The determinants of unsecured borrowing: evidence from the BritishHousehold Panel Survey

Ana Del-Rio*and

Garry Young**

Draft Working Paper

* Banco de España. Alcalá, 50. 28014 Madrid, SpainE-mail: adelrio@bde.es** Financial Stability, Bank of England, Threadneedle Street, London, EC2R 8AH.E-mail: garry.young@bankofengland.co.uk

The views expressed in this paper are those of the authors, and not necessarily those of the Bank ofEngland or Banco de España. We are grateful for helpful comments and suggestions to AndrewBenito, Olympia Bover and Merxe Tudela and the participants of seminars held at the Bank ofEngland and Banco de España.

Draft paper circulated for comment. Please do not cite or quote without written

permission of authors.

2

Contents

Abstract 3

1 Introduction 4

2 The theoretical determinants of debt 5

3 Estimating the empirical determinants of unsecured debt 8

4 A cross-sectional analysis of the determinants of unsecured debt 11

4.1 The data 11

4.2 Preliminary Data Description 12

4.3 Estimation results 13

5 Changes in levels of unsecured debt: panel estimation 19

6 Conclusions 21

References 23

Tables and Charts

A: Comparative Statics of Calibrated Model 24B: Number of debt instruments by age group 25C: Individual debt and indebtedness in BHPS 26D: Probit model for credit market participation 27E: Cross section regressions of debt levels 29F: Panel estimation for the change in debt levels 32

1: Debt and indebtedness by age and income 332: Debt and indebtedness by net housing wealth and housing tenure 343: Debt and indebtedness by financial wealth 354: Interest rate spreads over base rates 355: Sample weights by age, income, education, labour status, 36housing tenure and financial wealth6: Predicted unsecured debt by age and income 37

3

Abstract

Household indebtedness has risen sharply in recent years, with large increases in unsecuredborrowing. This paper uses waves 5 and 10 of the British Household Panel Survey (BHPS) toexamine the determinants of participation in the unsecured credit market, the amount borrowed andthe changes between 1995 and 2000, the years in which the surveys were carried out. We estimateprobit models for participation and find that age, income, positive expectations on the financialsituation and housing tenancy status are very significant and have the expected sign according to alife cycle model for consumption. However, regressions to explain the level of borrowing byborrowers show that income seems to be the main variable explaining cross-section differences inunsecured-debt stocks.

The increase in aggregate unsecured debt between 1995 and 2000 does not seem to be linked tochanges in the determinants of credit market participation and has been mainly driven by the largerdebt of participants. The increase in income, educational qualifications and good prospects regardingfinancial situation contributed significantly to this, but part of this process is not explained byindividual-level variables and is, instead, accounted for by common, unmodeled macroeconomicfactors such as credit market conditions and the course of interest rates.

Key words: Unsecured debt, British Household Panel Survey,

JEL classification: C21, D14, E21

4

1 Introduction

Borrowing by UK households has risen rapidly in recent years so that by 2002 the overall amountowed was worth over 120 per cent of annual post-tax income. The majority of this debt is accountedfor by mortgages, secured on the borrower’s home, but the proportion that is unsecured is now wortharound 20 per cent of household income, almost double what it was in 1994.

Unsecured debt differs from secured debt in terms of its purposes, cost, flexibility and risk.Traditionally, the main purpose of unsecured debt has been to finance durable consumption whilesecured debt has financed house purchase, but the purposes for which debt is used are changing.During the 1990s, more use was made of unsecured debt to finance holidays, clothing or specialoccasions1 while secured debt increasingly financed consumption through mortgage equitywithdrawal (see Davey (2001)). These developments suggest that debt is much less closely related toparticular purchases than in the past. Usually, unsecured debt has a relatively higher cost since lackof collateral leads to higher interest rates and shorter terms than secured debt. However, there is alsoan increasing amount of aggregate unsecured debt that does not bear any explicit interest, arisingfrom purchases offering interest-free credit or from the use of credit cards that do not bear interest ifsettled at the end of each month.

This paper attempts to assess what lies behind the greater use of unsecured debt by Britishhouseholds since this potentially has implications for both macroeconomic and financial stability. Itdoes this by means of a detailed investigation of the determinants of borrowing at the individual levelusing information from the 1995 and 2000 waves of the British Household Panel Survey (BHPS).This attempts to clarify the type of factors that influence borrowing and whether these have changedrecently. Is it that people are borrowing more because they feel more confident about the future, or isit simply more convenient for them to finance spending in this way? What are the characteristics ofborrowers and have these changed recently?

Overall levels of borrowing can be understood in life-cycle permanent income hypothesis modelswhere debt allows individuals to smooth consumption over the life cycle and to finance the purchaseof assets such as houses and consumer durables. Changing levels of borrowing can then be explainedwithin this framework as a response to factors affecting individual spending, including relaxation ofcredit constraints and other shifts on the supply-side of the financial market that might influence theway in which spending is financed. In this paper we extend this model to take account of differencesbetween secured and unsecured debt. Because secured debt is cheaper than unsecured debt, secureddebt will be used in preference to unsecured debt when it is available. This points to the importancein the empirical analysis in taking account of the position of the individual borrower in the housingmarket.

1 The weight of durable consumption as the end-use of unsecured debt decreased from near 68 per cent in 1995 to around54 per cent in 2002. Meanwhile, the weight of the financing of holidays, clothes and special occasions increased frombelow 3 per cent of the stock of unsecured debt in 1995 to near 10 per cent in 2002, and that of loan consolidationincreased from around 6 per cent to more than 12 per cent.

5

Empirical analysis on the determinants of debt using household-level data is quite extensive for UShouseholds. For instance, Cox and Japelli (1993) use the 1983 Survey of Consumer Finances (SCF)to estimate a cross-section demand for debt for US households. They find a positive relationship withpermanent income and net worth and a negative relationship with current income and age. Duca andRosenthal (1993) also use the 1983 SCF and find that the demand for debt of young households ispositively related to wealth, income and household size and negatively related to unemployment.Crook (2001) focuses on a more recent period and finds that US households’ demand for debt isrelated positively with income, home ownership, family size, and job status; and negatively to networth, age (when the head of the household is over 55 years old) and risk aversion.

For British households, Bridges and Disney (2002) examine the access to unsecured-credit of low-income households finding that it is positively associated with income-related and income generatingcharacteristics. Banks et al (2002) describe the distribution of British household debt according to theBHPS as a part of a very comprehensive analysis of the distribution of financial wealth in the UK inyear 2000. Cox et al (2002) also use the BHPS to analyse the changes in the distribution ofhousehold debt-to-income ratios, income and assets across borrowers and concluded that the increasein debt of British households during the second half of the 1990s was larger among the youngest andlowest-income households.

The paper is organised as follows. Section 2 extends a standard life-cycle model of consumption totake account of the relationship between secured and unsecured borrowing. Section 3 outlines theempirical method used. Section 4 describes unsecured debt in the BHPS and examines thedeterminants of debt in the cross section. Section 5 focuses on debt changes using the paneldimension in the BHPS. Section 6 concludes.

2 The theoretical determinants of debt

One of the most important characteristics of unsecured debt is that it is expensive relative to otherpossible methods of finance such as secured borrowing or running down asset holding. This wouldsuggest that its use be concentrated among those who do not have access to cheaper finance. In thissection we outline a calibrated version of the life-cycle model of consumption where individuals areable to borrow at relatively low rates against the security of their house, but have to pay higher ratesfor unsecured borrowing. This model captures many of the key characteristics of the UK debtmarket in that household financial decisions appear to be strongly tied to the individual’s position inthe housing market. The model can be used as a framework for understanding the effect of factorswhich vary across individuals, thereby accounting for cross-sectional differences in indebtedness,and factors which change over time.

Individuals are assumed to be economically active for three periods reflecting different phases of thelife cycle. During this time they earn an exogenous income stream and consume non-durable goodsand housing. They aim to maximise intertemporal utility and die solvent. Their intertemporal utilityfunction at the beginning of their lives is given by:

6

( )( )∑

=

−−

+−=

3

1

11

01)1(t

ttt ch

Vδγ

γαα

(2.1)

where h and c are their holdings of housing and consumption of goods respectively, α is a parameterindicating their preference for housing relative to goods, γ is the coefficient of relative risk aversionand δ indicates their rate of time preference.

Individuals face the following flow budget constraint:

st + ut = ptct + qtht + (1+rt-1)st-1 + (1 + rt-1 + ηt-1)ut-1 - qtht-1 - yt, t=1,2,3,4. (2.2)

where s and u are stocks of secured and unsecured debt respectively, y is exogenous nominal non-property income, pt and qt are the prices of goods and housing respectively, r is the rate of interest onsecured debt and η is the premium on unsecured borrowing. It is assumed that all households aim todie with zero net worth, so that at the beginning of the period after their death (at date 4) theproceeds from the sale of the house is sufficient to pay off all remaining debt.

Individuals can use secured and unsecured debt to smooth their spending over time. The use ofsecured and unsecured-debt is assumed to be constrained such that:

st ≤ φt qt ht (2.3)

0 ≤ ut (2.4)

The first expression states that secured debt cannot exceed a proportion, φ of the value of theborrowing individual’s house qh. The second expression states that unsecured debt cannot benegative (so individuals cannot lend at high unsecured interest rates).

The choice of how much secured and unsecured debt to issue is then determined jointly with that ofhow much to consume and how much to spend on housing. Optimal housing and non-housingconsumption is derived by maximising (2.1) subject to (2.2), (2.3) and (2.4). There are three possiblesolutions at any date depending on which of the borrowing constraints are binding. These arereflected in the first order conditions for intertemporal consumption over time (2.5) and the choicebetween housing and goods (2.6) written for the case when the secured debt constraint is binding:

11

1

)1(

)1(

++

+

+++

=∂∂

∂∂

ti

titt

it

it

it

it

p

pr

c

v

c

v

δη

(2.5)

ittitttitt

titt

it

it

it

it

qqqr

pr

h

v

c

v

φηηη

−−++++

=∂∂

∂∂

+1)1(

)1((2.6)

7

In this case, when the constraint on secured debt is binding, both inter-temporal and intratemporalconsumption decisions are affected by the rate of interest on secured borrowing. When the secureddebt constraint is not binding then the premium on unsecured debt drops out of these conditions. Thiswould have the effect of raising current relative to future consumption and changing the effectiverelative price of housing. The third possible outcome is a corner solution where household borrowingwould violate the secured borrowing constraint if consumption choices could be made at securedborrowing rates, but be within the constraint when choices are made at higher unsecured rates.

