θ nets 2008Q2,t t out any differences in birth rates induced by monetary policy that affect families who do not have an ARM. Similarly, α nets out any time-invariant differences across local authorities in l preferencesforchildrenwhichmightbecorrelatedwithexposure . So,iffamiliesinManchester lat have more children on average than families in London, and also are more likely to have an ARM, α captures those differences. l The vector X represents a vector of age-group by local authority controls. Importantly, la this includes characteristics of the mortgage contracts chosen by families in each age-group and local authority. This includes the home ownership rate and the share of families who initially chose an ARM for their current mortgage contract. We include these variables to facilitate a casualinterpretationofβ , incasegroupswithdifferentfertilitypreferencesmighthavedifferent 1 propensities to choose an ARM or own a home. We also control for the number of years since the mortgage was originated (in year-groups), the number of years before the loan will be paid off (in year-groups), whether the mortgage was used for a refinance or a purchase, and if it was a purchase, whether the buyer was a first time home owner. We include these variables to allow for the possibility that families with fertility intentions may have different rates of exposure because of different lengths of tenure in their homes. All of the mortgage variables are measured in the June 2008, just like exposure . lat The vector X also includes age-group by local authority measures of educational attainla ment, income, and wealth. Educational attainment is measured as the share of women in each group who have a higher-education qualification and is obtained from the 2001 Census. To capture incomes, we use the 5th, 25th, 50th, and 75th percentile of gross annual incomes, which we collect from the mortgage data. We include the bottom of the distribution to better capture 16

non-owners.23 We also control for differences in wealth and leverage using loan-to-values and loan-to-incomes at origination, which are discretized into 10 and 6 categories (plus non-owners), respectively. Ouranalysisalsoincludesanumberoftime-varyingcharacteristicsoflocalauthoritieswhich the previous literature has shown have an impact on fertility choice. In particular, the vector E includes controls for the quarterly unemployment rate at the local authority level and home lt prices, both obtained from the ONS.24 Since owners and non-owners will react differently to changes in house prices, we also interact house prices with the aforementioned LAU-age group home ownership rate. Tables 2 and 3 provide descriptive statistics of the data we use in our analysis. On average, there about 21 births per 1,000 women in each LAU-age-group quarter; across age-groups, the means range from 9.7 for 35-44 year olds to 28 for 25-34 year olds. 53 percent of women in each LAU-age-group are home owners and 36.2 percent of were on an adjustable rate during our analysis period. Across age-groups, adjustable rates shares range from 17 percent of 16-24 year olds to 49 percent for 35-44 year olds, which mostly reflects the fact the ownership rates increase as women get older. Median gross annual income for mortgagers at origination was £27,850. Most families had mortgage LTVs at origination above 85 percent. About 18 percent of families are in the first year of their mortgage, another 20 percent in the second, 13 percent in the third, and under 2 percent in the third or later year. 14 percent of families chose an ARM at origination, 22 percent of families had a refinanced mortgage and 31 percent had a purchase mortgage. House prices during this time averaged £180,460 and the unemployment rate averaged 7.9 percent. Our main identification assumption is that group-level exposure interacted with changes in themonetarypolicyrateisconditionallyexogenoustobirthrates. Violationstothisassumption would arise if there are omitted variables that coincide with the path of monetary policy and 23Incomes are calculated per capita by dividing by 2 when there is a joint mortgage. We convert gross incomes into 2008Q2 . 24The unemployment rate we use are the model-based estimates of unemployment based on the Annual Population Survey. The house price index we use is the UK HPI. Both were obtained from the ONS. 17

differentially affect the demand for children among groups with higher exposure rates. We probe the sensitivity of our results to this assumption by conducting a number of alternative specifications, such as replacing X with LAU-by-Age-group fixed effects and E with LAUla lt by-quarter fixed effects. We also present results from an event study analysis which does not uncover any evidence of a difference in pre-period birth rate trends by exposure level. A final threat to identification is possibility of reverse causality, wherein families who wish to have children manipulate their own exposure in anticipation of monetary policy. Recall that we construct our measure of exposure at the baseline to remove the possibility of endogenous lat home-buyingorrefinancing, andthatwecontrolforinitialadjustableorfixedratechoice, homeownership rates, and housing and mortgage tenure. Thus, the story of reverse causality is one in which families anticipate policy several years in advance and time the expiration of their fixed terms accordingly. Given that survey evidence suggests that less than 10 percent of families expected rates to fall in the months preceding the Great Recession, widespread behavior of this nature appears implausible. 5 Results 5.1 Main Results Table4presentstheresultsofestimatingEquation2.25 Column1isthemainspecification. The coefficient of interest, β which is the interaction between exposure and ∆bankrate 1 lat 2008Q2,t displays a point estimate of 0.0513, which is statistically significant at the 0.1 percent level. This implies that for family-groups with a 100 percent exposure rate (henceforth, families on an adjustable rate), a 1 percentage point decline in Bank Rate increases birth rates by 5 percent.26 At the mean exposure rate for the UK, this is equivalent to a 2 percent increase in birth rates. 25Coefficients on the control variables in the model are displayed in appendix table A1. 26Notethata100percentexposurerateisoutofsample. Still, itisinstructivetothinkaboutwhatthisimplies at the individual-level for families on an adjustable rate. This suggests that a 1 percentage point drop in the policy rate would increase the probability of conceiving a baby by 5 percent, or 0.1 percentage points at the mean birth rate of 2 percent. 18

Or, putdifferently, movingfromagroupinwhich25percentoffamiliesareonanadjustablerate to a group where 75 percent of families are on an adjustable rate increases the responsiveness of the birth rate to Bank Rate by 300 percent. The next two columns of Table 4 tests the robustness of those results to more flexible specifications of X and E . Column 2 of Table 4 adds LAU by age-group fixed effects to la lt Equation 1. This allows for the possibility that there is some omitted variable that is correlated with differences across groups in exposure rates and birth rates. For example, if older families in Manchester had higher exposure rates than younger families in London, and older families in Manchester had higher latent fertility rates than younger families in London for reasons not captured by X , this specification would allow for that possibility. Column 2 shows the point la estimates are little changed.27 Column 3 adds LAU by quarter fixed effects to Equation 2. These specification allows for thepossibilitythatdifferentLAUsmighthavebeenondifferentfertilitytrendsoverthequarters we study. In other words, this allows for the possibility of omitted variable bias arising from changes in some aspects of the local economy that are correlated with the path of monetary policy and fertility trends. However, column 3 show our point estimates are little changed when we include these controls. Finally, column 4 restricts the time period over which we estimate equation 2 to the six quarters from 2008Q3 to 2009Q4. As shown in Figure 4, all of the changes to Bank Rate occured between October 2008 and March 2009, at which point rates remained fixed at 0.5 percent. One motivation of our study is to identify how the initial monetary easing period affected birth rates in the short term. The specification in column 4 addresses that question. The coefficient on β is slightly attenuated from the estimate derived from the larger time frame 1 in column 1, but remains statistically significant at the 0.1 percent level. At the mean exposure 27Another possibility is owners and renters are differentially affected by monetary policy and our measure of exposure is capturing that effect. To allow for this, we ran a specification where we included an interaction term between bankrate and the LAU-age-group home ownership rate in addition all of the variables 2008Q2,t in equation 2. The coefficient on this term was small and insignificant, and the coefficient on our main point estimate was virtually unchanged. Results available upon request. 19

