Household Access to Mortgages in the UK

We employ the propensity score matching approach to investigate household access to mortgages in the UK using information on 29,732 households between 2003 and 2010. We find that, on average, the probability of obtaining a mortgage is similar for White and Non-White households. However, we find that Black households with low incomes are less likely to have mortgages compared to White households with similar characteristics. Asian households, in contrast, do not seem to have a lower probability of having a mortgage.


1
Financial inclusion refers to a process whereby people have access and/or use financial services (such as basic bank and saving accounts) and products (such as loans and mortgages) in the mainstream market that are appropriate to their needs and enable them to lead a normal social life in the society in which they belong (European Commission 2008). Literature also defines households with access to various financial services as Bbanked^.
economies find that households with lower usage of financial services are more likely to be from non-White background (see for example Hogarth et al. 2005;Devlin 2005;Carbo et al. 2007;Khan 2008;Honohan 2008;Simpson and Buckland 2009;Demirguc-Kunt and Klapper 2013). 2 In this paper we investigate differences in the probability of having a household mortgage in the UK. Previous evidence suggests that access to mortgage and other forms of household finance is influenced by ethnicity (Pollin and Riva 2001;Goodwin et al. 2000). Other evidence also suggests that in the UK various households may not have household credit at all (Kempson and Whyley 1999;Devlin 2005;Khan 2008;Deku et al. 2015). Recent public policy concerns have also been raised about the inability of ethnic groups to obtain SME finance (Clegg 2011) and in the literature access to finance has been cited as one of the most significant barriers facing ethnic minority businesses in the UK (Bates 2011;Carter et al. 2015).
While there is evidence suggestive of prejudice in ethnic group usage of financial services, the aforementioned UK empirical literature does not consider a major household financial market, namely, mortgages. Our analysis seeks to fill this gap in the literature. Previous research on the likelihood of obtaining household mortgage finance mainly focuses on the US and shows that non-White minorities face inequality in access to mortgages on noneconomic grounds. This empirical literature finds that non-White households have higher mortgage application rejection rates and are offered less attractive terms than Whites with similar credit and other features (Black et al. 1978;Munnell et al. 1996;Ross and Yinger 1999).
Building on this evidence, we investigate the probability of non-White households obtaining mortgages in the UK. Mortgages are important for households as the ability to obtain a mortgage is critical to a household's capacity to acquire a home, and it is typically the largest financial commitment most people make in their entire lives. Being a homeowner is an important component of wealth acquisition and can increase status and standing in society. Hence, the inability to obtain mortgage credit may significantly hinder household wealth and aggravate social exclusion. We seek to provide a better understanding of the key determinants of household access to mortgage credit in the UK which we hope can help inform public policy.
We use the Propensity Score Matching (PSM) approach and a large household sample compiled from the Living Costs and Food Survey gathered by the Office of National Statistics (ONS). The sample, collected between 2003 and 2010, consists of information on the economic, social and demographic features of 29,732 households, of which 17,398 have a mortgage and 12,234 are renting. 3 In general, we find that non-White households have the same chance of obtaining a mortgage as White households. However, Black households with low incomes are less likely to have mortgages compared to White households at similar income levels and other characteristics. Asian households, in contrast, do not seem to have a lower probability of having a mortgage. The 2007/08 financial crisis does not seem to have reduced mortgage access for Black households.