Depending on the exact specification of preferences, the model can be solved for the optimal timeprofile of consumption of housing and goods that satisfy the budget constraint and the terminalcondition. Iterative methods need to be used since behaviour at any date depends on whether securedborrowing constraints are expected to be binding in the future. The solution is illustrated here bymeans of a model calibrated roughly to the UK situation. The three periods of the model can bethought of as representing fifteen years each.

The results of the simulation are shown in Table A for different scenarios and for two different levelsof the premium on unsecured borrowing. In the first case, income in the first period is £300thousand, consistent with annual income of £20,000. This rises to the equivalent of £40,000 perannum in the second stage of life, before falling back to £15,000 per annum in the last stage of lifewhich includes retirement. The rate of time preference has been set equal to the rate of interest onsecured debt so that in the absence of constraints individuals would smooth consumption over theirlife-cycle by borrowing when young, saving when in middle age and running down their assets in oldage and at death. But the imposition of a limit on secured borrowing up to 90 per cent of the value ofowned housing prevents individuals from reaching this optimum.

In the case where the premium on unsecured debt is 0.1, consumption of goods in the first period isvirtually equal to income and the stock of housing is £54,400, just over 2.5 times annual income.This is financed by secured debt of £48,900, the maximum possible given the secured borrowingconstraint. Unsecured borrowing in the first period is relatively small at £500. After the first period,the borrowing constraint in the model no longer binds so that individuals choose the same level ofconsumption in the second and third stages of life. The pattern of income over time means thatindividuals build up financial assets during the second stage of life and run these down in the thirdstage so that when they die the value of their house is sufficient to pay off their secured debt. Only inthe first period of life is any unsecured debt borrowed.

In the same circumstances, but where the premium on unsecured debt is lower at 0.05, individualsare better able to smooth consumption over time. Consumption is higher in the first stage of life andlower thereafter. This is financed both by higher secured and unsecured borrowing; securedborrowing is higher since individuals choose to buy a larger stock of housing, with the additionalamount effectively financed by unsecured borrowing, which eases the secured borrowing constraintsomewhat. Note that despite the lower rate of interest on unsecured borrowing, those in the secondand third stage of life do not use it because cheaper secured borrowing is available.

8

The second case we consider is of a higher secured borrowing limit of 95 per cent of the value of theindividual’s house, rather than 90 per cent as in the base case. The main effect is in causingindividuals to substitute secured debt for unsecured debt with little or no noticeable effect on firstperiod spending in either case. The reason for this negligible impact is that the change in theborrowing limit does little to alleviate the constraint. In the absence of any restriction on securedborrowing, individuals in the same circumstances would choose to invest £80,000 in housing andborrow £140,000, a loan to value ratio of 175 per cent, using the additional resources to financeconsumption of goods. Hence, the relaxation of the constraint does little to move individuals to theiroptimum position.

The third case shows the behaviour of those who are not owner occupiers (other than to a trivialextent). Consumption smoothing is prevented by the higher rate of interest on borrowing which leadsto less consumption than optimal being chosen in the first stage of life. This is clearly less of aproblem when the premium on unsecured borrowing is lower. In the cross-section, in comparisonwith those who have a stronger preference for housing, unsecured borrowing is higher for those whodo not have access to the secured debt market, although their overall level of borrowing is lowersince they do not have to finance the purchase of a house. Their net worth is also lower since theyhave little taste for tangible assets.

The fourth case illustrates the importance of income expectations on borrowing. With second stageincome expected to be the same as in the first stage of life, it is possible for the individual to smoothconsumption. Note that the stock of housing purchased in the first stage of life is higher than forthose who have higher lifetime incomes but are constrained from borrowing as much as they wouldlike, indicating the effect of the borrowing constraint on the intratemporal consumption decision.

The fifth case shows the effect of less patience. In the high-unsecured premium case, this leads to ahump-shaped path of consumption, with the premium on borrowing causing impatient individuals torestrain their desire for present consumption. Despite this, their unsecured borrowing exceeds theirsecured borrowing. In the low unsecured premium case, there is no hump shape in consumption asindividuals are more able to achieve their preferred consumption path.

The implications of the model are that unsecured debt is likely to be used more by those who areyoung, impatient, with strong income expectations and no access to cheaper secured debt. It is likelyto be used most when unsecured borrowing costs least.

3 Estimating the empirical determinants of unsecured debt

The previous discussion showed that unsecured borrowing is related to the individual’s position inthe life-cycle, access to cheaper, secured finance, income and income expectations and the cost ofunsecured borrowing. In this section, we describe how the determinants of the demand for unsecureddebt may be estimated empirically. Suppose that the demand for unsecured debt by individual i atdate t, Dit, is of the following general form:

9

itittittittiit rZYD εφγβα +−++= (3.1)

where αi is an individual specific fixed effect, Y represents the income and other economiccircumstances of the individual, Z the individual’s demographic and other personal characteristicsincluding age related effects, r the individual-specific interest rate levied on unsecured debt and ε theunobserved determinants of unsecured debt. The coefficients, other than the fixed effect, αi, canpotentially vary over time.

Supply conditions in the unsecured credit market are reflected in the effective rate of interest chargedon unsecured debt. Where individuals are credit constrained, the effective rate can be thought of asbeing high enough that their demand for unsecured loans is equal to the supply. Thus, the effectiveinterest rate might be given by:

itttb

tit Yrr δϕ −+= (3.2)

where rb is the base rate, ϕt is the premium that financial institutions charge to the riskiest individualsand the negative relationship between the effective rate and individual income reflects lowerperceived lending risks at high levels of individual income. Substituting (3.2) into (3.1) then gives areduced form debt function:

ittb

ttittittttiit rZYD εϕφγδφβα ++−+++= )()( (3.3)

This expression helps to clarify a number of points relevant to how it is estimated. First, in cross-sections, it is impossible to estimate the individual specific fixed effects, αi, and the intercept termhas to be imposed at the same value across all individuals. This means that any genuine fixed effectsbecome part of the error term. If these are correlated with any of the right hand side variables thenthe relevant coefficients are biased. Thus if individuals with a particularly high rate of timepreference also choose to work more and so give themselves higher income, then the estimatedcoefficient on income will overstate the genuine effect of income on unsecured borrowing. Second,also in cross-sections, there is no variation across individuals in macroeconomic variables such as thebase rate of interest, rb, as such their impact becomes part of the overall intercept term. Third, withpanel data, it is possible to avoid the biases due to individual specific fixed effects and to identify theimpact of macroeconomic factors, but it is also necessary to assume that coefficients are eitherconstant over time, or have a relatively simple structure.

A key feature of the theoretical model and of the data is that many individuals will choose not to useunsecured debt. Then, the estimation of parameters in (3.3) with a Tobit approach will be highlyinfluenced by participation decisions. Failure to take account of this could lead to the estimatedcoefficients being affected by sample selection bias. For example, estimation of (3.3) for only thoseindividuals with unsecured debt would lead to biased estimates of the true underlying coefficients ifthe sample is partly self-selecting and includes individuals with unobservable characteristics that pre-dispose them to having unsecured debt. Biases would also arise if quantitative credit constraintsprevent some individuals participating in the market.

10

In principle, such biases can be avoided using the suggestion of Heckman (1979) which involvesestimating a model of the credit market participation decision where this is conditioned on factorsadditional to those that determine how much debt to have. Studies similar to ours, such as Duca andRosenthal (1993), Cox and Japelli (1993) and Crook (2001), use a two-step Heckman procedure toobtain parameters of the demand for debt by including two additional terms in the debt-equation tocapture the probability of an individual participating in the credit market and not being creditconstraint. These are estimated in a first step using probit models to estimate the probability ofparticipating in the market and the probability of being unconstrained2. Then the estimated hazardsare included as additional regressors in (3.3), so that the parameters can be interpreted as those of atrue demand function.

The BHPS, unlike the US surveys, does not provide any direct measure that would make it possibleto discriminate between constrained and unconstrained individuals. Although it is unlikely that manyindividuals in the UK are unable to borrow at all. 3 Furthermore, the implementation of the Heckmanprocedure is quite problematic since there is not a strong theoretical case for supplementary variablesaffecting the participation in the credit market without having much effect on the amount borrowed4.In our two-step Heckman procedure, we have added regional dummies, race and the interaction ofemployment with the occupational sector to the participation equation as factors that might influenceparticipation in the unsecured debt market without having much effect on the amount borrowed.However, the marginal effects in the Heckman approach are exactly the same as in the simple OLScross-section regressions. Given this and the results in previous studies we will focus separately onthe participation equation (using a probit model) and on debt equations, using simple OLS cross-section regressions including only debtors.5

2 According to Japelli (1990) credit constraints affect 19 per cent of households and this number raise to 30 per cent foryoung households (Duca and Rosenthal (1993)). Other related works are Hayashi (1985), Zeldes (1989), Cox andJappelli (1993), Crook (1996), Gross and Souleles (2002). For the UK, Davies and Weber (1991), using household-leveldata and identifying unconstrained households as those with savings, found evidence of liquidity constraints but not ofloosening credit contraints in the 1980s. Bayoumi (1993) found that softer liquidity constraints due to financialderegulation during the 1980s had a significant effect on U.K. consumption. More recently, Fernández-Corugedo andMuellbauer (2002) estimated an index of non-price credit conditions providing evidence of looser credit restrictionsduring the 1980s and second half of the 1990s.3 One possible proxy is given in wave 5 when individuals state whether they think that it was a right time to use credit inthe hypothetical case that they wanted to buy something bit. One of the possible answers to this question was ‘Can’t getcredit’ that was only selected by 2.4 per cent of 8,774 respondents.4 Cox and Japelli (1993) used years of education, occupation, area income, employment status, and rural/urban status assupplementary variables for the probability of having positive debt. Duca and Rosenthal (1993) and Crook (1996)assumed the same variables for determining the probability of having debt and the amount borrowing (allowing fordifferent parameters in the participation and debt equations).5 Results using the Heckman procedure are available upon request.