rate for that time period, the estimate in column 4 implies that the 4.5 percentage point drop in the policy rate led to a 6.8 percent increase in the birth rate. The effects we estimate are comparable to effects estimated in the literature on the effects of changes in income or prices on fertility. Our results imply that among families with an adjustable rate mortgage, a one percentage point decline in mortgage interest rates (which increases birth rates by 5 percent) reduces their mortgage payments £258 per quarter on average, equivalent to 1.7 percent their household income (see Table 1). For comparison, Milligan (2005) findsthata£900(inflatedtotoday’sdollars, andatthecurrentCAD-GBPexchangerate)child tax incentive leads to a 17 percent increase in birth rates, and Cohen et al. (2013) find that a 2 percent decline in income from a child tax benefit decreases the birth rate 10 percent. Kearney and Wilson (2018) find that a 3.8 percent increase in earnings associated with fracking in the U.S. increased the birth rate by 6 percent and Black et al. (2013) find that a 10 percent increase in income associated with Appalachian coal mining earnings due to changes in world energy prices led to a 7 percent increase in birth rates. In Section 2, we noted that whether a transitory shock will affect the timing of fertility depends on the extent to which families are able to borrow or save. Thus, we would expect the effects we estimate to vary depending on the extent to which families are liquidity constrained. In our data, we can proxy for liquidity constraints in a couple of ways, such as using income, loan-to-incomes (LTIs), and LTVs, since families who have lower incomes, or are more highly leveraged, are typically less able to borrow and save. Table 5 displays the results of splitting our sample according whether a group has above or below median income, LTIs or LTVs. The results of Table 5 indicate that indeed, families who are lower income, or had higher LTIs or LTVs are more responsive to a change in monetary policy. For example, columns 3 and 4 indicates that a 1 percentage point drop in Bank Rate leads to a 5 percent increase in births for families on an adjustable rate in the high LTI group, compared to a 2.7 percent increase for families in the low LTI group. Overall, this suggests liquidity constraints play an important role in explaining why families’ fertilty decisions are sensitive to changes in mortgage rates. 20

5.2 Robustness Checks We conduct a number of robustness checks on the sample construction and empirical specification, displayed in Table 6. The mortgage data includes information on the “lead borrower’s” age but not their sex. In our main specification we simply assume that is the mother’s age. But in practice it is likely to often be the father’s age if men are often listed as the lead borrower. Thus, column 1 presents the results where we match the mortgage information by the father’s age. In our data, around 94 percent of birth certificates list the father’s age, so this procedure should yield accurate results. The point estimates are slightly larger than the original point estimate, which could reflect attenuation bias in our original measure although it is well within the confidence interval of the original estimate. Column 2-3 presents estimates in which we conduct some alternative imputations of missing information in the mortgage data, described in more detail in the Appendix. In column 2 we presentestimateswhereweimputethenumberofage-groupbyLAUmortgagesthataremissing because they did not transact (either purchase or refinance) between 2005 and 2008. Recall these mortgages would virtually all be on an adjustable rate because their fixed periods (if present) would have mostly expired. We do this by isolating mortgages originated prior to 2005 from information on the stock of mortgages, which is available beginning in 2015. In this case, the point estimate is slightly smaller, but again well within the confidence interval of our original estimate. Column 3 present results where we vary how we impute information on the length of the initial period when that information is missing. As discussed in the appendix, in the main results we use a model to impute this information when it was not reported by the lender. In column 3 we instead simply drop all mortgages with missing initial periods and use a subsample of mortgages with known initial periods to calculate quarterly ARM shares.28 Columns 28As described in the appendix, omission of this information appears to lender-specific. The demographic characteristics of borrowers who used these lenders are very similar to those who did not, and the geographic distributionofthoselendersisverysimilartoothers. Thus,thesubsampleofborrowerswithouttermsappears to be similar to what might be achieved from a random sample. 21

3 shows the results are robust to these changes. Finally, columns 4-5 present results estimating the model in levels and unweighted. Again, the results are not sensitive to these changes in the specification. 5.3 Event Study Design The identification assumption underlying our empirical strategy is that in the absence of monetary policy, family-groups of varying levels of exposure would not have had different birth rates. To comprehensively examine pre-trends as well as dynamics over estimation period, we can estimate an alternative event-study style version of our estimating equation. To do so, we focus on exposure rates as of 2008Q2 (just prior to the beginning of our sample) and estimate the following regression for the time period 2005Q1 through 2011Q4: birthrate = β exposure ×θ l,a,t+3 t l,a,2008Q2 t +β E +β X +γ +α +θ +(cid:15) (3) 2 l,t 3 l,a a l t l,a,t wheretheleft-outcategoryofθ is2008Q2. Inthisspecification, X includesexposure . t l,a l,a,2008Q2 Figure 5 plots the coefficients on β and 99 percent confidence intervals, where the coefficients 1 can be interpreted as the conditional difference in trends in birth rates for LAU-age-groups with 100 percent exposure, relative to LAU-age-groups with no exposure.29 The coefficients on the pre-trends for 2005Q1 through 2008Q1 indicate that prior to the onset of monetary easing, there was essentially no difference in birth rate trends across exposure rates, and only 2 of the 12 coefficients are statistically different from zero. After monetary easing begins, however, there is a strong divergence in trends by exposure from the end of 2008 through to the end of 2013. This gap grows throughout the time period examined, presumably due to the fact that actual exposure rates are increasing (recall this measure is frozen in time, and actual exposure can only increase from there). The gap between groups with 100 percent and zero percent exposure 29We estimated this model in levels because it is easier to interpret the magnitude of the coefficients. For comparison, our main specification in levels is in column 5 of Table 5. 22

(as of 2008) grows from essentially zero in 2005-2008 to 15 additional births per 1000 women by 2011.30 Overall, Figure 5 provides no evidence that we can reject parallel trends by exposure rate in the three years leading up to monetary easing. 5.4 Extensions and Policy Implications 5.4.1 Would the UK have experienced a Great Recession “Baby Bust” without Mortgage Interest Rate Pass-through? Our results indicate that monetary policy increased the birth rate in the UK, but what might haveoccurredintheabsenceofmonetarypolicy? Thereisalongliteratureexamininghowbirth rates evolve over business cycles, which has tended to find that birth rates fall in recessions and rise during expansions. But as evidenced by Table 2, birth rates in the UK rose over the recessionary period we study. Although they are not our variables of interest in our main analysis,wecontrolforunemploymentandhouseprices. Thus,wecanconductasimilarexercise totherecentempiricalUSliteraturewhichexaminescross-sectionalvariationovertimetostudy the cyclicality of fertility (e.g., Dehejia and Lleras-Muney (2004); Dettling and Kearney (2014); Schaller (2016); Buckles et al. (2018)). To our knowledge, ours is the first study to do so for birth rates over the Great Recession in the UK. Column1ofTable7estimatesamodifiedversionofEquation2, whereweomitthemonetary policy and mortgage variables and estimate how local economic conditions correlate with birth rates in UK over the time period of our analysis. This specification is very close to Dettling and Kearney (2014). Column 1 of Table 6 displays the results of this analysis. The coefficients on economic conditions confirm the effects for the US housing boom period hold in the UK: local house price gains increase birth rates for owners and decrease birth rates for renters, and there is a negative but statistically insignificant effect of local unemployment rates on birth rates.31 30Of course, 100 percent exposure is out of sample as no group in our data has 100 percent exposure. At the mean exposure rate, the magnitude is 3.6 births per 1000 women. 31The estimates indicate that a £10,000 decrease in house prices leads to 15 percent decrease in births for owners, and a 7 percent increase for non-owners. At the USD-GBP exchange rate at that time, (Dettling and Kearney, 2014) find that a £10,000 increase in house prices lead to a 7.5 percent increase in births among 23