The remainder of the paper is organized as follows: Section 2 provides a brief background on the ethnic discrimination literature and also highlights the disadvantages faced by ethnic minorities in the UK. Section 3 discusses the developments in the UK housing market between 2001 and 2010. Section 4 describes the data sources, explains the methodology and provides descriptive statistics. The results are presented and discussed in Section 5. Section 6 concludes. discouragement and the possibility of loan denials rather than ethnic discrimination by banks (Coleman 2000;Fraser 2009). 4 One reason why ethnic minority business owners may feel discouraged from applying for bank loans is because they fear prejudicial treatment (Blanchflower et al. 2003;Kon and Storey 2003). Fraser (2009) argues that loan denials are linked to differences in creditworthiness rather than ethnic discrimination. 5 While there is a literature covering ethnic minority use of a variety of UK consumer financing, research on household mortgages remains limited. In contrast, there exists an established US literature. For instance, Black and Hispanic households are more likely to be rejected and offered less attractive terms for mortgages than Whites (Black et al. 1978;\Munnell et al. 1996;Ross and Yinger 1999). Non-Whites pay more for their mortgages even when factors such as income levels, property dates and the age of buyer are controlled for (Oliver and Shapiro 1997;Courchane and Nickerson 1997;Crawford and Rosenblatt 1999;Black et al. 2003;Cheng et al. 2015). Although these higher rates may be counteracted with more favourable terms, such as longer low rate lock-ins (Crawford and Rosenblatt 1999). Also, disparities in mortgage approval rates tend to fall substantially for Black households the longer their credit history (Han 2011). Redlining, outlawed in the US by the 1974 Equal Credit Opportunity Act, has also been widely researched. 6 Early work finds little evidence of such discriminatory mortgage lending practice (Schafer and Ladd 1981;Benston and Horsky 1992;Munnell et al. 1996;Tootell 1996). However, the majority of later studies document that minority neighbourhoods have lower access to mortgage funding (Phillips-Patrick and Rossi 1996;Siskin and Cupingood 1996;Ross and Yinger 1999) and are more likely to be subject to predatory lending practices (Calem et al. 2004;Williams et al. 2005;Dymski 2006). 7 Given the economic and social disadvantages faced by ethnic minorities in the UK, and evidence on their lower usage/access to various consumer financial services, we seek to expand this literature by examining the probability of different ethnic groups obtaining household mortgages. This evidence will add to evidence we already have on the US household mortgage market. Before we outline our data and approach the following section highlights developments in the UK housing market over our period of study. 4 Banks in the UK typically use credit scoring techniques to assess the riskiness of borrowers. Antidiscrimination legislation prevents the use of ethnicity, gender, disability or religious beliefs in determining credit scores. However, scoring techniques are often used as a complement to relationship lending, involving close contact between the entrepreneur and bank manager, which introduces the possibility for credit assessments to be tainted by personal prejudices (Fraser 2009). 5 Fraser (2009 finds that Black African firms are significantly more likely to miss loan repayments and/or exceed their agreed overdraft limit and this behavior seems to largely account for their much higher loan denial rates. 6 Redlining is the refusal to lend to certain neighborhoods due to non-economic features. In the US, in addition to the 1974 Equal Credit Opportunity Act, Community Reinvestment Act of 1977 made it illegal for lenders to have a smaller amount of mortgage funds available in minority neighborhoods compared to similar White neighborhoods. 7 There is also literature on discrimination in the US consumer credit market, however, the findings are not uniform. Some studies conclude that non-Whites are not discriminated against in terms of access to consumer credit (Lindley et al. 1984;Hawley and Fujii 1991). Others find that loan approval rates are lower for non-Whites (Duca and Rosenthal 1993) and that they pay higher interest rates (Edelberg 2007). Lenders chose to discriminate against non-Whites because, on average, they have higher default risk (Lin 2010). In addition, a number of studies look at auto loan pricing and find no evidence of discrimination (Goldberg 1996;Martin and Hill 2000) although this could be because non-price terms differ for minorities compared to Whites leading those discriminated against to drop out of the market (Ayres and Siegelman 1995).

Snapshot of developments in the UK housing market
The UK has one of the most persistently volatile housing markets, with four boom and bust cycles since the 1970s (Stephens 2011). This is important as the consequences of these cycles may have a disproportionate impact on different segments of the population. Boom and bust cycles may influence housing choices and bank lending behavior, inhibit house building and increase wealth inequalities. We present a set of key indicators of the housing market over 2001 to 2010 in Table 1.
Average house prices in the UK increased 65.6 % between 2001 and 2007 from £138,281 to £227,735 and dropped significantly afterwards with the impact of the financial crisis. UK housing supply, unlike the US, is argued to be much less sensitive to price changes and increases in demand for housing has a large impact on price but not volume (Meen and Andrew 2005). High levels of house completions were achieved in the period between 2003 and 2007. However, after the financial crisis new house construction fell notably, failing to cover the demand from newly formed households. Hence, the ratio of rented to owner occupied houses increased steadily, from 4.44 % in 2006 to 5.23 % in 2010 and average monthly rent increased 4.9 % during this period, above average yearly inflation level of 2.1 %. These figures signal an increase demand for rental properties perhaps because of tighter lending policies of banks during and after the financial crisis.