11

4 A cross-sectional analysis of the determinants of unsecured debt

4.1 The data

The BHPS6 is an annual national survey of the economic and demographic characteristics of Britishindividuals and households. The first wave covered a representative sample of the population ofGreat Britain in 1991. And this sample has remained broadly representative since the sameindividuals are re-interviewed each year and, if they split-off from original households, all adultmembers of their new households are also interviewed. In 1991 the survey included around 5,500households and 10,300 individuals (aged over 16 years old).7

Information on unsecured debt and financial assets is available only in waves 5 and 10 of the BHPScovering 1995 and 2000. Individuals are asked about the overall amount they owe excluding creditcard and other bills being paid off in the month of the interview.8 They are shown a card to promptthem about the forms in which they may have borrowed. In 1995, the prompt card contained thefollowing list of debt instruments: hire purchase agreements, personal loans (from bank, buildingsociety or other financial institution), credit cards, catalogue or mail order purchase agreements,DSS Social Fund loan, any other loan from a private individual, or anything else. In 2000, twoadditional instruments, overdrafts and student loans, were added to this list. This change in the list ofinstruments casts doubt on the comparability of responses across the two waves of the survey.Respondents could have included borrowing in these additional forms of debt in their answers to thesurvey in 1995 without being prompted. For example, they could have considered borrowing onoverdrafts as a form of personal loan. But the change in question must leave room for doubt that thiswas the case. As shown in Table B, in terms of quantity rather than value, overdrafts representednearly 7 per cent of the total number of credit instruments mentioned in 2000. Student loans were aless significant 1 per cent of total debt instruments. If borrowing using these instruments wereentirely omitted in 1995, but not 2000, then analysis would overstate the increase in unsecuredhousehold debt. There is some evidence against this in that Redwood and Tudela (2003) find thatunsecured debt is more underreported relative to aggregate figures in 2000 than in 1995. This wouldsuggest that the new listed instruments in 2000 were counted in other categories in 1995. Throughout

6 The British Household Panel Survey (BHPS) is being carried out by the ESRC UK Longitudinal Studies Centre withthe Institute for Social and Economic Research at the University of Essex. Detail information can be found in Brice(2002) available at http://www.iser.essex.ac.uk/bhps/.7 The sample excludes north of the Caledonian Canal in Scotland. Since 1997 new samples have been added to the BHPSaimed at extending the coverage of some particular regions and groups of population. We exclude them to keep thesample representative of the British population.8 If individuals do not know the exact amount they owe, they are asked to provide whether it is more than £100, morethan £500, more than £1,500, more than £5,000. Depending on the case we assign a debt of £50, £300, £1,000, £3,250 or£7,000. This affects 310 borrowers (out of 6,889). If individuals report that the debt is a joint commitment we assign halfof the value. In 2000 we can also know which part of the debt is a sole commitment but we do not use this informationsince it is not available for 1995. Joint commitments affect 984 and 709 individuals out of 3,481 and 3,458 debtors in1995 and 2000 respectively.

12

this analysis we assess the sensitivity of estimates to this potential problem by changing the samplein 2000.

4.2 Preliminary Data Description

Table C shows that the proportion of individuals reporting that they had any unsecured debt did notchange between 1995 and 2000 with around 39 per cent of individuals who answered this questionclaiming to have at least one form of unsecured debt in both years.9 Significantly, among those withsome debt the mean amount almost doubled from £1,718 in 1995 to £3,272 in 2000. Indeed,unsecured debt approximately doubled at most points of the distribution with the median rising from£700 per debtor in 1995 to £1,500 in 2000 and the 90th percentile rising from £4,000 to £8,000.

Indebtedness, defined as the unsecured debt to income ratio, also rose at most points of thedistribution. For individuals with some debt, the median increased from 8 per cent to 12 per cent andthe indebtedness of those at the 90th percentile rose from around 45 per cent in 1995 to nearer 80 percent in 2000.

As suggested by the life-cycle model, there are clear differences in credit market participation byage. In both 1995 and 2000, nearly 60 per cent of individuals aged 20 to 35 years old had at least oneform of unsecured-debt (see upper-left panel in Chart 1). This fraction decreases with age to 10 percent for individuals older than 60 years old. By contrast, differences across age groups in averagelevels of debt and debt-income ratios for borrowers are not very large, especially in 1995.

The increasing relationship between credit market participation and income is similar in both 1995and 2000 (see right panels in Chart 1).10 The unsecured debt-to-income ratio is fairly constant acrossall but the lowest income groups who had the highest levels of indebtedness.11 In 2000, the level ofborrowing is similar for those below the median income level. Debt levels and indebtedness arehigher for all groups in 2000.

Chart 2 shows the relationship between unsecured debt and various measures of secured debtcapacity. There is a clear negative relationship between participation in the unsecured debt marketand net housing wealth12, but the relationship between the amount borrowed and net housing wealthis less clear. In 1995 indebtedness appears to be independent of net housing wealth. This increasedmost between 1995 and 200 for those with low housing wealth such that there is a slight decreasingrelationship in 2000.

9 About 5% of individuals did not answer this question in both 1995 and 2000.10 Income groups are deciles of the income distribution of the total sample in 1995. In 2000 decile values are updatedwith the Retail Price Index.11 In Cox et al (2002) indebtedness seems to be negatively correlated with age and income. Discrepancies can arise sincetheir study focuses on households, not on individuals, and income variables and groups can differ. In addition ouranalysis excludes all new samples in BHPS since 1997.12 Net housing wealth is the value of the residential house net of mortgages. Since these are household variables in BHPSwe assign half value of the house and mortgage to the first and second person owning the accommodation.

13

Looking at housing status, owner-occupiers with a mortgage have a higher propensity to holdunsecured debt than other groups in both 1995 and 2000. They also have higher amount of unsecureddebt than other groups, although this increased for all groups between 1995 and 2000. Further,among households with mortgages, there seems to be a positive correlation between the level ofunsecured debt and the secured-debt-to-income ratio, especially in 2000. This general relationship isconsistent with the theoretical model of Section 2, although it is not clear why those with relativelylow secured debt to income ratios have any unsecured debt.13

As regards financial wealth14 there seems to be a negative relationship between total financial assetsand the fraction of debtors (see Chart 3).15 In 2000, there is a relatively clear decreasing relationshipbetween unsecured debt-to-income ratios and financial wealth.16 Those with a low level of financialassets are more likely to hold unsecured debt to finance their consumption.

All these figures point to a quite generalised increase in average unsecured-debt of borrowersbetween 1995 and 2000 while no large changes appear to have been in market participation.

4.3 Estimation results

While the preceding section provides a broad overview of unsecured debt and its correlation with thecharacteristics and circumstances of individual borrowers, a major limitation is the inability todisentangle the independent contributions of individual factors. In this sub-section, we useregression-based analysis to assess statistically the key factors determining participation and theamount borrowed in the unsecured debt market.

The explanatory variables in the debt and participation equations include dummies for age to takeinto account the life cycle stage of individuals and variables aimed at explaining the effect of currentand expected income on consumption and borrowing. These variables are actual income, educationalqualifications17 and whether the individual expects an improvement in his financial situation.Employment status is included to proxy income uncertainty. Some of the income-related variablescould also have an additional effect in the reduced form through their impact in the unsecured-debtpremium if they are correlated with the risk of default. We also include dummies to take into accountwhether individuals have access to the mortgage market and, in the case of mortgagees, distinguishthem by the level of mortgage-to-income ratio. Gross financial wealth is included in the form of a

13 This apparent relationship and the relative increase in the fraction of debtors for those most mortgage-indebted mightbe related with the introduction of exemptions of mortgage indemnity insurance premia in 1998 when loan to value ratioswere below 0.9. According to Fernández-Corugedo and Muellbauer (2002) the greater incentive of some individuals toincrease the unsecured component of borrowing raised the long-run stock of aggregate unsecured credit by 8 per cent.14 This variable does not included assets in the form of pension funds or insurance products.15 Financial wealth groups are percentiles for those with positive financial assets. We consider separately those with nofinancial wealth. We assign half values if savings are hold jointly.16 The same pattern is observed when considering only liquid financial assets.17 We consider three groups. The first one corresponds to individuals with the highest educational qualification being ahigher degree, first degree, teaching QF, other higher. Medium qualified are those with nursing, GCE A Levels, GCE OLevels or Equivalent, Commercial QF, CSE Grade 2-5,Scot G, Apprenticeship and Other QF.

14

dummy variable distinguishing those with no financial wealth and those with financial wealth belowand above the median in the sample population respectively. Finally we add other demographicvariables such as whether the individual is head of the household, gender, marital status and numberof dependent children. Other idiosyncratic differences in household preferences and interest rateswill be reflected in the error term because as noted earlier with cross-section data we cannot separatethe random and systematic component of the residuals.

Table D reports the estimation results of the probit model of credit market participation and Table Ereports the estimation results for the amount of unsecured debt of those who participate. For a bettercomparison of the changes in the parameters between 1995 and 2000 we include all variablesinteracted with the year (instead of carrying out two separate estimations). In the bottom lines ofeach table we present the p-values for the null hypothesis of equal parameters in 1995 and 2000 forsome selected variables. In addition, to ensure comparability between 1995 and 2000, we present twosets of results. In the participation equation, there are separate results including and excludingindividuals who report that their only form of debt is student loans or overdrafts.18 These types ofindividuals might have been included as debtors in 1995 depending on how they interpreted the debtcategories listed in the questionnaire in 1995. In the equation for the amount of unsecured debtModel 2 includes dummy variables interacting with age for individuals with overdrafts and studentloans.19 In the latter model, the combined effect of overdrafts and student loans in 2000 figures isdifficult to isolate given the strong correlation between having overdrafts and being a large debtor.Hence, dummy variables in Model 2 do not just capture debt in the form of overdrafts but therelatively large amount of debt of those holding overdrafts.

Since most of the explanatory variables take the form of dummy rather than continuous variables, theestimation is relative to a ‘reference group’ for whom only the constant term is evaluated. Thereference group appears in parenthesis in a separate column.20 According to the estimated probitmodel the probability of having unsecured debt for the reference group was 0.53 in 2000 and wecannot reject the null hypothesis that this probability is unchanged from that in 1995.

The age profile in the participation equation is consistent with the theoretical life-cycle model ofconsumption in that the probability of having debt decreases with age for all but those aged 16 to 20.Indeed, according to the estimated marginal effects, age is the variable that most strongly affectsparticipation. The results suggest that the probability of participating in the credit market is 25 to 30percentage points lower for individuals older than 60 years old compared to those aged 20 to 30.Those aged 16 to 20 have a significant lower probability of having debt (around 20 p.p.), perhaps

18 This reduces the sample by 185 observations, of which 147 are individuals reporting all their debt in overdrafts.19 See Table B to see the distribution of these instruments by age. We also made estimations with interactions withincome dummies and results were not altered qualitatively.20 White males, head of household, living in couple, with no dependent children, living in Inner London, aged 20-30,with a high level of education, employed, whose house is owned outright with a value below the 30th percentile in thesample population, with income between the 50th and 70th percentiles and no financial wealth. The head of the householdin the BHPS is the principal owner or renter of the residence, and if more than one potential head it is assigned to theeldest.

15

reflecting their economic immaturity. The age profile did not change significantly between 1995 and2000 except for the oldest group whose probability of participation rose by 8 percentage points21.

