The Bank of England / NMG Survey of Household Finances

Transcription

1 FISCAL STUDIES, vol. 37, no. 1, pp (2016) The Bank of England / NMG Survey of Household Finances GARETH ANDERSON, PHILIP BUNN, ALICE PUGH and ARZU ULUC Bank of England (Gareth.Anderson@bankofengland.co.uk) Bank of England (Philip.Bunn@bankofengland.co.uk) Bank of England (Alice.Pugh@bankofengland.co.uk) Bank of England (Arzu.Uluc@bankofengland.co.uk) Abstract Every year since 2004, the Bank of England has commissioned NMG Consulting to carry out a survey on household finances. This paper describes the NMG Survey, its methodology, and its advantages and disadvantages relative to other surveys. The NMG Survey is useful in providing a timelier guide to developments in the distribution of household balance sheets than other surveys, it appears better at measuring financial distress, and it includes questions on topical policy issues that are often not available in other surveys. Submitted July The views expressed in this paper are those of the authors, and not necessarily those of the Bank of England or its committees. The authors are grateful to Alan Castle, Tom Crossley, Cormac O Dea, Paul Robinson, Gavin Wallis, Garry Young, an anonymous referee and participants at the March 2015 conference on Household Wealth Data and Public Policy for helpful comments and suggestions. This work contains statistical data from the Office for National Statistics (ONS) which are Crown copyright. The use of the ONS statistical data in this work does not imply the endorsement of the ONS in relation to the interpretation or analysis of the statistical data. This work uses research data sets which may not exactly reproduce National Statistics aggregates. Keywords: household debt, financial distress, wealth, survey methodology. JEL classification numbers: C81, D31. C 2016 Bank of England.

2 132 Fiscal Studies A drawback of the NMG Survey is that there may be a greater risk of selection into the survey based on unobservable characteristics than is the case for some other household surveys. Policy points Understanding developments in households balance sheets is important for both monetary policy and financial stability. Aggregate data can provide only a limited assessment of balancesheet positions. There are large differences in financial positions across households, which can affect macroeconomic outcomes. The Bank of England / NMG Survey is one of a number of publiclyavailable household surveys providing information on balance-sheet positions. The NMG Survey is timelier than other surveys and it provides a relatively reliable guide to changes in the distribution of balance sheets. It should be viewed as a complement to other surveys that are more comprehensive but less timely. The NMG Survey also appears better at measuring financial distress than other surveys, and it includes questions on topical policy issues that are often not available in other surveys, such as around how households may respond to higher interest rates. The main drawback of the NMG Survey is that there is a greater risk that households may be selected into the survey based on unobservable characteristics than is the case in other surveys. I. Introduction Understanding developments in households balance-sheet positions has important implications for both monetary policy and financial stability and is therefore of particular interest to the Bank of England. A household s financial situation is likely to be an important determinant of how it responds to shocks, both in terms of how it adjusts spending and in terms of the likelihood of it suffering financial distress. Aggregate data can provide only a limited assessment of balance-sheet positions. Such aggregate data can mask large differences in financial positions across households because debt and wealth are both unevenly distributed. It is therefore important also to use household-level data to assess these differences, and in particular to monitor the proportion of households that are in a potentially vulnerable position. Furthermore, the emergence of heterogeneous agent models in macroeconomics, 1 which show how differences in the responses to 1 For example, Carroll (2012).

3 The Bank of England / NMG Survey of household finances 133 shocks across different types of households can affect aggregate outcomes, has increased the emphasis on using micro data to understand household finances. That is supported by empirical evidence on the heterogeneous responses of households to fiscal and monetary policy shocks, 2 which suggests that household surveys can offer valuable information for policymakers. The Bank of England uses data from a range of household surveys to help assess developments in the distribution of balance sheets. These include the English Housing Survey, Family Resources Survey, Living Costs and Food Survey, Wealth and Assets Survey, and Understanding Society. But they also include the Bank of England / NMG Survey (referred to as just the NMG Survey below), which is a household survey commissioned by the Bank and carried out on its behalf by NMG Consulting. This paper describes the NMG Survey, its methodology, and its advantages and disadvantages relative to other surveys. 3 The NMG Survey has been running since 2004 and includes questions on household balance-sheet positions and on attitudes to spending as well as ad hoc modules on topics of policy interest. Since 2012, the survey has been carried out online, having previously been done face-to-face. The shift to online sampling has facilitated the inclusion of a panel element to the survey. The NMG Survey micro data are publicly available on the Bank of England website. 4 The survey results are summarised in an annual article in the Bank of England Quarterly Bulletin. 5 As part of its Strategic Plan, launched in March 2014, the Bank committed to opening up its research agenda to external contributions by publishing its key research questions. 6 The NMG Survey is one of the data sets owned by the Bank that form part of that Agenda. The NMG Survey has advantages and disadvantages relative to other household surveys. These need to be considered in the context that the NMG Survey is run on a small fraction of the resources of most other household surveys. Three of the main benefits of the NMG Survey are as follows: It is timelier than other household surveys. The micro data from the NMG Survey are typically publicly available within three months of the survey being completed rather than around two years on surveys run by the Office for National Statistics (ONS). It appears to be better at measuring financial distress than other surveys. It includes questions on topical policy issues that are not typically available in other surveys, and these data are available on a timely basis. 2 For example, Jappelli and Pistaferri (2014). 3 A more detailed version of this paper will be presented in Anderson et al. (2016). 4 See 5 For example, see Anderson et al. (2014). 6 See