The number of households in the UK increased steadily over this period, an average of 0.77 % a year with average household size remaining unchanged at around 2.36. The average age of mortgage holders also gradually increased after the crisis. A combination of rising household numbers and lower housing completions together with tighter credit conditions are likely to have forced poorer and younger households out of the market.
In the pre-crisis period borrowing costs for mortgages became more expensive for households, increasing from 3.81 to 5.79 % between 2003 and 2008 (based on 2 year variable rates). However, other lending conditions became more favorable. Households were able to borrow more (mortgage payments to average pay increased two fold between 2001 and 2007) with the requirement of lower initial deposits. For first time borrowers the average deposit amount was 16.4 % in 2006 compared to 23 % in 2003. These favorable conditions indicate a relaxing of mortgage lending standards prior to the financial crisis and a tightening thereafter. 8 Looking at the age distribution of borrowers, younger households increased their share of the housing market pre-crisis. However, this trend soon reversed as the share of households under 34 years fell from 47.2 % in 2007 to 40 % in 2010.
Different types of mortgage products are available in the UK. Households can borrow for maturities ranging from 5 to 30 years. Borrowers may choose to repay the capital along with the interest or opt out for paying only interest. 9 Rates may be fixed (typically 2 to 5 years) or variable (determined by the lender or tracking an official rate-such as the Bank of England's base rate). Borrowers can switch from fixed to variable rates although penalty costs (often 3 months mortgage payments) are charged. Borrowers need to make deposits to obtain a mortgage and these vary according to the lender. After the financial crisis minimum deposit levels increased to around 25 % of property valuation, whereas pre-crisis deposits typically were around 5 %. The increase in initial deposits has disadvantaged poorer and younger House prices adjusted for retail prices. This uses the Office for National Statistics Retail Price Index (RPI) to convert nominal prices to current prices households who typically have more limited savings. Property can be repossessed if the borrower falls into arrears-under current law lenders have to give at least 3 months notice before repossession actions take place. 10 In the UK, unlike the US, households can still be pursued through the courts for any negative equity even after the repossession of the property by the lender. Overall one can see that the access to mortgage finance was more favorable to poorer households prior to the crisis and conditions have tightened thereafter. Given that ethnic minorities are disproportionately represented in the lower household income groups, it could be that they have also suffered more from the changed mortgage market environment.

Data and methodology 4.1 Data source
We collect our data from the Living Costs and Food Survey gathered by the Office of National Statistics (ONS) in the UK for the years between 2003 and 2010. This is an annual exercise to collect data on private household expenditure on goods and services. Most of the questions address issues relating to household characteristics such as, race, family relations, employment details, as well as information on household spending and income features. Following previous literature on the UK (Kempson and Whyley 1999;Devlin 2005;Deku et al. 2015), the household reference person is assumed to be the most influential within the household even though certain responses require that variables are aggregated for all household members. Our sample consists of 29,732 households, of which 17,398 have a mortgage and 12,234 are renting. We do not include observations where the household owns the property outright as the historical source of ownership (whether inheritance or paid mortgage) is unknown. Non-White households constitute 5.68 % (1,681 households) of the entire sample. Within the non-White group the distribution of Asian and Black households are 944 and 737, respectively. 11 We exclude mixed race and other race categories from our sample as it is not possible to identify ethnic background of these categories accurately.

Propensity score matching
Here we aim to answer the following question: Are non-White households, ceteris paribus, less likely to have mortgages than White households with comparable economic and other characteristics? A potential selection bias emerges as being a non-White household is likely to be endogenous and related to various other observable characteristics. As such we follow Rosenbaum and Rubin (1983) and use PSM as a way to reduce selection bias. Matching restricts inference to the sample of non-White households (the treatment group, denoted T i = 1 for household i) and White households (the control group, denoted T i = 0). The treatment group is matched with the control group on the basis of its propensity score which is a function of households' observable characteristics (X i ): Following Dehejia and Wahba (2002), we match the households based on the nearestneighbor with the replacement. Propensity scores are estimated via a probit model utilizing household head's characteristics (age, employment, occupational classification, education, gender and marital status) and other household characteristics (household size, income, benefits and regional location) as independent variables, which are drawn from the aforementioned literature.