The impact of age on the amount of unsecured debt of those who are borrowers is made unclear bythe confusion over whether overdraft and student loan borrowing was included in survey responsesin 1995. The basic results for 2000, shown as Model 1, do indicate that the amount borrowed issignificantly lower for older age groups, with the absolute value of the negative coefficientincreasing with age. This is in contrast to the results for 1995 which show that only those youngerthan 20 years old have significantly less debt than the reference group, in line with the raw datawhere the amount of unsecured debt of borrowers does not appear to vary systematically by age.While the change in coefficients between 1995 and 2000 is not statistically significant atconventional levels of significance, it is suggestive of an increase in the relative indebtedness of the20 to 30 year olds in the reference group. But it is not clear whether this is due to a change in theirbehaviour or a change in the survey question. Some evidence on this is gained by including dummiesfor individuals with overdrafts and student loans in 2000. In this case, shown as Model 2, thesignificance of the negative coefficient for individuals aged 30 to 60 disappears, suggesting that thereis no discernible age effect in 2000 for those who do not have student loans or overdrafts. Oneinterpretation of these results is that age effects are more apparent when individuals with studentloans and overdrafts are included in the sample since such borrowing is done predominantly byyoung people, this would be consistent with under-recording of unsecured debt in 1995 by those withoverdrafts and student loans. For individuals older than 60 years old the amount of debt issignificantly lower than the reference group in year 2000 in both Models 1 and 2 indicating that theincrease in average debt between the period 1995 and 2000 has been lower for these individuals thanfor the reference group.22

Differences in income also introduce significant differences in the probability of participating in thecredit market. Individuals with income below the 30th percentile have a significantly lowerprobability of having any unsecured debt, with an even lower probability for those below the 10th

percentile. The null hypothesis that the coefficients of each income group are similar in 1995 and2000 cannot be rejected at 5 per cent of significance. However for low-income individuals thishypothesis is accepted with a low probability.

There is also a strong positive relationship between the amount of unsecured debt and income.23 Thesize of coefficients show that income is the main variable explaining differences in unsecured debt.Comparing the results for 1995 and 2000, there is little change in the estimated coefficient for thosein the main body of the income distribution. For the lowest-income group the hypothesis of equalcoefficients in 1995 and 2000 is rejected at 10 per cent of significance in Model 1 and suggests more

21 The probability of accepting the null hypothesis that the coefficient for this variable is equal in 1995 and 2000 is0.1342 (See bottom lines in Table D).22 Note that this result is not very robust because the probability of accepting the null hypothesis of equal coefficient in1995 and 2000 for this age group is still quite high: 0.35.23 Alternatively, we also included income and the square of income as explanatory variables. Qualitative results were thesame.

16

relative demand for debt among those with low incomes or less discrimination against them withrespect the reference group. At the same time, the tendency of high-income groups to have higherunsecured debt levels is accentuated in 2000.

Positive expectations of the individual’s future financial position are also associated with a largerprobability of participation in the credit market. The marginal effect of this variable is stronger in2000 than in 1995 suggesting that the confident are more willing or able to borrow in 2000 than 1995(see p-values at the end of Table D). Good economic prospects are also important in determining theamount of unsecured debt. Table E shows that the dummy variable for positive expectations aboutthe future financial situation is highly significant and positive in both 1995 and 2000 with similarcoefficient.

The lack of educational qualifications is also associated with a lower probability of having unsecureddebt. In particular, individuals with no educational qualifications have around 10 percentage pointslower probability of having debt than high-educated ones. Qualification dummies also indicate thatthe higher the educational qualification the higher the debt level. There is no strong evidence of achange in the link between qualifications and indebtedness between 1995 and 2000.

As regards labour status, the retired do have a lower probability of having debt, around 10 percentagepoints lower than for the employed in 1995. This is in addition to the age effect and suggests thatretired people are less likely to have unsecured debt at every age than those in work. They also tendto have less unsecured debt when they borrow, but this appears to be much less important in 2000than in 1995. The unemployed also have a lower probability of participating than the employed,consistent with the possibility of being credit constrained, as well as with their greater uncertainty offuture income. Interestingly, as with the retired, the amount of unsecured borrowing by theunemployed, keeping constant all other characteristics, is lower relative to the reference group in1995 but not 2000. This is consistent with the more depressed labour market in 1995 when theunemployed would have faced more uncertainty about their prospects, while in 2000 unemploymentmight have been considered more of a temporary problem. Another possible interpretation would bethe presence of looser credit restrictions for unemployed in 2000. It is worth noting, however, that in1995, individuals unemployed for more than a year had significantly less debt, this was not true forthe short-term unemployed who might need time to restructure their debt balances according to theirnew situation.

The self-employed are less likely to borrow in the unsecured credit market but have more debt whenthey do. This might be because they have other sources of finance, but greater general demand forfinance for business reasons when they participate in the market.

The interpretation of the results for full time students is complicated by the change in the wording ofthe question in the survey.24 They appear to have had a lower propensity to participate in the credit

24 In 1995, full time students have near 50 per cent of their debt commitments in personal loans. In 2000, near 30 per centwere overdrafts and another 30 per cent was classified as the category something else in the show card.

17

market than the employed in 1995, but this may be because they were not asked to include studentloans in their answer to the question. When student loans and overdrafts are included in 2000, theestimated marginal effect of being a full time student is positive but insignificant, suggesting that thelikelihood of full-time students participating in the unsecured credit market is the same as for otherindividuals with the same characteristics.

For those who do participate, full-time students had a relatively larger demand for unsecured debt inboth 1995 and 2000, with similar marginal effects in the two years. This is likely to be due to theirbetter prospects and access to cheaper credit relative to their peers of similar observablecircumstances.

Housing tenure and access to secured debt affect both the probability of having unsecured debt andthe amount borrowed. In 1995, those who are not owner occupiers and those with a mortgage hadaround a 15 percentage points higher probability of having unsecured debt than those living inhouses owned outright. Differences in the ratio of mortgage-debt to income produce a slight hump-shaped pattern with respect the probability of being in the market with the highest participation ofthose with mortgage debt to income ratio around the median. In terms of the amount borrowed, onlythose with secured debt to income above the eightieth percentile of the distribution had significantlymore debt in 1995 and 2000, consistent with these individuals having used up cheaper sources offunds.

There are also differences among those without a mortgage but with housing equity. There appears tobe some evidence that people would prefer to borrow in the unsecured market than re-enter themortgage market. In 2000 those who had no mortgage and were in the top percentile in terms ofhousing equity had more unsecured debt than the reference group. This could reflect the presence ofcosts when withdrawing equity from the house value and the relatively low interest rate of someforms of unsecured debt.

Financial wealth is also an important determinant. Consistent with Banks et al. (2002) those with nofinancial wealth (reference group in the regression) are more likely to have unsecured debt than thosewith the largest financial assets holdings. However, having a moderate amount of financial assets isassociated with a larger probability of having some debt, especially in 2000. In terms of the quantityof debt, the empirical results indicate that those with financial assets have lower amounts ofunsecured debt. Between 1995 and 2000 there has been an increase in the probability of marketparticipation for those with financial wealth, but the increase in the amount borrowed by those withabove median financial wealth has been relatively lower.

Finally, the increase in the constant term in the debt level equation between 1995 and 2000 impliesan increase in the expected debt of the reference borrowers. Given the small changes in thecoefficients between 1995 and 2000 this increase can be considered as generalised and probablyexplained by changes in the macroeconomic environment and credit market structure that cannot bemodelled appropriately with our data set. On the other hand, the evolution of the explanatoryvariables is another key aspect to understand the increase on unsecured debt between 1995 and 2000.

18

With respect credit contract terms it is interesting to stress that both interest rates and the increase incompetition in the credit market might have played an important role that, unfortunately, cannot beaddressed with the BHPS. Some evidence of a reduction in the unsecured debt spreads and of longeraverage terms of consumer credit is shown in Chart 4. As can be seen, the spreads differ largely bytype of debt instrument and the reduction of spreads during the second half of the 1990s does notaffect all instruments equally. While the interest rate spread of personal loans (of more than £10,000)on the retail bank base rate has experienced a significant reduction from around 10 points to 6 points,in the case of overdrafts it has fluctuated between 12 and 10 points. As regards the maturity ofunsecured debt, the weight of personal loans with original maturity greater than four years increasedfrom 22 per cent in 1995 to 35 per cent in 2000. However the number of loans with original maturitybetween one and four years decreased from 64 per cent to around 50 per cent. Whether this is ademand or supply effect is something we do not know but it has allowed individuals to sustain higherlevels of debt without increasing regular re-payments of debt.

Regarding the evolution of explanatory variables Chart 5 shows the sample weights of differentpopulation groups by some of the key variables determining credit market participation and debtamount. As it can be seen, there are some important shifts in the population characteristics that mighthave affected the stock of aggregate debt. In particular the shift in the number of individuals towardshigher income groups (in real terms) is quite large as well as the increase in the proportion ofindividuals with high qualifications. These two characteristics are related with both a higherparticipation and amount of debt and can also contribute to explain the increase in the amount ofunsecured debt observed in an aggregate form.

A useful way of summarising the estimation results is to show the fitted value of debt conditional onhaving debt. Chart 6 shows the predicted age and income profile of unsecured-debt levels for thereference group. In contrast to the simple patterns that might be present in the raw data, theregression approach makes it possible to hold constant all other factors (such as housing tenure,labour market status) that might also vary with income and age. A quite flat hump-shaped profilewith respect to age is observed and the increase in the expected debt from 1995 is considerable for allage groups. With respect to income, the expected debt of borrowers increases very rapidly.

Summarising, cross-section differences in the demand for debt by borrowers is highly determined byincome and good financial prospects, while the impact of age, employment status and housingtenancy status are either not clearly consistent with the model or are not highly significant. Onepossible explanation is that results are distorted due to the large variety of debt instruments includedas unsecured debt in terms of maturity, interest rates and purposes such as the financing of expensivedurable consumption, human capital investment and non-durable consumption. In addition, life cyclepatterns might be obscured by the increasing relevance of borrowing against housing wealth tofinance non-housing consumption25.

25 When we use a two-step Heckman procedure the results are exactly the same if we focus in the marginal effect of eachvariable in the amount of debt. In 1995, the inverse of Mill’s ratio, included to control for possible sample selection bias,is not significant, and therefore estimated coefficients are the same as marginal effects. In 2000, it becomes negative and

19

When comparing 1995 and 2000 we do not observe substantial changes in the factors affecting creditmarket participation, with only a relatively larger increase in the probability of having unsecureddebt of those aged over 60 years old, self employed, with good financial prospects and positivefinancial wealth. When comparing expected values of debt between 1995 and 2000 we see that themarginal effects are larger for individuals in the tails of the income distribution in 2000, retired andunemployed. While they are lower for those aged over 60 years old and self-employed. In addition,the increase in debt is a quite generalised phenomenon captured to a large extent by the largeincrease in the constant in the equation. This suggest that the change in interest rates levied onunsecured debt and the change in other aspects of these contracts, such as the maturity, might beimportant factors explaining the higher level of debt of debtors in 2000. Equally important, theevolution of explanatory variables suggests that the shift of individuals towards higher incomegroups and high qualifications, which are positively related with the amount of debt, is also animportant factor explaining the increase debt levels in aggregate terms.