The second column of Table 7 adds the control variables available in our mortgage to the specification. To our knowledge, ours is the first paper to include as extensive a set of controls asweusehereinanexaminationofthecyclicalityoffertility. Ofparticularnotearethecontrols for housing tenure and leverage, which might bias previous estimates of the effects of economic conditions on birth rates if families who have been in their home longer or are more leveraged are more responsive to changes in house prices or unemployment. Still, column 2 confirms a sizable and statistically significant effect of house prices, though the effects on both owners and non-owners are attenuated from column 1. As before, unemployment rates do not enter significantly. There were large national changes in economic conditions during the period we study, which may swamp local changes, particularly relative to studies of larger and more heterogeneous countries like the US. Thus, the coefficients on θ , which can be interpreted as the national t change in birth rates (where the left out category is the third quarter of 2008), may be a better testofwhetherfertilityiscyclicalwithnationaleconomicconditions. Columns1and2,however, confirm that birth rates increased during the recession. The point estimates indicate birth rates were a statistically significant 2-6 percent higher in late 2008 through mid 2011 than in 2008Q3. Thus, without controlling for monetary policy, birth rates were strongly counter-cyclical at the national level over this time period of deteriorating economic conditions. Finally, in column 3 of Table 7 we add in our coefficient of interest on the effects of monetary policy. The coefficients on local house prices are essentially unchanged by the inclusion of the monetary policy exposure measure, indicating those variables have independent effects on birth rates. However, unlike column 1, the coefficients on θ have reversed sign and are now negative, t and statistically significant from late 2008 through mid 2011. In other words, controlling for the positive monetary policy effect on birth rates, the prevailing national trend during this period is that fertility rates would have been pro-cyclical. Next, we conduct a counterfactual analysis that investigates this notion more formally. In ownersanda3.6percentdeclineamongrenters. Wehavealsoexaminedalongertimeseriesofdataspanning 2005-2013 in the UK and found similar results, which are available upon request. 24

particular, we use the parameters from column 1 of Table 4 and quarterly means of all of our variables to predict the path of birth rates if monetary policy was unchanged. The solid line of Figure 6 indicates that birth rates are considerably lower in the counterfactual throughout the sample period and indeed appear to fall during the Great Recession period.32 In sum, this analysis appear to confirms the previous literature on pro-cyclical fertility, and suggests a Great Recession “baby bust” in the UK in the absence of monetary policy. 5.4.2 Mortgage Contract Flexibility and Monetary Policy Transmission to Birth Rates Acrossdevelopedcountries, thereisconsiderablevariationinmortgagecontracts, withcountries like the US, Germany, and France tending to have more fixed rate mortgages (FRMs) and countries like the UK, Australia, and Spain tending to have more ARMs. As we have discussed, our point estimates imply that the reason monetary policy has a positive stimulus effect on fertility is that a relatively large share of UK borrowers have ARMs. In this section, we use our model to consider how birth rates might have evolved if the share of borrowers with ARMs was lower (or higher) prior to monetary easing. To implement this, we distribute alternative baseline mean ARM shares across groups and over time using the distributions we view in our data, restricting the ARM share to be between 0 and 100 percent and the mean ARM share to line up with our alternative scenario.33 Figure 7 displays the results of this exercise. The vertical axis displays the change in the birth rate between the third quarter of 2008 and the third quarter of 2009 (we chose the same quarter of the year since birth rates display seasonality), and the horizontal access displays 32Of course, monetary policy also has an effect on economic conditions, and a more realistic simulation of the overallpathofbirthratesintheabsenceofmonetarypolicymightbeoneinwhichwealsofactorinthegeneral equilibrium effects of policy on house prices and unemployment. Because national changes in house prices andunemploymentareabsorbedintheθ terms,itisnotobvioushowwewouldimplementthisexercise. One t option might be to adjust the path of local house prices and unemployment rates only. In the Appendix we present the results of this exercise where we adjust each using the Bank of England’s forecasting model (as in (Pugh et al., 2018)) as inputs in our model. We distribute these across LAUs using the observed distribution of changes in house prices and unemployment rates. This does little to impact the time trend. 33We also adjust the initial contract choice to in a similar fashion. 25

the hypothetical share of mortgages which are ARMs as of the third quarter 2008 (holding the home ownership rate constant). The dashed line presents 95 percent bootstrapped confidence intervals. We can see that an increase in the ARM share is associated with an increase in the birth rate. Once about 25 percent of mortgages are ARMs, the change in the birth rate becomes positive. Once around 70 percent of mortgagers have ARMs, the birth rate rises nearly 5 percent.34 5.4.3 Comparing UK and US Birth Rates over the Great Recession The transmission of monetary policy is very different in the US and UK. In the US, more than 80 percent of mortgages are fixed rate mortgages (FRMs) with fixed terms of over 10 years, and the most popular mortgage is the 30 year FRM. In 2007, the share of mortgages with a fully floating rate was under 6 percent.35 There are limited pre-payment penalties, however, and the primary transmission mechanisms is voluntary refinance. In contrast, fewer than 5 percent of mortgages in the UK feature fixed terms longer than 10 years and refinance prior to the end of a fixed term is very costly, so the primary transmission mechanism is through ARMs.36 We can use these differences as an illustrative example of how differences in the transmission mechanism can affect birth rate trends across groups. As a case study, a comparison between the US and UK is particularly interesting because the countries faced reasonably similarly sized down-turns and the path of the central bank policy response was similar. For example, from trough-to-peak the US unemployment rate rose from 4.4 percent in 2007 to 9.5 percent in 2010, whereas the UK rate rose from 5.2 percent in 2007 to 8.4 percent in 2011. The Bank of England lowered interest rates 4.5 percentage points in late 2008 and early 2009, and the Federal Reserve Board lowered interest rates by the same amount in late 2007 and early 2008. 34Note that once we reach about 70 percent of mortgages being ARMs the line flattens out because age groups cannot have an arm share over 100 percent and few mortgages will have fixed terms which expire over the course of that year. 35Authors’ calculation from the 2007 Survey of Consumer Finances. 36Authors’ calculations based on 2007 Survey of Consumer Finances for the US and UK PSD data. 26

To examine trends in birth rates between the two countries we compiled data on quarterly birth rates by age-group in the US and UK from 2002-2015. The UK trends are based on the same data we used for our main analysis, but including additional years. The US trend data were obtained from microdata on conceptions resulting in live births for 2002-2015 from the US Vital Statistics Natality files, which was matched that to population data from the CDC SEER program.37 The gray connected lines in Figure 8 are seasonally adjusted quarterly birth rates in the US and UK over time for the same three age groups we used in our empirical analyses. As in Figure 1, for context, the shaded gray areas indicate the time period in which unemployment increased from its trough-to-peak and the dashed black line is the path of each country’s policy rate. A couple of baseline characteristics of different age-groups lead to predictions about the relative differences in the birth rates across the two countries. First, as indicated by Table 2, 16-24 year olds in the UK have relatively low rates of exposure because home ownership rates arelow. Thissuggeststhatfertilityratesamong16-24yearoldsshouldbesimilaracrossthetwo countries. Second, although 25-34 year olds and 35-44 year olds have relatively higher exposure rates in both countries, 25-34 year olds tend to be more leveraged than 35-44 year olds in both countries.38 As evidenced by Table 1, in the UK this led to a larger relative drop in mortgage payments (as a share of the mortgage payment or as a share of income) for 25-34 year olds. In the US, higher loan-to-value ratios imply there is a higher propensity to become underwater when house prices fall, which would have restricted active refinancing into a lower mortgage rate. This suggests the trends in the two countries should diverge for both age-groups, but more so for 25-34 year olds. Figure 8 indicates the predictions above are born out in the data and different age-groups had differing fertility patterns during the recessions in the US and UK. The middle panel of Figure 8 is the most striking: 25-34 year olds initially had similar trends in the two countries. 37Data was downloaded from www.nber.org 38Forexample,the2007USSurveyofConsumerFinancesindicatesthat50percentof25-34yearoldsarehome owners, with an average LTV of 67. Among 16-24 year olds, only 12 percent are owners and among 35-44 year olds, 83 percent are owners, with an average LTV of 55. Authors’ calculations. 27