Age is categorized into 10 year bands ranging from 16 to 65+. Employment status is defined as employed, unemployed, retired or unoccupied. Occupational classification indicates the skill level and content of the head of household's employment into six categories as: higher managerial, professional or working for a large employer; lower managerial; clerical and intermediate; small employers or self owned business; lower supervisory or technical; and routine/semi-routine manual or service. For education we use three UK educational qualification levels as GSCE (typically at 16 years of age), A-levels (typically at 18 years of age) and higher education (including further and higher university education). Marital status is grouped as married, co-habiting or single (including widowed, divorced and separated). Household size is the number of persons in a household. Income indicates the total weekly income of the household. Income gap equals 1 if household expenditure is more than its income and 0 otherwise. Benefits represent those households receiving any form of government benefit payments (from the Department for Work and Pensions or the Social Security Agency). Region is where the household is located. 12 We also use a set of year dummy variables to capture the effect of the macroeconomic environment.
We acknowledge the limitations of our analysis, which is common to the extant literature. PSM has advantages and shortcomings. It does not depend on the assumption of functional form and has the advantage of restricting inference to the sample of White and non-White households that are actually comparable in their observable characteristics. On the other hand, the PSM procedure relies on the assumption of selection on observables (namely the Conditional Independence Assumption), and it only corrects for selection bias among included observable characteristics. While we control for a rich set of covariates to explain access to mortgages, it cannot be completely ruled out that the existence of unobservable characteristics (such as social and cultural differences that could be correlated with racial differences) may still bias the treatment effect (Berkovec et al. 1996;Pager and Shepherd 2008;Han 2011). Controlling for all characteristics that affects the outcome requires very informative data on individuals, especially on the credit worthiness of borrowers. Unfortunately this information is not available in the database we are utilizing and this may lead to bias in our results. In an effort to minimize these potential biases, we examine subgroups of households where some economic indicators (such as being employed, having higher income or savings) aim to capture indirect dimensions of the creditworthiness of potential mortgage applicants. Table 2 provides summary statistics for the sample. In the first three columns, we display statistics for the whole sample and divide the sample by tenure. Non-White households constitute 5.7 % of all households and this ratio is higher, 7 %, for the renting group. The percentage of household heads that are employed for mortgage owners are 91.7 compared to 42.2 of the renting group. Average weekly income for mortgage owners is GBP 797, which is more than double the renting group (GBP 342). Household heads are more likely to hold a higher education qualification if they are home owners. Renting households are more likely to be female and less likely to be married when compared to mortgage owners.

Descriptive statistics
We also present characteristics of White and non-White households within each tenure choice. In the renting group White households are older than non-White households with mean ages 48.4 years and 40.8 years, respectively. A larger percentage of White renting households are over 65 and retired. In terms of size White households are smaller (mean size 2.18 members) than non-White ones (mean size 2.88 members), and a significantly lower percentage of White households have higher educational attainment. Although there are large observable differences between non-White and White household renters, these groups look much closer in terms of households' characteristics for households with mortgages. They have similar age distributions, educational and employment characteristics. One difference between the two groups is the average weekly income where White households earn on average £39 more per week.
In the latter four columns of Table 2 we divide the sample by racial origin. A higher percentage of non-White households (51.5 %) do not have mortgages compared to White households (40.9 %). Within the non-White group, the ratio seems to be driven by Black households, of whom 64.9 % do not have a mortgage. For Asian households this ratio is 41.6 % and close to the White households' mortgage ownership. Black households have a lower employment rate, are employed in lower paid occupations and have the lowest average weekly income. Compared to White households a lower percentage of Black households have a basic educational attainment (up to GCSE level), although they are more likely to have received (post-16 years of age) higher education. Furthermore, a higher proportion of Black household heads are female (55.2 %) and single (64.2 %). A larger fraction of Asian households are employed (73.3 %) with 20.4 % of them occupying managerial roles. Asian households' gross weekly income is the highest of all groups and they are more likely to have a degree in higher education and to be married.
Overall the descriptive statistics illustrate that non-White UK households are a diverse group with Asian households on average being economically better-off, while Black households are most economically disadvantaged.