5 Changes in levels of unsecured debt: panel estimation

Previous analysis made no use of the fact that the same individuals are present in both the 1995 and2000 samples so that their behaviour may be tracked over time. In this section we discuss evidenceon changes in the borrowing of the same individuals.

Table C shows that despite the apparent stability in credit market participation and general increasein the amount borrowed, there is considerable change in the position of individual borrowers as theymove into and out of debt. Among those in the top quartile of debt in 1995, 41 per cent of them werestill highly indebted in 2000, but 35 per cent of them had reduced their debt to zero. There appears tobe more persistence at low debt levels in that 78 per cent of people with no unsecured debt in 1995still had no debt in 2000, whereas only 8 per cent of them had moved to the top quartile.

We now use regression analysis to find out whether the changes in debt levels of individuals between1995 and 2000 are related to changes in the determinants of debt. Differencing (3.3), assuming nochange in the coefficients across the two years gives:

marginally significant, and this could suggest that the unobservable characteristics that induce people to participate in theunsecured credit market are negatively correlated with the unobservable factors determining how much they borrowhaving entered the market. This is the opposite of the effect that would be obtained if we think that market participationdecisions are affected by the presence of entry costs. Then this term should be positive, as in Crook (2001), to indicatethat those being in the sample of borrowers tend to demand larger funds. In our case, the negative term in 2000 meansthat the unobserved factors in the equation for positive debt are negatively related with the amount of debt. One possibleexplanation for this effect is that the unobservable (to the econometrician) characteristics of those who are the mosteligible borrowers from the perspective of lenders also reduce the amount that these individuals wish to borrow.However, it is not clear why this factor should be stronger in 2000 than 1995. An alternative possible interpretation isrelated to the presence of costs when withdrawing equity from the housing value. These costs may induce individualswith a higher demand for funds to borrow in the secured debt market while only those with a lower demand for fundswould borrow in the unsecured-debt market, even though they have to pay a premium. This effect would be stronger inthe stronger housing market conditions of 2000.

20

ittb

tititit rZYD εϕφγφδβ ∆++∆−∆+∆+=∆ )()( (5.1)

Thus, in general terms, the change in the unsecured debt of an individual embodies a general ageeffect as the individual moves further along the life-cycle and responds to changes in economiccircumstances (reflected here in the income term), changes in personal characteristics and changes inmacroeconomic conditions (reflected here in the change in interest rates).

We estimate this equation using the change in the level of unsecured debt as the dependent variable.Among the dependent variables we include the change in income allowing for a non-linear effect, thechange in educational qualifications,26 in labour status, in financial wealth and in the mortgage stock.The changes in the labour status are considered with different dummy variables indicating whetherthe individual found a job, whether he became unemployed, among others combinations of thedifferent labour status considered previously. The changes in the mortgage stock are split up in fourcategories. We consider separately the change in the mortgage of those increasing the stock ofsecured debt, the change of those decreasing the mortgage and dummy variables for those who havethe same mortgage in 1995 and 2000. We also include age dummies to capture the stage of the lifecycle and a dummy variable to consider whether the individual expects an improvement in hisfinancial situation. The constant term picks up the effect of general changes in macroeconomic andcredit market conditions relevant to the reference group who consist of individuals between the ageof 20 and 30 in 1995 who are employed in both years.

The estimation results are presented in Table F. Two adjustments are included to deal with thechange in survey question. In the first case we include a dummy variable for those with overdraftsand student loans in 2000. In the second case we exclude individuals with student loans andoverdrafts in 2000.

The results tend to bear out those found in the cross section although the effect of age is much moreapparent here once the individual specific fixed effects have been differenced out.27 Those agedbetween 45 and 60 reduced their unsecured debt by over £500 more than those between 20 and 30,while those over 60 reduced their debt by over £700 more than their younger counterparts.

The increase in income is statistically significant in explaining the increase in debt. When separatingthis variable between low and high-income individuals,28 this effect is larger for those with lowincomes in 1995 consistent with a non-linear effect.

As in the cross section results, there is clear evidence of the effect of expectations on unsecuredborrowing. Expectations of a better financial situation have a significant positive impact. Similarly

26 This variable is the change in a variable that takes the value 3 for high educational level, 2 for medium and 1 for noqualifications.27 Note that we keep the same age groups as in previous analysis. Since age dummies enter contemporaneously in thepanel equation the youngest individuals are 20 years old, that is, they were 16 years old in 1995.28 High (low) income individuals are those above (below) the median income distribution of 1995.

21

an increase in educational qualifications, likely associated with improved economic prospects, raisesunsecured borrowing.

Changes in the labour market position of the individual also have an impact on borrowing. The effectof being unemployed in both 1995 and 2000 appears to offset the general upward trend in borrowing,although this effect is not precisely determined statistically. Becoming employed or self-employed(from being without work) has a significant positive effect on the change in debt.

The impact of changes in the secured debt market is again surprising in that there is evidence ofincreasing unsecured debt among those reducing their mortgage. This bears out some of the crosssection evidence that unsecured borrowing is being undertaken by those who have unused secureddebt capacity.

As with the cross-section results, there is clear evidence of a general increase in unsecured debtcaptured in the constant term in the equation, worth £600 per individual. This is close to the medianincrease of £800 of those with unsecured debt in the whole sample, indicating that this factor is mostimportant in explaining the general increase. This effect is present even when individuals withstudent loans or overdrafts are excluded from the comparison.

6 Conclusions

According to the BHPS, the proportion of individuals with unsecured debt did not change between1995 and 2000 with around 39 per cent of individuals claiming that they had least one form ofunsecured debt. However, the average level of unsecured debt held by borrowers more than doubledduring this period of five years.

We use waves 5 and 10 of the BHPS to examine the determinants of credit market participation andthe amount of unsecured debt in 1995 and 2000 and assess whether these have changed between thetwo years.

Participation in the unsecured credit market tends to decrease with age and it is significantly lowerfor the retired and unemployed, those with no qualifications, and income below the median. It alsotends to be higher for those with optimistic expectations of their financial position and those withoutaccess to the mortgage market.

The main variable explaining cross-section differences in levels of unsecured debt appears to beincome, but also good prospects for the financial situation. The effect of age is less clear-cut andappears to be less important than theory would suggest.

There has been little change in the relationship between secured and unsecured borrowing over thesetwo years. Those with a high secured-debt-to-income ratio also tend to have a high demand forunsecured since they might have used up cheaper sources of funds. Contrary to what we expected,non-owners occupiers did not have a higher amount of unsecured debt since they probably areperceived as more risky, having access to unsecured-debt instruments with a higher premium. For

22

given circumstances, the self-employed and full time students had more unsecured debt than othergroups, though for students comparisons between 1995 and 2000 are sensitive to the introduction ofnew listed instruments in the 2000 questionnaire. Finally there appears to have been a generalincrease in debt that cannot be accounted for by cross-section differences. We show that the courseof interest rates could have also contributed to the increase in debt, but a longer longitudinal datawould be needed to assess their impact. The simulation model showed that a fall in the relative costof unsecured borrowing would raise unsecured borrowing among those participating in the marketand affecting marginally those who do not participate. This is the pattern we have observed inpractice.

On the other hand, the positive relationship between the amount of debt and income and educationalqualifications suggests that the increase in income, education qualifications and good prospects onfuture financial situation contribute significantly to explain the increase in unsecured-debt. This isalso confirmed in the panel analysis.

23

References

Banks J, Smith Z and Wakefield M (2002), ‘The Distribution of Financial Wealth In the UK:Evidence From 2000 BHPS’, The Institute for Fiscal Studies, Working Paper 02/21.

Bayoumi T (1993), ‘Financial Deregulation and Consumption in the United Kingdom’, The Reviewof Economics and Statistics’, pages 536-39.

Brice J, Buck N and Prentice-Lane E (2002), British Household Panel Survey User ManualVolume A: Introduction, Technical Report and Appendices. Taylor, Marcia Freed (ed). Colchester:University of Essex.

Bridges S and Disney R (2002), ‘Access to credit, and debt, among low income families in theUnited Kingdom: an Empirical Analysis’, Mimeo, available at:http://www.nottingham.ac.uk/economics/ExCEM/publications/index.html

Cox D and Jappelli T (1993), ‘The effect of Borrowing Constraints on Consumer Liabilities’,Journal of Money, Credit and Banking, 25, pages 197-203.

Cox P, Whitley J and Brierly P (2002),‘Financial Pressures in the UK Household Sector: Evidencefrom the British Household Panel Survey, Bank of England Quarterly Bulletin: Winter 2002.

Crook J (2001),‘The demand for Household Debt in the USA: Evidence from the 1995 Survey ofConsumer Finance’, Applied Financial Economics, 11, pages 83-91.

Crook J (1996), ‘Credit Constraints and US households’, Applied Financial Economics, 6, pages477-485.

Davey M (2001), ‘Mortgage Equity Withdrawal and Consumption’, Bank of England QuarterlyBulletin, Spring 2001, pages 100-103.

Davies A J and Weber G (1991), ‘Credit and British Consumers: Some Micro Evidence’. FiscalStudies, May, 12(2), pages 61-84.

Duca J V and Rosenthal S S (1993), ‘Borrowing constraints, household debt, and racialdiscrimination in loan markets’, Journal of Financial Intermediation, 3, pages 77-103.

Fernández-Corugedo E and Muellbauer J (2002), ‘Modelling Consumer Credit Conditions in theU.K.’, Bank of England. Mimeo.

Gross D B and Souleles N S (2002), ‘Do Liquidity Constraints And Interest Rates Matter ForConsumer Behavior? Evidence from Credit Card Data’, Quarterly Journal of Economics, 117(1),pages 149-85.

Japelli T (1990), ‘Who is credit constraint in the US economy?’, Quarterly Journal of Economics,pages 219-234.

Redwood V and Tudela M (2003), ‘Grossing up of Dissaggregate Household Data from the BHPS:How does this Match with Aggregate Data?’, Bank of England, mimeo.

Zeldes P S (1989), ‘Consumption and Liquidity Constraints: An Empirical Investigation’, Journal ofPolitical Economy, Vol 97(2), pages 305-346.