4 134 Fiscal Studies A drawback of the NMG Survey is that there may be a greater risk of selection into the survey based on unobservable characteristics than is the case for other household surveys. Although the sample is weighted to be representative of the population and the survey provider targets harder-to-reach groups to ensure coverage is as complete as possible, the survey is carried out online and people do have to opt in to being part of the survey provider s database. All surveys face the problem that non-response may not be random, even when the sampling probabilities are known, but selection on unobservables may be more of an issue with the NMG Survey. It is therefore important to assess the reliability of the NMG Survey and compare results with those from other surveys. Overall, the NMG Survey does a reasonable job of tracking aggregate data and it follows broadly similar trends to other surveys, although it does do a little less well than other surveys in some respects. But it does contain enough useful information to provide an approximate guide to balance-sheet developments in a timely way before other survey data are released. This paper provides an overview of the NMG Survey and highlights its main pros and cons. It starts in Section II with a summary of the topics covered by the survey. Section III describes the methodology and the implications of moving the survey online. Section IV compares the NMG Survey with other household surveys, while Section V gives an example of how the survey has been used to assess policy questions. Section VI concludes. II. Topics covered by the NMG Survey The NMG Survey contains a long-running core set of questions around balancesheet positions. The original motivation for starting the NMG Survey back in 2004 was to provide a timelier read on balance-sheet developments than was available then from the British Household Panel Survey (BHPS) and to collect data on topics typically not well captured by other surveys, such as unsecured debt. These continue to be the main objectives of the survey. That first survey is summarised in May, Tudela and Young (2004). Reflecting the fact that the NMG Survey was initiated to provide timelier updates on balance sheets than were available from the BHPS, the wording of many of the balance-sheet questions in the NMG Survey is the same as that used in the BHPS. Early NMG surveys focused largely on collecting information on the liabilities side of balance sheets for example, including questions on things such as mortgage debt, mortgage payments and unsecured debt liabilities. As well as including questions asking about amounts of debt outstanding, the survey asks households more subjective questions about how they are coping with their debts and whether they are finding loan repayments a burden or not. Respondents have also been asked about their income and housing wealth in every survey in order to help assess the sustainability of those debts, as well as questions about the characteristics and composition of their household.

5 The Bank of England / NMG Survey of household finances 135 Over time, the NMG Survey has included a wider set of questions around the state of household finances in addition to focusing on debt. For example, since 2010, it has included questions on financial wealth (although these were also asked in 2005). 7 More questions have been added on attitudes to spending and saving, although the NMG Survey does not contain actual consumption data. And more recently the survey has begun to include more questions on households expectations both for their own personal finances and for the wider economy. Alongside these more regular questions, the NMG Survey has typically included some more ad hoc questions on topics of particular policy interest at the time of the survey. Some of these questions have been included in only one survey, while others have been included a number of times. The more bespoke modules included in the NMG Survey are discussed in more detail in Section V. III. Survey methodology The Bank of England / NMG Survey is carried out on behalf of the Bank of England by NMG Consulting, which is a strategy, analytics and market research company. Stakeholders within the Bank decide which questions are included in each survey, with input on areas of current policy interest from the Monetary Policy and Financial Policy Committees. The NMG Survey has been carried out online since Previously, the survey was conducted face-to-face. Since 2013, the NMG Survey has included approximately 6,000 households in each wave. That number has increased from 2,000 a year when the survey was face-to-face. The survey was annual between 2004 and 2013 and was conducted during September. In 2014 and 2015, there have been separate April and September surveys (with some changes in the questionnaire between the two surveys), with 6,000 households in each survey. The survey covers households in Great Britain only and so excludes Northern Ireland. This section describes the current methodology and the most substantive methodological change that has been made, which was moving the survey online. 8 7 The survey only asks about wealth held directly and not in pension funds. 8 Other methodological changes have been made to the NMG Survey for example, asking households to report actual values in response to questions requesting financial values such as those around income and debt. Prior to 2013, respondents were given a list of ranges to choose from rather than being asked to provide a point estimate of the actual value because it was thought that would lead to higher response rates. But households have more time to think about their responses in an online survey and, following tests in the 2013 survey, all respondents have been asked for actual values since 2014, in line with the best practice used in most other major surveys. The differences associated with this change were relatively modest and are discussed in more detail in Anderson et al. (2014). Any methodological changes introduced in a survey are described in the annual Bank of England Quarterly Bulletin article that accompanies the release of the survey.