White versus non-White households
We present propensity score estimations for the whole sample as well as the sub-samples in Table 3. Briefly, we find that non-White households have lower incomes and a larger household size. They are also more likely to experience income gap and be a benefit recipient. Non-White household heads are more likely to be married and less likely to work in    ***, **, and * indicate statistical significance at the 1 %, 5 %, and 10 % levels, respectively managerial positions. The results, sign and significance of the coefficients, are consistent across sub-groups, the only exception being variables in occupation classification.
We match non-White households with one, four and eight corresponding White households. To verify the quality of matching we plot the distribution of the propensity score for both groups before and after matching for the whole sample (Fig. 1). In the unmatched sample, the propensity score distribution of the White households is skewed to the left, whereas it is very close to that of the non-White households in the matched sample. This result suggests that the matches are appropriate.
We present the average treatment effect on the treated (ATT) in Table 4. Estimations include sub-groups of households where we are interested to see whether not having a mortgage is prevalent in sub-groups that have similar income levels, employment status, age, savings and income gap. 13 In Column 1 we observe that for the whole sample ATT is significant. In other words, for a household, on average, the effect of being non-White increases the likelihood of not having a mortgage.
Subsequently, we match propensity scores for two sub-groups above and below median income. We find that the difference between non-White and White households remains statistically significant for the below median income group. In contrast, ATT loses its significance for households above median income. This suggests that at higher levels of income non-White households have similar chances having a mortgage as White households. Looking at the households with an employed head, we still find ATT to be statistically significant. Thus, employment itself does not alleviate the chances of non-Whites have a lower chance of obtaining a mortgage. Next we scrutinize the sample by age, specifically for younger households who need access to mortgages. In 2013 the average UK house buyer was likely to be aged 36 when they acquired their first property (Office of National Statistics 2014). Hence, we examine the sample with household heads younger than 37 years old. We find that ATT is still significantly present for this group.
It could be argued that the probability of holding a mortgage is also related to the households' ability to put down a deposit; therefore, having savings may be a precondition to borrowing. With this in mind we examine the sub-group of households reported to have savings. Results show that non-White households have a lower likelihood of having a mortgage even if they have savings. We also look at the circumstances where households  Fig. 1 Distribution of property score of white and non-white households before and after matching 13 We recognize that these variables are all endogenous in the model. While this is an important caveat to keep in mind for results presented in this section, we think it is nonetheless interesting to see the results for sub-groups.  ***, ** and * represents significance levels at 1 %, 5 % and 10 %, respectively are less likely to save a deposit. Income gap shows the cases where the household's expenditure is higher than its income. The result does not change. As shown in Section 2, the UK housing market changed dramatically following the start of the financial crisis. Bank mortgage lending contracted sharply in the post crisis period with the poorer households (namely, the young and ethnic groups) most likely being the most adversely affected. To capture differences in accessing mortgages between these two periods, we repeat our analysis separately for the pre- (2003-2006) and post-(2007-2010) crisis period. Results are presented in the first two columns of Table 5. We find that results are similar as the effect of being a non-White household increases the likelihood of not having a mortgage in both periods. However, it is worth noting that, in the post 2007 period, the statistical significance of ATT weakens.
We also examine whether our results differ in regions where larger communities of non-White households live. We divide the regions into two groups using a 5 % threshold of non-White household presence. Regions that have more than 5 % non-White households to total residents include the North West, Yorkshire and Humber, East Midlands, West Midlands, London and the South East (Office for National Statistics 2012). We hypothesize that in regions where there are more non-White residents lending institutions will be less likely to not provide mortgages to these households as they constitute a large customer base. In addition lending personnel may have developed expertise to assess mortgage risks and have learned to deal appropriately with specific issues that are distinctive to various ethnic groups. On the other hand, in regions where non-White residents are rare, lending institutions may not develop such expertise and therefore in such regions ethnic minorities are less likely to have mortgages. We present our results in the latter two columns of Table 5. We find that the ATT is larger and has higher statistical significance in the sample from regions where non-White households constitute less than 5 % of the total residents. This result provides some evidence that non-White households living in regions where fewer minorities reside are less likely to have a mortgage.