24

Table A: Comparative Statics of Calibrated Model£,thousands

Period c h s u y c H s u yMain case:

1 295.1 54.4 48.9 0.5 300 325.8 66.1 59.5 32.4 3002 409.8 93.5 -86.8 0 600 385.6 87.9 -71.5 0 6003 409.8 93.5 71.9 0 225 385.6 87.9 67.6 0 225

Higher secured borrowing limit ( φ = 0.95)1 297.3 54.8 52 0 300 325.8 66.1 62.8 29.2 3002 408.2 93.1 -85.8 0 600 385.6 87.9 -71.5 0 6003 408.2 93.1 71.6 0 225 385.6 87.9 67.6 0 225

Low ownership of housing ( α =0.0005)1 315.4 0.55 0.5 15.5 300 345.5 0.66 0.6 45.6 3002 424.4 0.92 -152.9 0 600 401.9 0.87 -135.6 0 6003 424.4 0.92 0.7 0 225 401.9 0.87 0.67 0 225

Low income expectations1 267.1 60.9 28.1 0 3002 267.1 60.9 3.6 0 3003 267.1 60.9 46.9 0 225

Less patience ( δ=0.4)1 376.9 69.4 62.5 83.8 300 410.8 83.4 75.1 119.1 3002 389.2 88.8 7.1 0 600 361.4 82.4 18.9 0 6003 289.3 66 50.8 0 225 268.7 61.3 47.1 0 225

Main case: δ = 0.3, α = 0.05, γ=0.25, φ = 0.90.

Unsecured premium = 0.1 Unsecured premium = 0.05

25

Table B: Number of debt instruments by age group

agehire purchase

personal loan

credit cards

mail order purchase

DSS Social Fund loan

loans from individuals

something else overdraft

student loan Total Total (%)

16-20 18 74 30 74 11 15 5 - - 227 4.7%20-25 110 258 144 136 19 33 8 - - 708 14.7%25-30 146 239 189 149 17 28 7 - - 775 16.1%30-35 192 210 215 179 12 24 3 - - 835 17.3%35-40 130 166 157 128 7 14 7 - - 609 12.6%40-45 115 138 150 89 10 6 5 - - 513 10.6%45-50 101 120 139 95 2 9 6 - - 472 9.8%50-55 59 73 104 63 3 5 4 - - 311 6.5%55-60 49 31 48 25 1 1 - - 155 3.2%60+ 58 26 76 47 1 3 4 - - 215 4.5%

Total 978 1335 1252 985 83 138 49 4820 100%

agehire purchase

personal loan

credit cards

mail order purchase

DSS Social Fund loan

loans from individuals

something else overdraft

student loan Total Total (%)

16-20 13 44 51 51 10 11 56 94 6 336 5.8%20-25 65 175 170 87 15 28 152 173 11 876 15.1%25-30 142 233 238 114 20 25 113 77 9 971 16.7%30-35 136 239 257 134 15 17 106 28 3 935 16.1%35-40 136 212 278 125 14 19 76 12 6 878 15.1%40-45 91 157 170 86 8 4 63 3 6 588 10.1%45-50 72 110 124 55 3 4 33 3 8 412 7.1%50-55 57 85 121 58 2 3 29 2 8 365 6.3%55-60 28 61 69 44 3 17 1 3 226 3.9%60+ 46 39 76 53 11 6 231 4.0%

Total 786 1355 1554 807 90 111 656 393 66 5818 100%

1995

2000

26

Table C: Individual debt and indebtedness in BHPS (1995-2000)

1995 2000 (a) 2000 (b)

Individuals with no debt 5,353 5,182 5,182Individuals with debt 3,431 3,273 3,458Proportation of debtors 0.39 0.39 0.40

Debtmedian 700 1500 1500

sample size 3,310 3,133 3,316mean 1,718 3,272 3,281

10th percentile 60 100 10030th percentile 250 500 50050th percentile 700 1,500 1,50070th percentile 1,600 3,500 3,50090th percentile 4,000 8,000 8,000

Indebtedness (% income)median 0.08 0.12 0.12

sample size 3,287 3,090 3,257mean 0.28 1.44 2.93

10th percentile 0.87 1.30 1.2730th percentile 3.14 4.99 4.8550th percentile 7.95 12.03 11.7270th percentile 17.20 26.67 25.5690th percentile 45.42 76.92 66.74

(a) excluding from the sample individuals with only overdrafts or student loans in 2000(b) including as debtors individuals with only overdrafts or student loans in 2000

Distribution of individuals in the panel by debt levels in 1995 and 2000 (number of individuals) (*)1995

2000 no debtdebt< percent25

p25th<debt<

p50th<debt< p75th

p75th<debt Total 00

no debt 3,119 310 268 213 247 4,157debt< percentile25th 255 140 94 59 39 587between 25th and 50th 219 86 110 90 76 581between 50th and 75th 207 58 85 117 121 588<percentile 75th 197 49 84 117 219 666

Total 95 3,997 643 641 596 702 6,579

(*) Debt percentiles correspond to £180, £650 and £2,000 in 1995, and £400, £1,500 and £4,000 in 2000

27

Table D: Probit model for credit market participation

No. observations 14369 14205

chi2(117) 2658 2649

Prob>chi2 0.0000 0.0000

Pseudo R 20.1687 0.1765

log likelihood -8080.0 -7938.3

Correctly classified 70.3% 70.6%

variable reference group coeff std.err z-ratio mg effect (a) coeff z-ratio

Not head of household (Head of the hhold) 95 nhohint95 -0.03 0.049 -0.5 -0.01 -0.03 -0.500 nhohint00 -0.15 0.051 -2.9 -0.06 -0.16 -3.0

Females (Male) 95 notmaleint95 0.15 0.042 3.7 0.06 0.15 3.700 notmaleint00 0.14 0.043 3.2 0.06 0.16 3.5

Divorce/separated 95 divint95 0.02 0.069 0.2 0.01 0.02 0.300 divint00 0.03 0.071 0.5 0.01 0.03 0.4

Widow 95 widowint95 -0.28 0.091 -3.1 -0.11 -0.28 -3.100 widowint00 -0.37 0.100 -3.7 -0.15 -0.39 -3.9

Never married 95 nvrmarint95 -0.12 0.053 -2.2 -0.05 -0.12 -2.200 nvrmarint00 0.04 0.057 0.7 0.01 0.02 0.4

1 dep. Child No dep. children 95 kids1int95 0.00 0.048 0.0 0.00 0.00 0.000 kids1int00 0.07 0.049 1.4 0.03 0.08 1.6

2 dep. Children 95 kids2int95 0.03 0.053 0.6 0.01 0.03 0.500 kids2int00 0.04 0.054 0.7 0.01 0.07 1.3

3 or more dep. Children 95 kids3int95 0.13 0.074 1.8 0.05 0.13 1.800 kids3int00 0.20 0.079 2.5 0.08 0.22 2.7

Not white 95 nowhitedi~95 -0.39 0.096 -4.1 -0.15 -0.39 -4.100 nowhitedi~00 -0.37 0.095 -3.9 -0.14 -0.37 -3.8

Aged 16 to 20 ( Aged 20 to 30) 95 aged1int95 -0.53 0.079 -6.8 -0.21 -0.53 -6.800 aged1int00 -0.53 0.088 -6.1 -0.21 -0.51 -5.6

Aged 30 to 45 95 aged3int95 -0.18 0.049 -3.6 -0.07 -0.18 -3.600 aged3int00 -0.19 0.052 -3.6 -0.07 -0.16 -3.1

Aged 45 to 60 95 aged4int95 -0.33 0.058 -5.8 -0.13 -0.33 -5.800 aged4int00 -0.33 0.061 -5.5 -0.13 -0.31 -5.0

Aged 60 or more 95 aged5int95 -0.87 0.091 -9.6 -0.32 -0.87 -9.600 aged5int00 -0.67 0.098 -6.9 -0.25 -0.65 -6.5

Medium qualification (high level) 95 edd2int95 -0.08 0.040 -2.0 -0.03 -0.08 -2.000 edd2int00 -0.15 0.039 -4.0 -0.06 -0.12 -3.1

No qualifications 95 edd3int95 -0.30 0.052 -5.8 -0.12 -0.31 -5.800 edd3int00 -0.29 0.056 -5.2 -0.11 -0.24 -4.3

Self-employed (employed) 95 jbstatd1i~95 -0.14 0.063 -2.2 -0.06 -0.14 -2.300 jbstatd1i~00 0.05 0.067 0.7 0.02 0.04 0.6

Unemployed 95 jbstatd3i~95 -0.20 0.081 -2.5 -0.08 -0.20 -2.500 jbstatd3i~00 -0.24 0.100 -2.4 -0.10 -0.27 -2.6

Retired 95 jbstatd4i~95 -0.29 0.085 -3.4 -0.11 -0.29 -3.400 jbstatd4i~00 -0.36 0.087 -4.2 -0.14 -0.35 -4.0

Full time student 95 jbstatd5i~95 -0.18 0.084 -2.2 -0.07 -0.18 -2.200 jbstatd5i~00 0.09 0.097 0.9 0.03 -0.23 -2.2

Other labour status 95 jbstatd6i~95 -0.17 0.056 -3.0 -0.07 -0.17 -3.000 jbstatd6i~00 -0.20 0.059 -3.3 -0.08 -0.19 -3.2

Income (y) <= perc 10th (p10) 95 yd1int95 -0.39 0.076 -5.1 -0.15 -0.38 -5.100 yd1int00 -0.32 0.080 -4.0 -0.13 -0.34 -4.0

between 10th and 30th 95 yd2int95 -0.20 0.055 -3.7 -0.08 -0.20 -3.700 yd2int00 -0.14 0.060 -2.3 -0.06 -0.16 -2.6

between 30th and 50th 95 yd3int95 -0.04 0.053 -0.8 -0.02 -0.04 -0.800 yd3int00 -0.14 0.055 -2.5 -0.05 -0.15 -2.7

between 70th and 90th 95 yd5int95 0.11 0.051 2.2 0.04 0.11 2.200 yd5int00 0.15 0.051 3.0 0.06 0.15 2.9

Excluding those with only overdrafts or

student loans

(per 50th < income

< perc 70 th )

(Couples no dep.children)

Total sample

28

Table D continued

variable reference group coeff std.err z-ratio mg effect (a) coeff z-ratio

Not owner-occupier 95 tenured3i~95 0.38 0.092 4.1 0.14 0.38 4.100 tenured3i~00 0.20 0.086 2.3 0.08 0.17 2.0

Living with mortgage-debtors 95 tenure2d2~95 0.14 0.107 1.3 0.05 0.14 1.300 tenure2d2~00 0.01 0.102 0.1 0.00 -0.01 -0.1

Living with owner occupiers 95 tenure2d12~5 -0.05 0.122 -0.4 -0.02 -0.05 -0.400 tenure2d12~0 -0.07 0.112 -0.6 -0.03 -0.08 -0.7