However, in the UK that trend reverses once monetary easing begins, at which point the birth rate subsequently rises. In the US, the downward trend continues. As predicted above, the pattern for 35-44 year olds is similar to 25-34 year olds, albeit somewhat more muted, and the broad pattern for 16-24 year olds is nearly identical in both countries. This highlights how differences in mortgage contract design can lead to differences in the economic incidence of monetary policy across groups. 6 Conclusion This paper has examined how monetary policy affects fertility decisions. We exploit variation across groups and over time in the share of families whose mortgages were eligible for a rate adjustment, and find that every 1 percentage point decline in the policy rate increases birth rates by 5 percent for families with an adjustable-rate mortgage. On average for the UK, a 1 percentage point decline in the policy rate increases birth rates by 2 percent. Overall, our paper deepens our understanding of the transmission mechanism of monetary policy to household behavior and the role of mortgage-structure rigidities for the macroeconomy. We have focused on differences across space and time in the share of mortgages with an adjustable rate in order to estimate the size of mortgage rate pass-through. But in other countries, there are other aspects of mortgage contracts that allow for interest rate pass-through, for example, lower pre-payment penalties on fixed rate mortgages so that borrowers can refinance (Beraja et al., 2018; Wong, 2016). That said, when mortgage refinance is the main transmission mechanism an asymmetry can arise between housing boom and bust cycles, since in order to refinance, the borrower must have sufficient equity to be approved for a new mortgage. With adjustable rate resets, re-approval is not necessary to obtain a lower mortgage rate. At face value, it is plausible then that the prevalence of fixed rate mortgages and negative equity in the US could offer one explanation for the divergence in birth rate trends between the US and UK (Figure 1). Though we leave a formal analysis of whether our results might hold in other 28

countries and time periods to future work, our descriptive comparisons with the US suggest that if more families had been able to obtain a lower interest rate, the U.S. might not have experienced as severe of a “baby bust” in the Great Recession. Our paper is written within the paradigm of an active literature on the cyclicality of fertility, which is a literature about the timing of fertility decisions. Ultimately, the recentness of the episode we study makes it difficult to ascertain with certainty whether the effects we have estimated represent changes in the timing of births or changes in the number of births a woman will have over the course of her life. However, there is suggestive evidence based on interest rate expectations and time trends in older women’s birth rates supporting the plausibility of a change in completed fertility. Birth rates feature prominently in standard models of economic growth. Thus, a change in completed fertility due to rate pass-through would imply a new supply-side mechanism via which monetary policy can affect long run economic outcomes. Still, short-term fluctuations in birth rates can also have important effects on the economy. Families who move forward their child-bearing plans will also move forward any associated consumer spending. And year-to-year fluctuations in cohorts and class sizes can have important implications for educational attainment and future labor market outcomes. Overall, our paper suggests monetary policy can have spillover effects on a host of economic and social outcomes. References Ahearne, Alan G, John Ammer, Brian M Doyle, Linda Kole, and Robert F Martin, “Monetary policy and house prices: a cross-country study,” FRB International Finance Discussion Paper, 2005, (841). Angrist, Joshua and William Evans, “ChildrenandTheirParents’LaborSupply: Evidence from Exogenous Variation in Family Size,” American Economic Review, 1998, 88 (3), 450–77. Angrist, Joshua D and Victor Lavy, “Using Maimonides’ rule to estimate the effect of class size on scholastic achievement,” The Quarterly journal of economics, 1999, 114 (2), 533–575. Auclert, Adrien, “Monetary policy and the redistribution channel,” American Economic Review, 2019, 109 (6), 2333–67. 29

Autor, David, David Dorn, Gordon Hanson et al., “When Work Disappears: Manufacturing Decline and the Falling Marriage-market Value of Young Men,” American Economic Review: Insights, 2018. Bank of England, “Bank of England/TNS Inflation Attitudes Survey,” Bank of England website 2018. Becker, Gary S, “An economic analysis of fertility,” in “Demographic and economic change in developed countries,” Columbia University Press, 1960, pp. 209–240. and H Gregg Lewis, “On the Interaction between the Quantity and Quality of Children,” Journal of political Economy, 1973, 81 (2, Part 2), S279–S288. Beraja, Martin, Andreas Fuster, Erik Hurst, and Joseph Vavra, “RegionalHeterogeneity and the Refinancing Channel of Monetary Policy,” The Quarterly Journal of Economics, 2018, p. qjy021. Berg, Kimberly, Chadwick Curtis, Steven Lugauer, and Nelson C Mark, “Demographics and Monetary Policy Shocks,” NBER Working Paper, 2019, (w25970). Bhutta, Neil, Daniel R Ringo et al., “The Effect of Interest Rates on Home Buying: Evidence from a Discontinuity in Mortgage Insurance Premiums,” Technical Report, Board of Governors of the Federal Reserve System (US) 2017. Black, Dan A, Natalia Kolesnikova, Seth G Sanders, and Lowell J Taylor, “Are Children “Normal”?,” The review of economics and statistics, 2013, 95 (1), 21–33. Buckles, Kasey, Daniel Hungerman, and Steven Lugauer, “Is Fertility a Leading Economic Indicator?,” Technical Report, National Bureau of Economic Research 2018. Calza, Alessandro, Tommaso Monacelli, and Livio Stracca, “Housing Finance and Monetary Policy,” Journal of the European Economic Association, 2013, 11 (suppl_1), 101– 122. Cava, Gianni La, Helen Hughson, and Greg Kaplan,“TheHouseholdCashFlowChannel of Monetary Policy,” Dec 2016, (rdp2016-12). Chakraborty, Chiranjit, Mariana Gimpelewicz, and Arzu Uluc, “A Tiger by the Tail: EstimatingtheUKMortgageMarketVulnerabilitiesfromLoan-LevelData,”BankofEngland working papers 703, Bank of England December 2017. Cloyne, James, Kilian Huber, Ethan Ilzetzki, and Henrik Kleven,“TheEffectofHouse Prices on Household Borrowing: A New Approach,” American Economic Review, 2019, 109 (6), 2104–2136. 30

Cohen, Alma, Rajeev Dehejia, and Dmitri Romanov,“Financialincentivesandfertility,” Review of Economics and Statistics, 2013, 95 (1), 1–20. Cumming, Fergus, “Mortgages, Cash-flow Shocks and Local Employment,” Bank of England working papers 773, Bank of England Dec 2018. Dehejia, Rajeev and Adriana Lleras-Muney, “Booms, Busts, and Babies’ Health,” The Quarterly Journal of Economics, 2004, 119 (3), 1091–1130. 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, 2014, 110, 82–100. Flodén, Martin, Matilda Kilström, Jósef Sigurdsson, and Roine Vestman, “Household Debt and Monetary Policy: Revealing the Cash-Flow Channel,” Technical Report 2017. Guren, Adam M, Arvind Krishnamurthy, and Timothy J McQuade, “Mortgagedesign in an Equilibrium Model of the Housing Market,” Technical Report, National Bureau of Economic Research 2018. Hotz, V Joseph, Jacob Alex Klerman, and Robert J Willis, “The economics of fertility in developed countries,” Handbook of population and family economics, 1997, 1, 275–347. Kearney, Melissa S and Riley Wilson, “MaleEarnings, MarriageableMen, andNonmarital Fertility: Evidence from the Fracking Boom,” Review of Economics and Statistics, 2018, 100 (4), 678–690. Korenman, Sanders and David Neumark, “Cohort crowding and youth labor markets: A cross-national analysis,” in Blanchflower and Freeman, eds., Youth Employment and Unemployment in Advanced Countries, University of Chicago Press, 2000, pp. 57–105. Lisack, Noëmie, Rana Sajedi, and Gregory Thwaites, “Demographic Trends and the Real Interest Rate,” Technical Report 2017. Liverpool Victoria, “Cost of a Child,” LV website 2014. Lovenheim, Michael F and Kevin J Mumford, “Do Family Wealth Shocks affect Fertility Choices? Evidence from the Housing Market,” Review of Economics and Statistics, 2013, 95 (2), 464–475. Maggio, Marco Di, Amir Kermani, Benjamin J. Keys, Tomasz Piskorski, Rodney Ramcharan, Amit Seru, and Vincent Yao, “Interest Rate Pass-Through: Mortgage Rates, Household Consumption, and Voluntary Deleveraging,” American Economic Review, November 2017, 107 (11), 3550–88. 31