Asian and Black households
This section compares Asian and Black household separately to White households as they have been shown to have different socio-economic characteristics noted earlier in this paper. Our findings are presented in Table 6. The main results for the whole sample are shown in the first two columns. We find that ATT is insignificant when Asian and White households are Table 5 Pre-and post-crisis periods and regional analysis. This table reports results for the propensity score matching estimates of the average treatment effect (ATT) of being a non-White household on the likelihood of not having a mortgage. Robust standard errors are bootstrapped

Number of controls matched
Pre-crisis period (2003)(2004)(2005)(2006) During and post-crisis period (2007)(2008)(2009)(2010) Regions with more than 5 % non-White population Regions with less than 5 % non-White population ***, ** and * represents significance levels at 1 %, 5 % and 10 %, respectively ***, ** and * represents significance levels at 1 %, 5 % and 10 %, respectively  ***, ** and * represents significance levels at 1 %, 5 % and 10 %, respectively matched. The effect of being an Asian household does not reduce the likelihood of not having a mortgage when compared to White households. On the other hand, ATT is positive and significant when we match White households with Black households. Black households are less likely to have a mortgage when compared to White households. We re-do our analysis for sub-groups pre-and post-crisis period. Our results are presented in columns two to five in Table 6. Similar to the aforementioned findings, we find that, when compared to White households, only ATTs for Black households are significant. Results for Asian households are not statistically significant. These findings are consistent in both the pre and post 2007 periods, with ATTs becoming slightly larger in the latter period when we compare Black to White households. We also separate the regions with higher or lower non-White household populations (again using the 5 % threshold) and compare the sub-groups of households with White households within these two regional groups. Results, presented in the latter four columns of Table 6, show that Black households, whether or not they live in a region with a higher portion of non-White residents, are less likely to have a mortgage. It is worthwhile to note that ATT are larger for the regions with less than a 5 % non-White population.
Overall, our sub-group results reveal that our main findings for the non-White households are driven by the Black household sample. Asian households do not seem to have a lower probability of having a mortgage when compared to White households. Hence, we shift our focus to Black households and stratify this sample by income levels, employment status, age, presence of savings and income gap. Results are presented in Table 7. We find that the difference between Black and White households remains statistically significant when we examine households that have below median income, have an employed household head and have savings. Younger Black households also seem to be less likely to have mortgages in comparison to their White peers. In contrast, ATTs are not significant for above median income groups. Black households with higher levels of income seem to have similar chances of obtaining a mortgage as White households.

Conclusion
We investigate the probability of households obtaining mortgages in the UK using a sample of 29,732 households between 2003 and 2010. Using PSM we compare mortgage ownership of White and non-White households which are very similar in terms of their observable characteristics. On average, we find no difference in the probability of obtaining a mortgage for White and non-White households. However, we do find that low income Black households are less likely to hold mortgages. We do not observe differences between Black and White households at higher income levels. The 2007/08 financial crisis does not seem to have had any influence in making the situation worse for lower income Black households as they have a higher chance, compared to Whites, of not having a mortgage pre-and post-crisis. There are geographical disparities in our findings: Black households are less likely to have mortgages in regions where fewer minorities reside. In contrast, we find that Asian households, unlike Black households, do not seem to have a lower probability of having a mortgage compared to White households.
Despite possible limitations of our methodology and data, we still argue that more work needs to be done trying to explain why it appears that Black households have poorer use of mortgage finance. One shortcoming of our analysis is that we only observe tenure status of households as a reduced form outcome, therefore, we cannot determine whether the lower probability of having a mortgage is a result of demand or supply-side factors. 14 It may also be claimed that the results capture self-exclusion due to Black households' possible cultural or religious differences. Our findings parallel those from the literature on SME financing in the UK, where Black business owners are found to have the lowest usage of bank finance. Perhaps, similar to the Black business case, Black households with low incomes may feel discouraged and do not apply for mortgages due to the belief that they would be rejected. An avenue for research may be to further investigate these issues.
Policy makers in the UK should seek to develop mechanisms aimed at lenders so they have to demonstrate that they are not discriminating against certain groups from the mortgage marked based on non-economic criteria as well as consider ways to reduce possible barriers that may inhibit Black households' use of mortgage market.