Secured-debt/income < 20th perc 95 risdi2d1i~95 0.32 0.104 3.1 0.12 0.32 3.100 risdi2d1i~00 0.09 0.099 0.9 0.04 0.08 0.8

between 20th and 40th 95 risdi2d2i~95 0.36 0.104 3.4 0.14 0.36 3.400 risdi2d2i~00 0.21 0.099 2.2 0.08 0.20 2.0

between 40th and 60th 95 risdi2d3i~95 0.50 0.105 4.7 0.19 0.49 4.700 risdi2d3i~00 0.31 0.100 3.2 0.12 0.31 3.1

between 60th and 80th 95 risdi2d4i~95 0.44 0.105 4.2 0.17 0.43 4.100 risdi2d4i~00 0.30 0.101 3.0 0.12 0.30 2.9

larger than the 80th percentile 95 risdi2d5i~95 0.39 0.106 3.7 0.15 0.39 3.700 risdi2d5i~00 0.27 0.102 2.7 0.11 0.27 2.6

No mortgage, p30 < housing equity< p50 95 hwid3int95 0.02 0.132 0.2 0.01 0.02 0.100 hwid3int00 -0.20 0.131 -1.6 -0.08 -0.20 -1.5

between 50th and 70th 95 hwid4int95 0.24 0.123 2.0 0.09 0.24 2.000 hwid4int00 -0.10 0.124 -0.8 -0.04 -0.11 -0.9

between 70th and 90th 95 hwid5int95 0.12 0.141 0.9 0.05 0.12 0.900 hwid5int00 -0.19 0.131 -1.4 -0.07 -0.24 -1.8

larger than the 90th percentile 95 hwid6int95 0.13 0.177 0.7 0.05 0.13 0.700 hwid6int00 -0.18 0.171 -1.0 -0.07 -0.16 -1.0

positive fcial wealth < median (zero fin.wealth) 95 fwd2int95 0.06 0.040 1.6 0.02 0.06 1.600 fwd2int00 0.17 0.043 4.0 0.07 0.20 4.5

positive fcial wealth > median 95 fwd3int95 -0.32 0.045 -7.0 -0.13 -0.32 -7.000 fwd3int00 -0.22 0.048 -4.6 -0.09 -0.22 -4.4

Positive expectations on future financial situation 95 poexpint95 0.17 0.036 4.6 0.07 0.16 4.500 poexpint00 0.27 0.037 7.1 0.10 0.27 7.2

constant (1995) 95 v95 -0.10 0.207 -0.5 -0.04 -0.04 -0.2constant 00 _cons 0.07 0.147 0.5 0.00 0.01 0.1

null hypothesis p-values null hypothesis p-values

aged5int95=aged5int00 0.134 risdi2d1int95=risdi2d1int00.115

yd1int95=yd1int00 0.544 risdi2d2int95=risdi2d2int00.316

yd2int95=yd2int00 0.438 risdi2d3int95=risdi2d3int00.209

yd3int95=yd3int00 0.213 risdi2d4int95=risdi2d4int00.358

yd5int95=yd5int00 0.597 risdi2d5int95=risdi2d5int00.417

yd6int95=yd6int00 0.833 poexpint95=poexpint00 0.056

jbstatd1int95=jbstatd1int00 0.040 fwd2int95=fwd2int00 0.069

jbstatd5int95=jbstatd5int00 0.036 fwd3int95=fwd3int00 0.140

(a) dF/dx is for a discrete change of dummy variables from 0 to 1.

(owners-occupiers with no mortgage

and hw < perc.30th )

Note: Estimation is carried out together for 1995 and 2000 interacting each variable with the year. Standard errors are heteroskedastic robust. Theestimation also includes 17 region dummies. Percentiles values are calculated for total sample. Income percentiles are, respectively, £1,328, £4,246,£7,376, £12,027, £21,316 in 1995. In 2000 they are updated according to RPI. Secured-debt to income ratios percentiles are 0.73, 1.21, 1.71, 2.5 in 1995,and 0.81, 1.25, 1.77, and 2.52 in 2000. Housing wealth percentiles for owner occupiers with no mortgage are £20,000, £30,000, £42,000, £60,000,£90,000, in 1995, and £25,000, £42,500, £60,000, £90,000, £150,000 in 2000. Financial wealth includes liquid and non-liquid financial assets and medianvalues correspond to £1,560 in 1995 and £2,250 in 2000. The category 'other' in labour status include maternity leave, family care, school, LT sick,disabled, Government training scheme and Other.

29

Table E: Cross-section regressions of debt levels

Model 1 Model 2number obs 5576 5576R-squared 0.2589 0.2778Adj R-squared 0.2464 0.2652

coeff std error t-stat coeff std error t-stat

Not head of the household nhohint95 -0.05 0.08 -0.58 -0.05 0.08 -0.59nhohint00 -0.11 0.08 -1.26 -0.07 0.08 -0.87

Female notmaleint95 -0.36 0.07 -5.08 -0.36 0.07 -5.15notmaleint00 -0.33 0.07 -4.45 -0.32 0.07 -4.36

Divorce/separated divint95 0.05 0.11 0.41 0.05 0.11 0.42divint00 0.03 0.12 0.21 0.05 0.12 0.43

Widow widowint95 -0.18 0.21 -0.83 -0.18 0.21 -0.84widowint00 -0.34 0.24 -1.39 -0.39 0.24 -1.63

Never married nvrmarint95 0.14 0.09 1.55 0.14 0.09 1.57nvrmarint00 0.21 0.10 2.15 0.14 0.09 1.45

1 dep. Child kids1int95 -0.17 0.08 -2.25 -0.17 0.08 -2.27kids1int00 -0.26 0.08 -3.36 -0.19 0.08 -2.45

2 dep. Children kids2int95 -0.17 0.08 -2.03 -0.17 0.08 -2.06kids2int00 -0.21 0.09 -2.43 -0.16 0.09 -1.81

3 or more dep. Children kids3int95 -0.33 0.11 -2.89 -0.33 0.11 -2.93kids3int00 -0.09 0.12 -0.77 -0.01 0.12 -0.05

Not white nowhitedi~95 0.32 0.18 1.78 0.32 0.18 1.80nowhitedi~00 0.22 0.19 1.15 0.14 0.19 0.75

Aged 16 to 20 aged1int95 -0.75 0.14 -5.43 -0.75 0.14 -5.50aged1int00 -0.72 0.15 -4.65 -0.84 0.17 -5.09

Aged 30 to 45 aged3int95 0.01 0.07 0.14 0.01 0.07 0.15aged3int00 -0.16 0.08 -2.07 -0.06 0.08 -0.70

Aged 45 to 60 aged4int95 -0.14 0.09 -1.56 -0.14 0.09 -1.58aged4int00 -0.29 0.10 -2.89 -0.15 0.10 -1.45

Aged 60 or more aged5int95 -0.27 0.20 -1.35 -0.27 0.20 -1.37aged5int00 -0.53 0.21 -2.54 -0.40 0.21 -1.92

Medium qualification edd2int95 -0.14 0.06 -2.27 -0.14 0.06 -2.30edd2int00 -0.22 0.06 -3.45 -0.12 0.06 -1.88

No qualifications edd3int95 -0.28 0.09 -3.05 -0.28 0.09 -3.09edd3int00 -0.38 0.11 -3.65 -0.29 0.10 -2.81

Self employed jbstatd1i~95 0.43 0.11 3.99 0.43 0.11 4.04jbstatd1i~00 0.25 0.11 2.29 0.27 0.11 2.43

Unemployed jbstatd3i~95 -0.38 0.14 -2.68 -0.38 0.14 -2.71jbstatd3i~00 -0.12 0.19 -0.63 -0.01 0.19 -0.07

Retired jbstatd4i~95 -0.43 0.20 -2.16 -0.43 0.20 -2.18jbstatd4i~00 -0.14 0.21 -0.67 -0.11 0.21 -0.52

Full time student jbstatd5i~95 0.60 0.15 3.95 0.60 0.15 4.00jbstatd5i~00 0.54 0.20 2.70 -0.27 0.21 -1.24

Other labour status jbstatd6i~95 -0.27 0.10 -2.69 -0.27 0.10 -2.73jbstatd6i~00 -0.41 0.11 -3.76 -0.35 0.11 -3.28

Income (y) <= perc 10th (p10) yd1int95 -0.79 0.14 -5.72 -0.79 0.14 -5.79yd1int00 -0.45 0.15 -2.99 -0.61 0.15 -4.04

between 10th perc and 30th perc yd2int95 -0.46 0.10 -4.79 -0.46 0.10 -4.85yd2int00 -0.46 0.11 -4.13 -0.55 0.11 -4.96

between 30th perc and 50th perc yd3int95 -0.30 0.09 -3.43 -0.30 0.09 -3.48yd3int00 -0.25 0.10 -2.56 -0.26 0.10 -2.68

between 70th perc and 90th perc yd5int95 0.41 0.08 5.03 0.41 0.08 5.10yd5int00 0.45 0.08 5.48 0.42 0.08 5.18

larger than 90th perc yd6int95 0.64 0.11 5.98 0.64 0.11 6.06yd6int00 0.90 0.11 8.59 0.91 0.10 8.75

Income of other members (coef *104) ahmiint95 0.08 0.00 3.18 0.08 0.00 3.22ahmiint00 0.05 0.00 2.50 0.05 0.00 2.13

30

… table E continuedModel 1 Model 2

coeff std error t-statistic coeff std error t-statistic

Not owner-occupier tenured3i~95 0.21 0.20 1.04 0.21 0.20 1.05tenured3i~00 -0.01 0.18 -0.03 -0.06 0.18 -0.32

Living with mortgage-debtors tenure2d2~95 0.21 0.22 0.95 0.21 0.22 0.96tenure2d2~00 0.15 0.20 0.74 0.15 0.20 0.75

Living with owner occupiers tenure2d12~5 0.36 0.26 1.42 0.36 0.25 1.44tenure2d12~0 0.42 0.23 1.77 0.36 0.23 1.57

Secured-debt/income < 20th perc risdi2d1i~95 0.14 0.21 0.68 0.14 0.21 0.69risdi2d1i~00 -0.13 0.20 -0.65 -0.12 0.19 -0.60

between 20th perc and 40th perc risdi2d2i~95 0.34 0.21 1.62 0.34 0.21 1.64risdi2d2i~00 0.16 0.19 0.83 0.17 0.19 0.89

between 40th perc and 60th perc risdi2d3i~95 0.30 0.21 1.40 0.30 0.21 1.42risdi2d3i~00 0.11 0.19 0.57 0.13 0.19 0.67

between 60th perc and 80th perc risdi2d4i~95 0.29 0.21 1.38 0.29 0.21 1.40risdi2d4i~00 0.25 0.19 1.27 0.26 0.19 1.34

larger than 80th perc risdi2d5i~95 0.57 0.21 2.66 0.57 0.21 2.69risdi2d5i~00 0.40 0.20 2.00 0.41 0.20 2.09