Martins, Nuno and Ernesto Villanueva, “Does high cost of mortgage debt explain why young adults live with their parents?,” Journal of the European Economic Association, 2009, 7 (5), 974–1010. Milligan, Kevin, “Subsidizingthe stork: Newevidence ontax incentives andfertility,” Review of Economics and statistics, 2005, 87 (3), 539–555. Pugh, Alice, Philip Bunn, and Chris Yeates, “The distributional impact of monetary policy easing in the UK between 2008 and 2014,” 2018. Rachel, Lukasz and Thomas D Smith, “Are Low Real Interest Rates Here to Stay?,” International Journal of Central Banking, 2017, 13 (3), 1–42. Rubio, Margarita, “Fixed- and Variable-Rate Mortgages, Business Cycles, and Monetary Policy,” Journal of Money, Credit and Banking, 2011, 43 (4), 657–688. Schaller, Jessamyn, “Booms, Busts, and Fertility Testing the Becker Model using Genderspecific Labor Demand,” Journal of Human Resources, 2016, 51 (1), 1–29. Sommer, Kamila, “Fertility choice in a life cycle model with idiosyncratic uninsurable earnings risk,” Journal of Monetary Economics, 2016, 83, 27–38. Stapleton, David C and Douglas J Young, “Educational attainment and cohort size,” Journal of Labor Economics, 1988, 6 (3), 330–361. Wong, Arlene, “Transmission of monetary policy to consumption and population aging,” Technical Report 2016. 32

Figures and Tables Figure 1: Trends in Birth Rates and Monetary Policy in the UK and US United Kingdom United States Solid lines are seasonally adjusted quarterly birth rates by age-group dated to the quarter of conception and expressed per 1,000 women. The dashed line shows the path of Bank rate (left column) and the Federal Funds Rate (right column). The gray bar indicates the period in which unemployment increased from its trough to its peak in each country during the Great Recession. Sources: ONS and Bank of England (left column) and NCHS, Census, BLS, and Federal Reserve Board (right column). 33

Figure 2: The Path of Interest Rates Source: Bank of England and own calculations. 34

Figure 3: Data and Empirical Strategy Timeline Mortgage originations Conceptions Q1 Q2 Q3 Q1 Q1 Q1 Q1 2005 2008 2008 2009 2010 2011 2012 Birth certificates Bank Rate 8 Oct 6 Nov 4 Dec 8 Jan 5 Feb 5 Mar 4.5% 3.0% 2.0% 1.5% 1.0% 0.5% Figure shows the time-line for the data and central specification. We construct an estimate of the stock of mortgages using the flow of new loans between April 2005 and June 2008. The experiment begins in 2008Q3 when Bank Rate is reduced from 4.5% to 0.5% over the next five months. We measure the number of conceptions in each quarter from 2008Q3 to 2011Q2, which appear in the birth registry around nine months later. 35

Figure 4: Exposure Rates as of July 2008 Proportion of ARMs among all ages, percent 15.5 27.4 30.3 33.6 43.3 Source: PSD, ONS and own calculations. Figure shows proportion of families in each LAU with adjustable-rate mortgages as of July 2008. Color breaks denote quartiles and 343 local authorities are shown. 36

Figure 5: Event Study Figure plots coefficients and 99 percent confidence intervals on β based on estimating equation 3. t 37

Figure 6: Counterfactual Simulation of Birth Rates with and without Monetary Policy The dotted lines show the actual path of birth rates and the dashed line shows the predicted path with no change in Bank Rate based on estimates in Table 4. The capped bars indicate 95 percent confidence intervals. Birth rates are seasonally adjusted. 38

Figure 7: Counterfactual Simulation of Birth Rates by Share of Mortgages that are ARMs The black line shows the predicted change in conceptions between the third quarter of 2008 and the third quarter of 2009 based on the results in Table 4. The horizontal axis displays the assumption about the share of mortgages that were an ARM at the end of the second quarter of 2008, which we use to simulate the change in birth rates, as described in the text. The dashed lines show bootstrapped 95 percent confidence intervals. 39

Figure 8: Trends in Birth Rates and Monetary Policy in the UK and US by Age-Group United Kingdom United States Solid lines are seasonally adjusted quarterly birth rates by age-group dated to the quarter of conception and expressed per 1,000 women. The dashed line shows the path of Bank rate (left column) and the Federal Funds Rate (right column). The gray bar indicates the period in which unemployment increased from its trough to its peak in each country during the Great Recession. Sources: ONS and Bank of England (left column) and NCHS, CDC, BLS, and Federal Reserve Board (right column). 40

Table 1: Simulated Payment Changes for families with an Adjustable Rate Variable Mean Median SD p25 p75 Panel 1: Impact of a 100bp fall in mortgage interest rates Quarterly payment decrease (£) 257.93 178.53 313.11 103.75 306.25 Change relative to payment (%) 9.50 8.08 5.14 5.58 12.50 Change relative to income (%) 1.72 1.53 1.02 1.05 2.20 Age group 16-24 Quarterly payment decrease (£) 185.76 155.25 140.92 109.14 224.69 Change relative to payment (%) 9.77 8.74 4.05 6.57 11.71 Change relative to income (%) 1.97 1.74 1.02 1.33 2.39 Age group 25-34 Quarterly payment decrease (£) 242.77 190.16 223.67 123.75 294.76 Change relative to payment (%) 9.61 8.57 4.40 6.37 11.46 Change relative to income (%) 1.85 1.66 0.94 1.22 2.29 Age group 35-44 Quarterly payment decrease (£) 276.45 188.42 338.30 108.69 328.76 Change relative to payment (%) 9.34 8.00 4.94 5.60 11.50 Change relative to income (%) 1.70 1.52 0.98 1.05 2.17 Panel 2: Impact of a 450bp fall in mortgage interest rates Quarterly payment decrease (£) 1130.31 765.55 1406.39 446.78 1228.28 Change relative to payment (%) 41.63 33.58 23.59 24.01 56.71 Change relative to income (%) 7.51 6.60 4.53 4.50 9.52 Age group 16-24 Quarterly payment decrease (£) 806.12 662.22 630.80 464.60 970.91 Change relative to payment (%) 42.39 36.62 18.72 27.93 50.08 Change relative to income (%) 8.51 7.52 4.49 5.69 10.67 Age group 25-34 Quarterly payment decrease (£) 1053.56 810.25 1001.81 526.33 1275.00 Change relative to payment (%) 41.72 35.40 20.36 27.20 47.46 Change relative to income (%) 7.97 7.09 4.16 5.22 9.83 Age group 35-44 Quarterly payment decrease (£) 1208.29 804.77 1518.59 464.24 1425.00 Change relative to payment (%) 40.79 33.06 22.74 24.07 51.51 Change relative to income (%) 7.40 6.50 4.38 4.49 9.32 Data Source is PSD. Simulation considers families on the initial period of an adjustable-rate mortgage (full interest rate pass-through) or on the SVR (two thirds pass-through). 41