6 136 Fiscal Studies 1. Current methodology Since moving online, the sample for the survey has been provided to NMG Consulting by Research Now, which is a specialist provider of samples for research surveys. Households are recruited to the Research Now panel through a variety of marketing channels, including online partner campaigns and ing, with specific invitations to particularly target those with characteristics that are more difficult to reach. Respondents are offered incentives for completing surveys. The incentive options allow panellists to select from a large range of gift cards, points programmes and partner products/services. The precise incentives offered depend on the survey length and the respondent s profile the incentives are larger for longer surveys and for individuals with harder-to-reach characteristics. Survey participation within the Research Now panel is limited to prevent survey fatigue and to prevent respondents from seeking to be professional panellists. The sample for the NMG Survey is weighted to be representative of the GB population in terms of age, gender, region, housing tenure and employment status, but within these criteria it is drawn randomly from the wider Research Now panel. However, the exact sampling probabilities are not known in the way that they are in other household surveys run by the ONS, which tend to use the Postcode Address File to draw their samples. 9 In the NMG Survey, respondents have to opt in to being part of the Research Now panel and, as the survey is now run online, they must have access to the internet since no alternative way of completing the survey is offered. Whilst all surveys face the problem that non-response may not be random, even when the sampling probabilities are known, selection on unobservables may be more problematic within the NMG Survey if households with internet access who sign up to the Research Now panel are not representative of households with similar observable characteristics (the ones used in the survey weighting) in the wider population. The data quality is checked by the survey provider to screen out illogical responses and extreme outliers. 10 However, the survey does not attempt to impute missing values or make any corrections to the data for possible nonresponse bias. By making no adjustments, the results implicitly assume that households who choose not to respond have similar characteristics to those who do respond. The issues around non-response in the NMG Survey are discussed in more detail in Barwell, May and Pezzini (2006). 9 The Post Code Address File is a list of all addresses in the country. 10 As part of this, respondents who complete the survey in less than 30 per cent of the median response time are subject to additional sense checks.

7 The Bank of England / NMG Survey of household finances Moving the NMG Survey online Prior to 2012, the NMG Survey was run as a face-to-face survey. The faceto-face survey was carried out by adding questions to a regular omnibus survey, Capibus, carried out by Ipsos MORI. Interviews were conducted in the respondents homes using Computer-Assisted Personal Interviewing. To avoid embarrassment in revealing sensitive information to the interviewer, replies to questions were coded on showcards and recorded on a computer in such a way that the interviewer would not know the content of respondents answers. The sample was chosen using a two-stage random location sample design to select a set of sampling points within regions and sub-regions. Those sampling points were different for each survey. Within each sampling point, interviewers were required to achieve interviews within a relatively small set of homogeneous streets. Quotas were applied to each interview assignment, to ensure a good spread of interviews across demographics, although final results were still weighted. In 2010 and 2011, online pilots of 500 and then 1,000 households were run alongside face-to-face surveys of 2,000 households. In 2012, the main survey moved online and covered 4,000 households, with a face-to-face parallel survey of 2,000 households containing only a small subset of questions. Those parallel runs can be used to assess how moving the survey online affected the results. It is not, however, possible to distinguish between the impact of asking questions online rather than face-to-face and the impact of the change in sampling methodology associated with moving the survey online. That could only be done by asking the same households the same questions in a different environment. Since 2013, the survey has been run entirely online with 6,000 households in each wave. Moving the NMG Survey online had a number of benefits: Asking households questions in a less time-pressured situation without the presence of an interviewer appears to have helped to increase disclosure about sensitive issues. For example, the proportion of mortgagors reporting problems paying for their accommodation was significantly higher in the online survey (Figure 1). Households were also more likely to report that they had unsecured debt (Table 1), and those with debt also reported higher average levels of unsecured debt. This finding is consistent with other research showing that households are more likely to disclose sensitive information in online surveys. 11 As discussed in Section IV, these higher levels of reported financial distress and unsecured debt from the online survey are closer to what is implied by aggregate data. But some of that increase in reported financial distress and unsecured debt could potentially 11 For example, Dayan, Paine Schofield and Johnson (2007).

