How reliable are household surveys? Phil Bunn, Gareth Anderson, Alice Pugh, Arzu Uluc Bank of England Household Wealth Data and Public Policy Conference, 9-10 March 2015 1 Motivation • Need to understand how reliable surveys are if using them to inform policy decisions • Evaluate the Bank’s own household survey, run by NMG Consulting 2 Outline • Comparing household surveys to each other and aggregate data: income and mortgage debt • What have we learnt about methodology from running the Bank of England/NMG survey: moving online 3 Previous literature • IFS work shows consumption has been increasingly under-recorded in the Living Costs and Food Survey, (Banks and Johnson (1998) and Barrett et al (2013)) • Previous Bank work looked at reliability of balance sheet data in the British Household Panel Survey, (Redwood and Tudela (2004)) • Our focus is on balance sheet data and financial distress 4 Surveys considered Survey Annual sample Main focus Family Resources Survey 25,000 Income Living Costs and Food Survey 5,000 Consumption Understanding Society/ BHPS 25-30,000/ 5000 Social & economic change Wealth and Assets Survey 20-30,000 per 2 year wave Wealth Bank of England/NMG Survey 2-6,000 Balance sheets/ Attitudes to spending 5 Bank of England/NMG Survey • Annual survey since 2004, run during September • Data on Bank website and annual Quarterly Bulletin article • Run online since 2012 • Now covers 6,000 households a year, previously 2,000 • Focus on balance sheets and attitudes to spending • Main benefits for us are timeliness and ability to add bespoke questions 6 A measure of gross cash income + + + + = Employee wages and salaries Self-employment income Government benefits Private pension payments Investment income Total gross cash income 7 Average household income Annual income per household, £ 50000 45000 40000 35000 30000 ONS aggregate FRS LCFS USoc/BHPS NMG WAS (2010-12) 25000 20000 15000 10000 5000 0 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 8 Distribution of individual incomes (2011) Percentage of income 30% HMRC 2011-2012 USOC 2011 WAS 2010-12 25% 20% 15% 10% 5% 0% Lower value of income band 9 Income coverage by component (2011) Percentage coverage relative to aggregate data 250 FRS LCFS 200 Usoc WAS 150 100 50 0 Wages and salaries Self employment Benefits Investment income 10 Reliability of income data: summary • Average income in surveys about 20% lower than implied by aggregate data, except WAS, which is in line • Degree of under-recording fairly constant over time • WAS better captures high debt to income households, although also has higher investment income • NMG survey 5-10% below other household surveys 11 Average mortgage debt Outstanding mortgage debt per household, £ 160000 ONS aggregate FRS LCFS USoc/BHPS NMG WAS 140000 120000 100000 80000 60000 40000 20000 0 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 12 Average mortgage debt Outstanding mortgage debt per household, £ 160000 ONS aggregate FRS LCFS USoc/BHPS NMG WAS ONS less BTL estimate 140000 120000 100000 80000 60000 40000 20000 0 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 13 Mortgage debt to income distribution (2012) Percentages of mortgagors 35 FRS LCFS NMG WAS (2010-12) 30 25 20 15 10 5 0 <1 1-2 2-3 3-4 Mortgage debt to income ratio 4-5 >5 14 Reliability of mortgage data: summary • All surveys under-record growth of mortgage debt since early 2000s • Some, but not all, of that reflects buy-to-let lending • Differences in share of high debt households quite small between surveys, although proportion is a bit higher in NMG survey 15 What have we learnt moving the NMG survey online 16 Moving the NMG survey online • What did we do? • Measurement of financial distress • Measurement of other financial variables • Response rates • Who has internet access? • Bespoke questions 17 Moving the NMG survey online • 2 online pilots in 2010 (500 households) and 2011 (1000 households) alongside full face to face surveys (2000 households) • Full survey online from 2012 (4000 households) and face-to-face parallel run of small number of questions (2000 households) • Increased sample size from 2,000 to 6,000 by 2013 18 Mortgage payment problems Percentages of mortgagors having problems paying for accommodation 25 BHPS/USOC NMG face to face NMG online 20 15 10 5 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 0 2014 19 Mortgage arrears Percentage of mortgages in arrears for at least 3 months 6 CML aggregate data BHPS NMG WAS 4 2 0 1987 1992 1997 2002 2007 2012 20 Average income and mortgage debt Online 2012 Face-to-face Mean gross annual income £31,524 £33,228 Mean mortgage debt (for mortgagors) £85,189 £84,004 21 Mortgage debt distribution Income distribution Percentages of mortgagors Percentages of households 12 20 NMG online NMG face-to-face FRS 18 NMG online 10 16 14 12 NMG face-to-face 8 FRS 6 10 8 4 6 4 2 2 Annual income (£, thousands) 0 <10 10 20 30 40 50 60 70 80 90 100 110 130 150 170 190 210 230 250 275 300 400 500 750 1000+ 0 Mortgage debt (£, thousands) 22 Response rates Proportion of refusals/don’t knows Online 2012 Face-to-face Income question 9% 35% Mortgage debt question 5% 29% Mortgage payment problems question 3% 7% 23 Household with internet access By income By age Percentages of households 100 20-30 30-40 40-50 50-60 60-70 70+ Percentages of households 100 80 80 60 60 40 40 20 20 0 0 5-10 10-2020-3030-4040-5050-6060-7070-80 80+ Annual income (£, thousands) 24 Impact of a 1pp rise in interest rates Borrowers vs savers Impact on post-tax income Impact on consumption via cashflow effect By age Per cent 4 Impact on post-tax income Impact on consumption via cashflow effect Per cent 4 3 3 2 1 2 0 1 -1 0 -2 -1 -3 -4 Mortgagors Unsecured borrowers Savers All -2 18-24 25-34 35-44 45-54 55-64 65+ All 25 Benefits of moving the NMG survey online • Cheaper/easier to administer • More accurate reporting of financial distress, not much effect on other variables • Higher response rates • Easier to include more innovative questions • Downside is that not all households have internet access and no opportunity to clarify questions 26 Conclusion • Household surveys appear reasonably, but not perfectly, reliable. • Surveys under-represent high income households, except WAS • All surveys under-record the growth of mortgage debt • NMG survey broadly follows other surveys. Does less well at margin but is better at measuring financial distress • There are advantages to conducting household surveys online – experience with the NMG survey has been positive 27