Table 2: Summary Statistics by Quarter and Age Variable Mean SD Min Max Panel 1: Birth and exposure rates over time 2008Q3 Birth rate (per 1,000 women) 20.5 8.39 2.79 39.1 Exposure (% households) 20.9 8.65 3.06 44.4 2009Q3 Birth rate (per 1,000 women) 20.9 8.62 3.08 37.1 Exposure (% households) 34.2 12.5 5.34 61.0 2010Q3 Birth rate (per 1,000 women) 21.3 8.83 2.86 37.0 Exposure (% households) 43.1 14.9 7.11 73.1 Panel 2: Birth, ownership and exposure rates across age groups Age group 16-24 Birth rate (per 1,000 women) 14.7 4.03 3.08 28.2 Ownership rate (% households) 27.1 8.20 8.31 56.0 Exposure (% households) 17.4 7.05 3.06 50.4 Age group 25-34 Birth rate (per 1,000 women) 27.9 3.74 14.0 39.5 Ownership rate (% households) 58.3 10.7 29.0 85.3 Exposure (% households) 39.8 11.6 12.8 74.0 Age group 35-44 Birth rate (per 1,000 women) 9.70 3.01 2.44 19.4 Ownership rate (% households) 70.3 11.2 33.8 90.7 Exposure (% households) 49.2 12.5 16.9 77.1 Data Sources are ONS, PSD and 2001 Census. Statistics are weighted by the number of births in each cell. 42

Table 3: Summary Statistics Variable Mean SD Variable Mean SD Mortgage Variables (PSD) Interest Rate Pass-through Loan to Value Ratio (% households) Exposure (%) 36.2 15.7 Under 60% 13.7 10.2 Exposure (%) x Bank Rate (%) 1.48 0.86 60-65% 2.87 1.63 65-70% 3.18 1.63 Type of Mortgage (% households) 70-75% 4.18 1.96 ARM at Origination 14.3 7.16 75-80% 4.11 1.69 First Time Buyer 14.6 5.44 80-85% 4.62 1.75 Re-mortgager 21.6 13.3 85-90% 6.30 2.02 Mover 16.9 8.08 90-95% 7.70 2.43 Years since origination (% households) 95-100% 5.39 2.37 Less than 1 year 17.7 6.64 Over 100% 1.16 0.63 1-2 years 20.0 7.10 Mortgagers’ personal income (£1,000s) 2-3 years 13.1 4.51 5th percentile 14.7 3.64 More than 3 years 2.21 0.72 25th percentile 21.0 6.27 Loan to Income Ratio (% households) 50th percentile 27.8 9.99 Under 200% 7.02 4.90 75th percentile 38.7 17.5 200-250% 6.93 3.45 Years until loan paid off (% households) 250-300% 10.4 3.98 Under 15 years 2.81 3.30 300-350% 11.3 3.72 15-20 years 6.99 6.37 350-400% 8.91 3.04 20-25 years 28.8 10.9 Over 400% 8.54 3.15 More than 25 years 14.5 7.31 Other Variables (ONS and Census) Birth rate (per 1,000 women) 21.1 8.65 Median House Price (£1000s) 18.46 8.15 Female higher qualification (%) 23.4 12.3 Unemployment Rate (%) 7.91 2.46 Ownership rate (%) 53.1 18.5 Data Sources are ONS, PSD and 2001 Census. Stastistics are weighted by the number of births in each cell. 43

Table 4: Monetary Policy and Birth Rates (1) (2) (3) (4) Dep Var: ln(birth rate ) lat+3 Exposure x Bank Rate 0.0513*** 0.0517*** 0.0637*** 0.0493*** lat 2008Q2,t (0.0042) (0.0042) (0.0056) (0.0082) LAU Fixed Effects x x Age Fixed Effects x x x Quarter Fixed Effects x x x LAU-Age Controls (X ) x x la LAU-Quarter Controls (E ) x x x lt LAU x Age Fixed Effects x LAU x Quarter Fixed Effects x Only 2008q3-2009q4 x N 12,348 12,348 12,348 6,174 Estimated according to Equation 2, as described in the text. Data Sources are ONS, PSD and 2001 Census. StandardserrorsadjustedforclusteringatLAUlevel. Regressionsareweightedbythenumberofbirthsineach cell. * p<0.05 ** p<0.01 *** p<0.001. 44

Table 5: Heterogeneity Analysis (1) (2) (3) (4) (5) (6) Below Above Above Below Above Below Median income Median income Median LTI Median LTI Median LTV Median LTV Dep Var: ln(birth rate ) lat+3 Exposure x Bank Rate 0.0708*** 0.0557*** 0.0514*** 0.0267*** 0.0585*** 0.0433*** lat 2008Q2,t (0.0072) (0.0077) (0.0056) (0.0075) (0.0055) (0.0116) N 6,180 6,168 6,180 6,168 6,180 6,168 Data Sources are ONS, PSD and 2001 Census. Standards errors adjusted for clustering at LAU level. Regressions are weighted by the number of births in each cell. * p<0.05 ** p<0.01 *** p<0.001. 45

Table 6: Robustness Checks (1) (2) (3) (4) (5) Father’s Impute Exclude Levels Unweighted Mortgage Missing Stock Missing Periods Dep Var: ln(birth rate ) lat+3 Exposure x Bank Rate 0.0525*** 0.0513*** 0.0449*** 0.8063*** 0.0473*** lat 2008Q2,t (0.0042) (0.0042) (0.0038) (0.0734) (0.0047) N 12,348 12,348 12,348 12,348 12,348 Estimated according to Equation 2, as described in the text. Data Sources are ONS, PSD and 2001 Census. Standards errors adjusted for clustering at LAU level. Regressions are weighted by the number of births in each cell. * p<0.05 ** p<0.01 *** p<0.001. 46

Table 7: Effect of Unemployment Rates and House Prices on Birth Rates (1) (2) (2) Dep Var: ln(birth rate ) lat+3 House price x Own 0.1113*** 0.0579*** 0.0576*** lt la (0.0100) (0.0112) (0.0113) House price -0.0506*** -0.0272*** -0.0262*** lt (0.0053) (0.0052) (0.0054) Unemployment rate -0.0008 -0.0002 0.0014 lt (0.0019) (0.0020) (0.0021) Exposure x Bank Rate 0.0513*** lat 2008Q2,t (0.0042) 2008Q4 0.0348*** 0.0278*** 0.0072 (0.0049) (0.0046) (0.0049) 2009Q1 0.0257*** 0.0138* -0.0408*** (0.0074) (0.0065) (0.0080) 2009Q2 0.0157 0.0030 -0.0681*** (0.0080) (0.0068) (0.0091) 2009Q3 0.0215** 0.0109 -0.0692*** (0.0069) (0.0063) (0.0092) 2009Q4 0.0532*** 0.0453*** -0.0424*** (0.0056) (0.0053) (0.0089) 2010Q1 0.0454*** 0.0388*** -0.0548*** (0.0055) (0.0052) (0.0092) 2010Q2 0.0023 -0.0036 -0.1012*** (0.0050) (0.0048) (0.0090) 2010Q3 0.0222*** 0.0181*** -0.0828*** (0.0049) (0.0048) (0.0099) 2010Q4 0.0474*** 0.0425*** -0.0609*** (0.0051) (0.0048) (0.0097) 2011Q1 0.0227*** 0.0157** -0.0899*** (0.0057) (0.0056) (0.0104) 2011Q2 0.0100 0.0022 -0.1066*** (0.0058) (0.0057) (0.0104) LAU Fixed Effects x x x Age Fixed Effects x x x Education and Income x x x Housing Tenure and Leverage x x N 12,348 12,348 12,348 Estimated according to Equation 2, as described in the text. Data Sources are ONS, PSD and 2001 Census. StandardserrorsadjustedforclusteringatLAUlevel. Regressionsareweightedbythenumberofbirthsineach cell. * p<0.05 ** p<0.01 *** p<0.001. 47