8 138 Fiscal Studies FIGURE 1 Mortgage payment problems Percentage of mortgagors having problems paying for accommodation Understanding Society / BHPS NMG face-to-face NMG online Note: Question In the past twelve months, would you say you have had any problems paying for your accommodation?. Source: NMG Consulting Survey; Understanding Society (British Household Panel Survey (BHPS) before 2009); authors calculations. TABLE 1 Comparison of online and face-to-face surveys Online Face-to-face Online Face-to-face Average gross annual income ( ) 32,040 34,860 31,524 33,228 - Refusal rate (%) Average outstanding mortgage debt ( ) 78,899 99,625 85,189 84,008 - Refusal rate (%) Share with unsecured debt (%) Refusal rate (%) Sample size 1,000 2,000 4,000 2,000 Note: Outstanding mortgage debt is for mortgagors only. Source: NMG Consulting Survey; authors calculations.

9 The Bank of England / NMG Survey of household finances 139 also be due to selection associated with those types of households being more likely to be part of the online panel. Online surveys make it easier to ask longer and more complex questions for example, by tailoring questions based on responses to earlier questions in the survey. Online surveys make it easier and cheaper to cover a larger sample, which is particularly relevant when looking at subsamples of the survey. A larger sample should also reduce sampling volatility between successive surveys. Moving online allowed the sample size of the NMG Survey to increase from 2,000 to 6,000 without a significant increase in costs. Rates of item non-response were materially lower in the online survey than face-to-face, further increasing the number of observations available for analysis. Face-to-face, around 30 per cent of households typically refused to answer questions about their income and debts, whereas that figure fell to 5 10 per cent when the survey was carried out online (Table 1), although this could also reflect the respondents more actively opting in to being part of the survey. Moving online facilitated the introduction of a panel element to the survey since it is much easier to invite the same households to participate online. On average, around 50 per cent of respondents to online NMG surveys also responded to the previous survey. That panel element allows an assessment of how circumstances have changed from one year to the next for individual households. One disadvantage of moving the NMG Survey online was that the risk of selecting certain types of household into the survey based on unobservable characteristics may have increased. That is both because it required households to have internet access and because the respondents now had to select themselves into being part of the panel from which respondents were drawn, although the sample is now drawn from across the country rather than clustered sampling points. But the survey is still run in line with industry best practice and the Research Now panel complies with the ICC/ESOMAR International Code of Marketing and Social Research Best Practice. Other central banks also use online-only panels to provide them with timely information on developments in the economy for example, the New York Fed Survey of Consumer Expectations. 12 If moving the survey online and so requiring households to have internet access made the NMG Survey less representative, older and lower-income households may have been the groups most affected by this change. According to Understanding Society, around three-quarters of households had internet 12 See Armantier et al. (2013) for more details.

10 140 Fiscal Studies Percentage of households FIGURE 2 Distribution of gross annual household income in < Lower limit of annual income band ( thousand) NMG online NMG face-to-face Family Resources Survey Source: NMG Consulting Survey; Family Resources Survey; authors calculations. access in , but that proportion was lower for the over-70s and for those with an annual income below 10,000, at around 45 per cent. Since the NMG Survey is weighted by age and so contains a representative number of older households, moving the survey online would only affect the results to the extent that households with internet access had different characteristics from those without. But even if there are differences, they may not be particularly important for analysis around household debt, since older households typically do not hold large amounts of debt. It would be of potentially more concern if the survey under-represented low-income households, who could be susceptible to financial distress. There were fewer low-income households in the online survey than face-to-face, although not noticeably fewer than in other household surveys such as the Family Resources Survey (Figure 2), although this still relies on the assumption that the lowincome households with internet access are otherwise similar to those without. Aside from increased disclosure around financial distress and unsecured debt, differences in other variables such as income and mortgage debt were typically smaller between the online and face-to-face surveys (Table 1). Average income was 5 10 per cent lower online, largely because there were fewer high-income households, although these differences were not

11 The Bank of England / NMG Survey of household finances 141 statistically significant. Average mortgage debt was significantly lower online in the 2011 small-sample parallel run, but not in 2012, which illustrates how it is difficult to distinguish between potentially small changes associated with the methodology and sampling volatility. Overall, it was felt that the advantages of moving the NMG Survey online outweighed the disadvantages. Greater disclosure of unsecured debt and financial distress is of particular importance to the Bank given its responsibilities for financial stability, and there are no other surveys assessing households finances using an online approach. And whilst there may be a greater risk of selection into the sample based on unobservable characteristics, the differences in the survey results were relatively modest and offset to some extent by increases in the sample size and fewer households refusing to answer key questions. Kamath et al. (2011) and Bunn et al. (2012) contain further discussion of moving the NMG Survey online. IV. Validating the NMG Survey against aggregate data and other surveys Given some of the limitations of the NMG Survey methodology described above, it is important to validate the survey against aggregate data and other household surveys in order to assess how useful information from the NMG Survey is. Table 2 summarises the proportion of aggregate National Accounts data that can be accounted for by comparable aggregates constructed from a range of household surveys. For the NMG Survey, these are shown as an average over the full sample period and for 2014 separately as some of those estimates for earlier years will also have been affected by methodological changes. We attempt to compare the results with the aggregate data that most closely correspond to the data available in household surveys, although there are still some differences in definitions and methodologies between the aggregate data and different surveys that could account for some of the differences we observe. Average income and debt in the NMG Survey are typically lower than implied by aggregate data. That is also the case for most of the other UK household surveys and is typically a common finding across micro-to-macro studies. However, the degree of under-recording has tended to be greater in the NMG Survey than for other surveys. Among the other UK household surveys, the Wealth and Assets Survey has tended to be the best at matching aggregate data over the recent past. In part, that is likely to reflect the fact that its methodology is specifically designed to try to better capture wealthier households. The main area in which the NMG Survey, at least in its current form, does better than other surveys at matching up to aggregate data is in the measurement of sensitive issues around financial distress. As discussed above, households

12 142 Fiscal Studies TABLE 2 Percentages of aggregate data accounted for by household surveys (2005 latest) Gross income Mortgage debt Unsecured debt NMG, NMG, EHS FRS LCFS USoc/BHPS WAS Note: Comparisons are made using calendar-year data wherever possible. Aggregate National Accounts gross income includes only components of income received in cash and not those imputed or received by pension funds on behalf of households. Survey measures of mortgage debt refer to mortgage on main residence only. Aggregate data exclude an estimate of buy-to-let loans, but include secured loans to nonprofit institutions serving households and unincorporated businesses. Unsecured debt question was only asked in the BHPS in Aggregate unsecured loan data include unsecured lending to individuals from monetary financial institutions and student loans only. The table uses face-to-face NMG data from 2005 to 2010 and online data from 2011 onwards. Wealth and Assets Survey estimates for Wave 1 (mid-2006 to mid-2008) are compared with aggregate data for 2007, Wave 2 (mid-2008 to mid-2010) are compared with 2009 and Wave 3 (mid-2010 to mid-2012) are compared with Source: Bank of England; English Housing Survey (EHS; Survey of English Housing (SEH) before 2008); Family Resources Survey (FRS); Living Costs and Food Survey (LCFS); Office for National Statistics; NMG Consulting Survey; Understanding Society (USoc; British Household Panel Survey (BHPS) before 2009); Wealth and Assets Survey (WAS); authors calculations. reported significantly higher levels of financial distress in the online survey than they did face-to-face. It is hard to compare these qualitative measures with aggregate data, but it is possible to examine how well household surveys capture mortgage arrears, and the NMG Survey appears to do better than other surveys in this respect. Mortgage arrears questions have only recently been introduced into the NMG Survey, but estimates of the proportion of households in arrears are close to those from Council of Mortgage Lenders (CML) aggregate data, whereas estimates from other face-to-face surveys have typically been well below those implied by aggregate data (Figure 3). That suggests the online survey does better capture financial distress, although we cannot rule out the possibility that, at least in part, higher reported levels of financial distress reflect households that are more prone to such distress being more likely to select themselves into the online sample. Moving the survey online has also increased the coverage of unsecured debt in the NMG Survey relative to aggregate data (shown by comparing the 2014 NMG Survey with the average of all years in Table 2). Again that may relate to households being more willing to disclose more sensitive information in an online environment, although in this case that only takes the NMG Survey up to a level that is comparable to the Wealth and Assets Survey.

13 The Bank of England / NMG Survey of household finances 143 FIGURE 3 Mortgage arrears 6 Percentage of mortgages in arrears for at least three months CML aggregate data EHS BHPS NMG WAS Note: CML data represent the proportion of mortgages more than three months in arrears at a point in time. NMG and BHPS questions ask about arrears for two months or more at any point over the last year. The WAS question asks about two or more months in arrears at a point in time and EHS three months in arrears at a given point. To make the survey data comparable to the aggregate data, 70 per cent of those who are in arrears for at least two months are assumed to also be there for at least three months and the average duration in arrears conditional on missing three payments is six months. These assumptions are based on estimates from other questions in the NMG Survey. Source: BHPS; Council of Mortgage Lenders (CML); EHS (SEH before 2008); NMG Consulting Survey; WAS; authors calculations. The main benefit of the NMG Survey is to provide a timely update on the distribution of household balance sheets given that NMG data are typically available up to two years ahead of data from other surveys. But for this to be useful, it is important to know how balance-sheet distributions from the NMG Survey compare with those from other surveys. Figure 4 compares the distribution of mortgage debt to income ratios in 2011 from a range of surveys. Whilst the NMG Survey distribution looks broadly similar to that from other surveys, the main difference is that it implies that there are more households with mortgage debt to income ratios above 4 than in other surveys and fewer with debt to income ratios between 0 and 1. That has also been the case in most other years (Figure 5). One reason why the NMG Survey may have had a higher proportion of highdebt-to-income households than other surveys is that the NMG Survey asks