Appendix (for online publication) 6.1 Data Appendix 6.1.1 PSD data This section describes how we construct our measures of interest rate exposure based on the PSD mortgage origination data. The PSD is administrative data, reported by lenders to the Financial Conduct Authority for all but a few primary residential mortgages in the United Kingdom. To minimise measurement error we remove mortgages with a mortgage term of less than 5 years, a loan-to-income ratio greater than 20 or a loan-to-value ratio of greater than 200. This removes about 3.7 percent of the data. We also drop business mortgages (only 0.25 percent of mortgages), as well those held by a social buyer, not known, or other (another 4.8 percent of mortgages). In order to estimate the stock of mortgages in the summer of 2008 we must remove superseded mortgages from the origination data to prevent double counting. Removing previous refinancing transactions requires matching birth dates and post codes and keeping the latest transactionbeforeJuly2008. Sincepostcodestypicallycoveraround15households, thisprocess rarely yields false matches (which arise in a small number of cases when the designated primary borrower changes between refinancing events). Removing loans that were paid off when a home owner moved houses is a little more involved. We do this using a three-way match on the (1) birth date, (2) transaction date and (3) post code following the steps outlined in Chakraborty et al. (2017). Because the data relies on lender reporting, there will sometimes be gaps and errors in the data due to mis-reporting or fields that are optional. For example, some mortgages will not include the initial interest rate or the length of initial period governing the behavior of the interest rate. Note that very few mortgages are missing information about whether the interest rate is fixed or adjustable (less than 0.5 percent). Oneimportantmodificationwemustmakeiswhenthelengthoftheinitialperiodismissing. This is important in our model because we use it to determine when a fixed rate mortgage will reset to an adjustable rate. For a mortgage originally on an adjustable rate, the initial period will determine when the mortgage reverts to the Standard Variable Rate (also an adjustable rate, but usually with a higher level and slightly slower pass-through from policy rates). 52 percent of our final sample of fixed-rate mortgages are missing the initial period (this field was optional before 2015). When it is missing, in our main specifications we impute it using a model on the observed distribution depending on whether the mortgage was originally on a fixed or adjustable rate at origination. The models are an ordered probit model, where the initial period can be 1, 2, 3, 4, or 5+ years. Included in the model are lender fixed effects, borrower age fixed effects, borrowing income category, the property value category, and an LTV category. We then 48

use the predicted probabilities arising from this model to assign borrowers’ their most likely initial period length (for all types of mortgage). Alternatively,inourrobustnesschecks,wesimplydropthemissingdataandtreatthesample of mortgages that have information on the missing period as though it were a random sample. There is good reason to think this is a fairly safe assumption. First, inspection indicates that the reporting of initial periods was a lender-specific decision. The structure, competitiveness andnationalreachofUKmortgagelendersduringthistimemeantthatthesampleofmortgages with a known initial period has very similar characteristics to that of the full sample. Table A3 considers the full sample used in our main analysis in the first column. In the second column we construct the set of mortgages that have a known initial period. In order to match the proportion of fixed and adjustable-rate mortgages in the main sample as of July 2008 we take all of the known adjustable-rate mortgages and a random sample of the known FRMs. Almost all categories are similar across the samples. In particular, the two samples have a very similar initial-period, geographical and mortgage-purpose distribution. Table A4 shows that a more detailed breakdown of mortgage characteristics across samples shows considerable similarities. Together these tables suggest our initial-period modelling specification is unlikely to be important for our main results. Once we have our estimated stock of mortgages and their initial periods, we proceed by simulating the loan-level evolution up until July 2008, and then quarter-by-quarter using the information we have regarding the mortgage characteristics at origination and the likely behavior of interest rates at each point in time. The primary variable of interest for each mortgage is whether the interest rate was fixed or adjustable in each quarter. A mortgage could be adjustable if it was specified at origination or the mortgage had run beyond the initial and therefore reverted to the Standard Variable Rate. For context, it is also useful to measure in money terms how mortgage payments changed over time. Quarterly mortgage payments are a function of the mortgage balance outstanding, the interest rate and the mortgage term and can be estimated using standard annuity calculations. Where we know the mortgage was a so-called interest only mortgage, the payment reduces to the balance outstanding multplied by the quarterly interest rate. For some mortgages the initial interest rate is missing (about 37 percent). Though we do not use this in our main analysis, we must use it to simulate the payment changes in Table 1. Because the UK mortgage market was very competitive during the mid-2000’s, we can run a simple model based on the date of origination and loan-to-value to predict the interest rate using observed data. 6.1.2 Car-buying among new parents in the LCFS The Living Costs and Food (LCFS) is an annual survey carried out by the ONS to gather information about the expenditure on goods and services. The LCFS uses multi-stage stratified random sampling but does not involve re-sampling so we cannot track households across time. 49

Respondents keep a detailed diary of spending for two weeks but in Table A2 we use the questions that ask families about the past twelve months. We use the surveys from 2006 to 2017. Table A2 reports OLS regressions where the dependent variable is a categorical variable based on whether or not the households bought a new or used car in the last year. The sample consists of mortgaged and all-tenure households with one child under 2 or no children under 18 at all. We drop all households that do not have an adult of child-bearing age (defined here as under 45) and households with more than one child. 50

6.2 Additional Tables and Figures Table A1: Coefficients on Control Variables (1) (2) (3) (4) Birth rate Birth rate Birth rate Birth rate Age 25-34 0.4284*** 0.0000 0.4420*** 0.3850*** (0.0962) (.) (0.1169) (0.0968) Age 35-44 -0.2243 0.0000 -0.2154 -0.3044* (0.1430) (.) (0.1729) (0.1423) 2008Q4 0.0072 0.0033 0.0069 (0.0049) (0.0045) (0.0052) 2009Q1 -0.0408*** -0.0479*** -0.0403*** (0.0080) (0.0070) (0.0089) 2009Q2 -0.0681*** -0.0764*** -0.0678*** (0.0091) (0.0080) (0.0111) 2009Q3 -0.0692*** -0.0763*** -0.0699*** (0.0092) (0.0086) (0.0126) 2009Q4 -0.0424*** -0.0475*** -0.0424** (0.0089) (0.0086) (0.0135) 2010Q1 -0.0548*** -0.0596*** (0.0092) (0.0090) 2010Q2 -0.1012*** -0.1055*** (0.0090) (0.0089) 2010Q3 -0.0828*** -0.0860*** (0.0099) (0.0099) 2010Q4 -0.0609*** -0.0640*** (0.0097) (0.0097) 2011Q1 -0.0899*** -0.0947*** (0.0104) (0.0103) 2011Q2 -0.1066*** -0.1116*** (0.0104) (0.0103) Female degree -0.0019 -0.0020 -0.0020 lt (0.0026) (0.0032) (0.0027) Ownership rate 3.1718 2.8724 2.2525 la (2.5907) (3.1258) (2.7246) ARM at origination 0.0397*** 0.0395*** 0.0397*** la (0.0075) (0.0091) (0.0077) First time buyer -0.7614 -0.7326 -0.8265 la (0.6816) (0.8262) (0.6977) Continued on next page 51