14 144 Fiscal Studies 35 FIGURE 4 Distribution of mortgage debt to income ratios in Percentage of mortgagors > Mortgage debt to income ratio NMG LCFS WAS EHS FRS Note: Outstanding mortgage on main residence relative to pre-tax annual household income. NMG Survey data are from the online survey. WAS data shown are for Wave 3 (mid-2010 to mid-2012). Source: EHS (SEH before 2008); FRS; LCFS; NMG Consulting Survey; WAS; authors calculations. about income in a different way. Figure 2 highlights that while the distribution of income in the online NMG Survey is broadly similar to that of the Family Resources Survey, the proportion of households with high incomes is lower in the NMG Survey. The NMG Survey has always asked respondents for their total household income, whereas other surveys typically ask for each component of income for each member of the household and then aggregate those responses. It is possible that some respondents to the NMG Survey forget to report either some components of income or the income of some household members. Tests in the Spring 2015 NMG Survey asked households to report income using four different approaches that were randomly allocated across respondents. Approaches asking for the income of each household member separately yielded significantly higher average income, while asking for components of income made less difference to the averages but significantly reduced response rates (Table 3). Those higher averages when asking about the income of all household members separately reflect there being more households with incomes above 40,000 a year and fewer reporting incomes between 17,500 and 40,000 (Figure 6). If a similar pattern had been evident in 2012, that would have moved the NMG income distribution closer to that of

15 The Bank of England / NMG Survey of household finances 145 FIGURE 5 Percentage of mortgagors with debt to income ratios above 4 25 Percentage of mortgagors NMG BHPS/USoc LCFS FRS EHS WAS NMG new method Note: Outstanding mortgage on main residence relative to pre-tax annual household income. Uses face-toface NMG data from 2005 to 2010 and online data from 2011 onwards. WAS data shown are for Wave 3 (mid-2010 to mid-2012). Source: EHS (SEH before 2008); FRS; LCFS; NMG Consulting Survey; USoc (BHPS before 2009); WAS; authors calculations. TABLE 3 Average income from experiments using different income questions Income question Mean income ( ) Share of National Accounts aggregate (%) Non-response rate (%) 1. Total household income only 33, Total income for each household member 39, Income components for household only 34, Income components for each household member 38, Note: Non-response includes those who failed to complete the survey as well as those who completed the survey but did not answer the income question. Data are from Spring 2015 NMG Survey. Source: NMG Consulting Survey; authors calculations.

16 146 Fiscal Studies FIGURE 6 Distribution of pre-tax annual household income with different income questions Percentage of households < Lower limit of annual income band ( thousand) Total income only Total income for each household member Note: Data are from the Spring 2015 Survey. Source: NMG Consulting Survey; authors calculations. the FRS (the actual differences in 2012 are shown in Figure 2). This result that the design of household income questions affects the responses is consistent with the findings of Micklewright and Schnepf (2010). In future surveys, the income question will change so that it asks about the total income of each household member separately (option 2 in Table 3). The new question reduces the proportion of households with high mortgage debt to income ratios (shown by the circle on Figure 5 the NMG line for 2015 is based on the subsample asked the old income question). It is likely that the different approach to asking about household income explains much of the levels difference in the proportion of high-debt-to-income households between the NMG Survey and other surveys over the past, although it will only be possible to fully evaluate this once a parallel run of data for the new NMG question and other surveys becomes available. While the NMG Survey may not necessarily provide the most accurate estimate of the proportion of households with high debt to income ratios, changes in that share do follow a broadly similar pattern of gradual increases up to around There are not yet enough data available from other surveys to assess whether the fall back in the share of highly indebted mortgagors

17 The Bank of England / NMG Survey of household finances 147 FIGURE 7 Percentage of mortgagors with loan to value ratios above 75 per cent 25 Percentage of mortgagors NMG BHPS/USoc WAS Note: Outstanding mortgage on main residence relative to self-reported property valuation. NMG data are from the face-to-face survey for 2007 to 2010 and from the online survey for 2011 onwards. WAS data for Wave 1 (mid-2006 to mid-2008) are shown as 2007, Wave 2 (mid-2008 to mid-2010) as 2009 and Wave 3 (mid-2010 to mid-2012) as Source: NMG Consulting Survey; USoc (BHPS before 2009); WAS; authors calculations. seen in more recent NMG Surveys will be reflected in the other household surveys, although that downward trend is at least consistent with further falls in the aggregate National Accounts mortgage debt to income ratio. Similar conclusions to those for the distribution of mortgage debt to income apply to analysis of the distribution of mortgage repayments relative to income. The NMG Survey has tended to more closely match balance-sheet distributions in other surveys for metrics that do not include income. For example, Figure 7 shows that the NMG Survey has provided similar estimates of the proportion of households with outstanding mortgage loan to value ratios above 75 per cent to other surveys, in terms of both the level and changes. The timelier data from the NMG Survey are likely to be informative in this respect. V. Using the survey to assess policy questions Over time, the NMG Survey has included modules of questions on a range of policy issues where there is little information available in other household surveys. Importantly, the responses to these questions are available to policymakers on a timely basis. Examples of policy modules included in

18 148 Fiscal Studies the NMG Survey include estimates of marginal propensities to consume and how they vary between households, reasons for saving decisions, intentions to purchase property, the impact of fiscal tightening, holdings of buy-to-let investments and responses to changes in pension regulations. The NMG Survey has been used to conduct a range of policy analysis both within and outside the Bank of England. Much of the Bank analysis is published in annual Bank of England Quarterly Bulletin articles, 13 but it is also regularly used in the Financial Stability Report and the Inflation Report and has been used in other Bank research. 14 The data have been used outside the Bank too. 15 To give a flavour of the types of analysis that have been done using policy modules in the NMG Survey, this section summarises work assessing the potential impact of higher interest rates, which is described in more detail in Anderson et al. (2014). Assessing the impact of higher interest rates The impact of higher interest rates has important implications for both monetary and financial stability. From a monetary policy perspective, it is important to understand by how much aggregate spending may be likely to change following a rise in interest rates. And from the viewpoint of financial stability, it is important to know to what extent higher rates might increase financial distress. At the time of the Autumn 2014 NMG Survey, financial market prices suggested that Bank Rate was likely to start rising within the following six months. That survey therefore included a range of questions aimed at assessing the implications of higher interest rates, although subsequent developments in the economy meant that Bank Rate ended up remaining unchanged at 0.5 per cent, at least up to the time of writing in November Amongst other things, the survey asked households how they would respond to a 2 percentage point rise in interest rates. Households were told how much their interest payments/receipts would increase, in pounds, if interest rates were to increase by 2 percentage points, assuming no change in their incomes. That increase in interest payments/receipts was calculated using responses for the amount of debt and deposits held from other questions in the survey. This should have made it easier for households to respond accurately, by placing the impact of higher rates in the context of their own financial situation. Around 60 per cent of borrowers reported that they would cut spending in response to a 2 percentage point rise in interest rates, while only 10 per cent of savers expected to increase spending. Some borrowers also expected to respond in other ways such as by working more or requesting a change to 13 Such as Anderson et al. (2014). 14 For example, Bunn et al. (2015) and Bunn and Rostom (2015). 15 For example, Gathergood (2012), Whittaker (2012) and Gibbons and Vaid (2015).

19 The Bank of England / NMG Survey of household finances 149 FIGURE 8 Impact of a 1 percentage point rise in interest rates on income and spending of borrowers and savers Per cent Mortgagors Unsecured-only borrowers Savers All Impact on post-tax income Impact on consumption via cash-flow effect Note: Questions If your monthly mortgage / unsecured loan payments / interest you receive on your savings were to increase for a sustained period by x [which is calculated automatically from software as the payment increase under a 2 percentage point increase in interest rates], how do you think you would respond? Please assume your income would not be any higher unless you take action to increase it. Respondents who reported that they would cut/increase spending were then asked How much would you reduce/increase your monthly spending by in this situation?. Mortgagors are defined as households with a mortgage who have positive net debt, unsecured borrowers are non-mortgagors with positive net debt and savers are households with net savings. Post-tax income is calculated by deducting estimates of National Insurance and income tax from reported total gross household income. Current consumption is estimated as post-tax income less regular saving. See Anderson et al. (2014) for further details on these calculations. Source: NMG Consulting Survey; authors calculations. the loan, but few savers expected to take other forms of action. Households that reported they would change spending were also asked by how much they would change spending. That allowed marginal propensities to consume to be constructed and an impact on aggregate spending to be estimated. Estimates based on analysis from the 2014 NMG Survey implied that a 1 percentage point increase in interest rates could reduce aggregate spending by around 0.5 per cent via a cash-flow effect of redistributing income from borrowers to savers. This was not considered to be an unusually large effect. Figure 8 illustrates the estimated impact of a rise in interest rates on the income and spending of both borrowers and savers. But since the NMG data