Table A1 – Continued from previous page (1) (2) (3) (4) Birth rate Birth rate Birth rate Birth rate Remortgager 0.1122 0.0775 -0.0474 la (0.6890) (0.8341) (0.7126) 1-2y seasoned -4.1315*** -4.0871*** -3.9013*** (0.9516) (1.1479) (0.9714) 2-3y seasoned -3.4107** -3.3937** -3.3574** (1.0458) (1.2658) (1.0717) >3y seasoned -1.6250 -1.5998 -2.2317 (2.1533) (2.5942) (2.2368) LTI <2 1.6535 1.9264 1.4557 (1.2352) (1.5130) (1.2946) LTI 2-2.5 5.0150*** 5.2751*** 4.6932*** (1.2944) (1.5833) (1.3438) LTI 2.5-3 3.4415* 3.6629* 3.0754* (1.3914) (1.6938) (1.4596) LTI 3-3.5 5.1673*** 5.2862** 5.1211*** (1.3596) (1.6338) (1.4313) LTI 3.5-4 4.1125** 4.1797* 4.1888** (1.4097) (1.6955) (1.4535) LTV <60% -1.3076 -1.4125 -0.1451 (2.3870) (2.8846) (2.4730) LTV 60-65% 0.9056 1.0082 1.7436 (2.8322) (3.4232) (2.9022) LTV 65-70% -0.1095 -0.0557 1.1507 (2.8268) (3.4079) (2.9070) LTV 70-75% -3.9152 -3.8846 -2.4613 (2.7308) (3.2843) (2.8479) LTV 75-80% -2.2619 -2.2832 -1.3858 (2.8071) (3.3854) (2.9011) LTV 80-85% -3.9062 -3.8431 -3.3280 (2.6605) (3.2070) (2.7754) LTV 85-90% -2.3429 -2.2864 -1.4096 (2.4987) (3.0056) (2.5774) LTV 90-95% -3.6044 -3.5846 -2.3963 (2.5727) (3.1010) (2.6612) LTV 95-100% -1.0038 -0.9937 0.1054 (2.6239) (3.1588) (2.7322) 5th percentile income -0.0481 -0.0631 -0.1242 Continued on next page 52

Table A1 – Continued from previous page (1) (2) (3) (4) Birth rate Birth rate Birth rate Birth rate (0.2047) (0.2482) (0.2085) 25th percentile income 0.1000 0.0778 0.1205 (0.2192) (0.2705) (0.2182) 50th percentile income 0.1724 0.1725 0.2136 (0.1399) (0.1698) (0.1429) 75th percentile income -0.0439 -0.0454 -0.0522 (0.0497) (0.0610) (0.0528) Share <15 year term -7.1893*** -7.0992*** -6.9395*** (0.9551) (1.1541) (0.9873) Share 15-20 year term -7.0150*** -6.9305*** -7.0424*** (0.9773) (1.1820) (1.0069) Share 20-25 year term -3.8872*** -3.9123*** -3.6939*** (0.4971) (0.6021) (0.5050) House price x Own rate 0.0576*** 0.0266*** 0.0636*** 0.0577*** lt la (0.0113) (0.0040) (0.0154) (0.0131) House price -0.0262*** -0.0126*** 0.0000 -0.0254*** lt (0.0054) (0.0025) (.) (0.0064) Unemployment rate 0.0014 0.0021 0.0000 0.0046 lt (0.0021) (0.0021) (.) (0.0031) Constant 2.4872*** 2.8823*** 2.0200*** 2.4240*** (0.2719) (0.0266) (0.2615) (0.2942) N 12,348 12,348 12,348 6,174 Estimated according to Equation 2. Data Sources are ONS, PSD and 2001 Census. Standards errors adjusted for clustering at LAU level. Regressions are weighted by the number of births in each cell. * p<0.05 ** p<0.01 *** p<0.001. 53

Table A2: Microdata Evidence on Car Spending Total Cars Used Cars New Cars All Households Has Children Under 2 0.033** 0.037*** -0.004 (0.012) (0.011) (0.005) N 15,355 15,355 15,355 Mortgaged Households Has Children Under 2 0.041* 0.048** -0.008 (0.018) (0.017) (0.009) N 6,915 6,915 6,915 Income Controls x x x Income Sq. Controls x x x Age Dummies x x x HH Size Dummies x x x Year Dummies x x x Sample includes households with at most one child under 2, no other children and at least one adult under 45. Age dummies represent the same age categories as our main analysis and household size is split between 1 or more then one adults aged between 18-45 in the houshold. Source: ONS Living Costs and Food Survey (LCFS). * p<0.05 ** p<0.01 *** p<0.001. 54

Table A3: Mortgage Characteristics Across Samples Variable Full Sample Known-Initial-Period Sample Region East Anglia (%) 4.5 4.3 East Midlands (%) 6.3 6.0 London (%) 14.5 15.0 North (%) 5.7 5.9 North West (%) 12.6 12.7 South East (%) 22.7 23.2 South West (%) 9.0 9.1 Wales (%) 5.2 5.3 West Midlands (%) 8.9 8.4 Yorkshire (%) 10.6 10.1 Employment status Employed (%) 79.1 76.9 Self-employed (%) 17.7 19.5 Other (%) 3.2 3.6 Rate type at 2008Q3 On SVR (%) 19.8 25.8 Mortgage type First time buyer (%) 16.4 17.9 Home mover (%) 29.3 29.0 Remortgage (%) 51.1 50.4 Other (%) 3.2 2.7 Initial period 1 year or less (%) 3.0 8.5 2 years (%) 48.2 48.1 3 years (%) 20.7 18.4 4 years or more (%) 28.1 25.0 N 4.58m 1.27m Data Source is PSD. There are 4.58m mortgages in the full sample, of which 58.6% are on a fixed rate in July 2008. The Known-period sample is made up of all mortgages on an adjustable rate in July 2008 with a declared initial period (0.53m), and a random sample of mortgage on a fixed rate with a declared initial period at that time (0.74m), ensuring the rate-type balance between the two categories matches the full sample. 55

Table A4: Detailed Mortgage Characteristics Across Samples Variable Mean Median SD p25 p75 FRMs Full Sample Age 37.9 37.0 10.1 30.0 44.0 Household income (£000s) 49.6 39.6 93.5 28.7 55.8 Joint-income mortgages (%) 53.1 100 49.9 0 100 Loan value (£000s) 128.9 110.2 94.6 75.6 156.8 Property value (£000s) 204.4 170.0 169.4 125.0 240.0 Interest rate (where available) 5.53 5.49 0.63 4.99 5.90 Known-period Sample Age 38.0 37.0 10.2 30.0 45.0 Household income (£000s) 52.3 40.6 90.4 29.5 57.5 Joint-income mortgages (%) 55.5 100 49.7 0 100 Loan value (£000s) 136.4 115.0 105.5 78.5 165.0 Property value (£000s) 210.3 170.0 187.8 125.0 245.0 Interest rate (where available) 5.63 5.59 0.65 5.09 5.99 Non-FRMs Full Sample Age 40.5 40.0 10.4 33.0 47.0 Household income (£000s) 67.1 46.2 204.7 32.0 70.3 Joint-income mortgages (%) 50.9 100 50.0 0 100 Loan value (£000s) 146.5 116.0 139.9 73.3 176.3 Property value (£000s) 256.9 200.0 252.5 140.0 292.6 Interest rate (where available) 5.40 5.34 0.64 4.94 5.83 Known-period Sample Age 40.1 39.0 10.5 32.0 47.0 Household income (£000s) 57.7 43.5 150.8 30.7 64.7 Joint-income mortgages (%) 50.8 100 50.0 0 100 Loan value (£000s) 136.8 114.0 107.4 72.2 170.0 Property value (£000s) 231.8 185.0 197.4 135.0 269.6 Interest rate (where available) 5.20 5.00 0.66 4.74 5.59 Data Source is PSD. There are 4.58m mortgages in the full sample, of which 58.6% are on a fixed rate in July 2008. The Known-period sample is made up of all mortgages on an adjustable rate in July 2008 with a declared initial period (0.53m), and a random sample of mortgage on a fixed rate with a declared initial period at that time (0.74m), ensuring the rate-type balance between the two categories matches the full sample. 56