20 150 Fiscal Studies FIGURE 9 Impact of a 1 percentage point rise in interest rates on income and spending by age 3 2 Per cent All Age Impact on post-tax income Impact on consumption via cash-flow effect Note: Calculated using the methodology explained in the note to Figure 8. Results for borrowers and savers within each age group are then aggregated. Source: NMG Consulting Survey; authors calculations. are constructed from the household level, they can also be used to assess the distributional impact of raising interest rates along other dimensions such as age, region and income decile (Figure 9 shows the impact on spending and income by age). VI. Conclusion Understanding developments in households balance sheets is important for both monetary policy and financial stability. The Bank of England / NMG Survey is one of a number of publicly-available household surveys providing information on balance-sheet positions. The key benefits of the NMG Survey relative to other surveys are that: it is timelier than other surveys, with the micro data typically available around two years earlier; it appears better at measuring financial distress; and it includes questions on topical policy issues that are often not available in other surveys. These factors are all important for policymakers. The main drawback of the NMG Survey is that there is a greater risk that households may be selected into the survey based on unobservable

21 The Bank of England / NMG Survey of household finances 151 characteristics than is the case in other surveys. That is because respondents have to have internet access and opt in to being part of the survey provider s panel, although the survey is weighted and the survey provider does target groups that are harder to reach in order to try to make the survey as representative as possible. The NMG Survey should not be seen as a substitute for more comprehensive balance-sheet surveys such as the Wealth and Assets Survey. It should be viewed as a complement to those other surveys, providing a timelier, and relatively reliable, guide to changes in the distribution of balance sheets than is available from those other surveys, at a reasonably low cost. References Anderson, G., Bunn, P., Pugh, A. and Uluc, A. (2014), The potential impact of higher interest rates on the household sector: evidence from the 2014 NMG Consulting survey, Bank of England Quarterly Bulletin, vol. 54, pp ,, and (2016), Evaluating the Bank of England/NMG Survey of households, Bank of England Staff Working Paper, forthcoming. Armantier, O., Topa, G., van der Klaauw, W. and Zafar, B. (2013), Introducing the FRBNY Survey of Consumer Expectations: survey goals, design, and content, eteconomics.newyorkfed.org/2013/12/introducing-the-frbny-survey-of-consumer-expectat ions-survey-goals-design-and-content.html. Barwell, R., May, O. and Pezzini, S. (2006), The distribution of assets, income and liabilities across UK households: results from the 2005 NMG Research survey, Bank of England Quarterly Bulletin, vol. 46, pp Bunn, P., Drapper, E., Pugh, A. and Rowe, J. (2015), How important are households expectations for spending?, Johnson, R., Le Roux, J. and McLeay, M. (2012), Influences on household spending: evidence from the 2012 NMG Consulting survey, Bank of England Quarterly Bulletin, vol. 52, pp and Rostom, M. (2015), Household debt and spending in the United Kingdom, Bank of England Staff Working Paper no Carroll, C. (2012), Implications of wealth heterogeneity for macroeconomics, Johns Hopkins University, Department of Economics, Working Paper no Dayan, Y., Paine Schofield, C. and Johnson, A. (2007), Responding to sensitive questions in surveys: a comparison of results from online panels, face to face and self-completion interviews, World Association for Public Opinion Research, Berlin. Gathergood, J. (2012), How do consumers respond to house price declines?, Economics Letters, vol. 115, pp Gibbons, D. and Vaid, L. (2015), Britain in the red: provisional report, Centre for Responsible Credit, liminary%20report.pdf. Jappelli, T. and Pistaferri, L. (2014), Fiscal policy and MPC heterogeneity, American Economic Journal: Macroeconomics, vol. 6, no. 4, pp Kamath, K., Nielsen, M., Radia, A. and Reinold, K. (2011), The financial position of British households: evidence from the 2011 NMG Consulting survey, Bank of England Quarterly Bulletin, vol. 51, pp

22 152 Fiscal Studies May, O., Tudela, M. and Young, G. (2004), British household indebtedness and financial stress: a household-level picture, Bank of England Quarterly Bulletin, vol. 44, pp Micklewright, J. and Schnepf, S. V. (2010), How reliable are income data collected with a single question?, Journal of the Royal Statistical Society: Series A (Statistics in Society), vol. 173, pp Whittaker, M. (2012), On borrowed time? Dealing with household debt in an era of stagnant incomes, Resolution Foundation, uploads/2014/08/on_borrowed_time.pdf.