No mortgage, p30 < housing equity< p50 hwid3int95 0.22 0.30 0.74 0.22 0.30 0.75hwid3int00 0.29 0.30 0.98 0.30 0.30 1.03

between 50th perc and 70th perc hwid4int95 0.43 0.27 1.57 0.43 0.27 1.59hwid4int00 0.39 0.28 1.38 0.40 0.28 1.47

between 70th perc and 90th perc hwid5int95 0.08 0.32 0.25 0.08 0.31 0.25hwid5int00 -0.04 0.31 -0.12 -0.02 0.30 -0.06

larger than 90th perc hwid6int95 0.14 0.37 0.38 0.14 0.37 0.38hwid6int00 0.94 0.37 2.55 0.97 0.36 2.68

Positive fcial wealth < perc 50th fwd2int95 -0.12 0.07 -1.88 -0.12 0.07 -1.90fwd2int00 -0.14 0.07 -1.95 -0.11 0.07 -1.60

Positive fcial wealth > perc 50th fwd3int95 -0.11 0.08 -1.35 -0.11 0.08 -1.36fwd3int00 -0.26 0.09 -2.97 -0.21 0.09 -2.51

Positive expectations on future fcial situat poexpint95 0.24 0.06 4.21 0.24 0.06 4.26poexpint00 0.23 0.06 3.75 0.22 0.06 3.71

constant 1995 v95 -0.85 0.35 -2.42 -0.58 0.35 -1.68Constant _cons 7.24 0.24 29.63 6.98 0.24 28.72Overdraft (aged 16-20) 2.65 0.33 7.99Overdraft (aged 30-60) 1.23 0.14 9.10Student loans 0.88 0.33 2.66

For each variable the first row corresponds to the interaction with year 1995, and the second for 2000. The estimation include 7 region dummies. See also notes in Table D.

31

table E continued

Null hypothesis

Model 1 Model 2

nhohint95=nhohint00 0.607 0.821notmaleint95=notmaleint00 0.719 0.639divint95=divint00 0.895 0.982widowint95=widowint00 0.622 0.507nvrmarint95=nvrmarint00 0.595 0.996kids1int95=kids1int00 0.405 0.869kids2int95=kids2int00 0.727 0.906kids3int95=kids3int00 0.152 0.048nowhitedint95=nowhitedint00 0.694 0.486aged1int95=aged1int00 0.883 0.667aged3int95=aged3int00 0.109 0.539aged4int95=aged4int00 0.287 0.993aged5int95=aged5int00 0.355 0.645edd2int95=edd2int00 0.387 0.804edd3int95=edd3int00 0.451 0.922jbstatd1int95=jbstatd1int00 0.266 0.296jbstatd3int95=jbstatd3int00 0.274 0.118jbstatd4int95=jbstatd4int00 0.316 0.258jbstatd5int95=jbstatd5int00 0.816 0.001jbstatd6int95=jbstatd6int00 0.340 0.563yd1int95=yd1int00 0.105 0.379yd2int95=yd2int00 0.979 0.567yd3int95=yd3int00 0.703 0.749yd5int95=yd5int00 0.699 0.896yd6int95=yd6int00 0.083 0.072tenured3int95=tenured3int00 0.429 0.320tenure2d22int95=tenure2d22int00 0.839 0.836tenure2d12int95=tenure2d12int00 0.882 0.999risdi2d1int95=risdi2d1int00 0.347 0.363risdi2d2int95=risdi2d2int00 0.526 0.544risdi2d3int95=risdi2d3int00 0.516 0.550risdi2d4int95=risdi2d4int00 0.876 0.902risdi2d5int95=risdi2d5int00 0.552 0.576hwid3int95=hwid3int00 0.867 0.846hwid4int95=hwid4int00 0.909 0.946hwid5int95=hwid5int00 0.789 0.822hwid6int95=hwid6int00 0.127 0.107fwd2int95=fwd2int00 0.864 0.922fwd3int95=fwd3int00 0.214 0.367poexpint95=poexpint00 0.835 0.791

p-value

32

Table F: Panel estiamtion for the change in debt levels

Dependent variable Dt-Dt-1

year 2000 2000 (a) 2000 (b)

Number of obs. 4875 4875 3462.9F test 8.22 10.7 5.79Prob F test 0.0000 0.0000 0.0000R2 0.0646 0.0928 0.0452Root MSE 3585 3531 3463

explanatory varaibles reference group coeff t-stat coeff t-stat coeff t-stat

Increase in income (high income individuals in 1995) 0.02 1.54 0.02 1.61 0.02 1.57Increase in income (low income individuals in 1995) 0.05 2.92 0.03 2.26 0.03 2.17

Expectations of better financial situation 506.81 3.06 434.98 2.65 399.81 2.39

Increase in educational qualification 527.93 3.52 327.79 2.28 212.25 1.47

20 years old in 2000 (30 to 45 years old) 249.19 0.29 308.50 0.58 -307.03 -0.9830 to 45 years old -40.71 -0.20 142.37 0.71 63.67 0.3345 to 60 years old -522.04 -2.46 -287.58 -1.36 -419.66 -2.10Older tan 60 -724.68 -2.70 -474.52 -1.78 -590.55 -2.25

Becoming unemployed in 2000(employed in 95 and 00) -195.97 -0.56 -187.50 -0.57 -277.73 -0.86

Uemployed in 1995 and 2000 -518.43 -1.70 -422.22 -1.43 -455.45 -1.53Becoming employed/self employed in 2000 467.67 2.05 137.69 0.61 33.38 0.15From self-empl to employed -298.85 -1.00 -300.10 -1.01 -286.91 -0.96From employed to sel-employ 221.29 0.34 226.97 0.35 216.97 0.33Self-employed in 1995 and 2000 654.83 1.10 661.88 1.11 639.30 1.07Retired in 2000 125.24 0.49 89.36 0.35 56.45 0.22Full time student in 2000 531.56 1.15 -678.65 -2.06 -980.02 -5.38Other -140.65 -0.96 -180.63 -1.24 -288.03 -2.11

Change in mortgage stock (if positive) 0.01 1.03 0.00 0.80 0.00 0.57Change in mortgage stock (if negative) -0.03 -1.95 -0.03 -1.95 -0.03 -1.97No chang in mortgage stock (positive mortgage) -444.56 -1.36 -436.38 -1.33 -410.42 -1.25No change in mortgage stock (zero mortgage) -146.29 -0.96 -134.81 -0.89 -98.93 -0.66

Changes in financial wealth -0.01 -1.27 -0.01 -1.22 -0.01 -1.20

Changes in net wealth (dummies) 57.29 1.02 64.62 1.16 73.25 1.31

Constant 598.68 2.00 327.57 1.11 563.11 1.98Individuals with overdrafts 4437 9.82

(a) including a dummy for individuals with overdrafts and student loans.(b) excluding individuals with overdrafts and student loans.

exc. Individuals with overdraftsdummy for overdrafts

33

Chart 1: Debt and indebtedness by age and income

Fraction of borrowers and mean unsecured debt of debtors (constant prices 1995)

0

500

1000

1500

2000

2500

3000

3500

4000

16-2

020

-25

25-3

030

-35

35-4

040

-45

45-5

050

-55

55-6

060

+

age

£

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70% mean debt, 95

mean debt, 00fraction of debtors 95fraction of debtors 00

Unsecured-debt-to-income ratio (Median)

0.00

0.05

0.10

0.15

0.20

0.25

16-2

020

-25

25-3

030

-35

35-4

040

-45

45-5

050

-55

55-6

060

+

age

%income indebtedness, 95

indebtedness, 00

Fraction of borrowers and mean unsecured debt of debtors (constant prices 1995)

0

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

1 2 3 4 5 6 7 8 9 10

income deciles

£

0.00

0.10

0.20

0.30

0.40

0.50

0.60%mean debt, 95

mean debt, 00fraction of debtors, 95fraction of debtors, 00

Unsecured-debt-to-income ratio (Median)

0.00

0.05

0.10

0.15

0.20

0.25

0.30

1 2 3 4 5 6 7 8 9 10

income deciles

%incomeindebtedness, 95

indebtedness, 00

34

Chart 2: Debt and indebtedness by net housing wealth of the household and housing-tenure status

Fraction of borrowers and mean unsecured debt of debtors (constant prices 1995)

0

500

1000

1500

2000

2500

3000

3500

4000

nohousingwealth

<10 10-30 30-50 50-70 70-90 morethan 90

£

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

Net housing wealth (percentiles)

%mean debt, 95 mean debt, 00fraction of debtors, 95 fraction of debtors,00

Unsecured-debt-to-income ratio (Median)

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

no housingwealth

<10 10-30 30-50 50-70 70-90 more than90

Net housing wealth

%income

1995 2000

Fraction of borrowers and mean unsecured debt of debtors (constant prices 1995)

0

500

1000

1500

2000

2500

3000

3500

Owneroccupier, no

mortage

Owneroccupier with

mortgage

Other Living withowner-

occupiers

Living withmortgagers

£

-0.05

0.05

0.15

0.25

0.35

0.45

0.55%

Unsecured-debt-to-income ratio (Median)

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

Owneroccupier, no

mortage

Owneroccupier with

mortgage

Other Living withowner-

occupiers

Living withmortgagers

%income

1995

2000

Fraction of borrowers and mean unsecured debt of debtors (constant prices 1995)

0

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

< percentile20

20-40 40-60 60-80 more than80Secured-debt-to-income ratio

£

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70%

Unsecured-debt-to-income ratio (Median)

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

< percentile20

20-40 40-60 60-80 more than80Secured-debt-to-income ratio

%income

1995

2000

35

Chart 3: Debt and indebtedness by financial wealth

Note: in 2000 the sample excludes individuals reporting only overdrafts or student loans. New samples in BHPS from 1997 are not included. Income is the annual individual income (sum of labour and non-labour income).

Fraction of borrowers and mean unsecured debt of debtors (constant prices 1995)

0

500

1000

1500

2000

2500

3000

3500

No financialwealth

1 2 3 4

Financial wealth quartiles

£

0.00

0.10

0.20

0.30

0.40

0.50

0.60%

mean debt, 95 mean debt, 00fraction of debtors,95 fraction of debtors, 00

Unsecured-debt-to-income ratio (Median)

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

No financialwealth

1 2 3 4

Financial wealth quartiles

%income

1995

2000

Source: Bank of England

Chart 4: Spread between unsecured debt interest rates and the retail bank base rate: personal loans, overdrafts and credit cards (quoted interest rates)

5

7

9

11

13

15

17

1995

1996

1997

1998

1999

2000

5

7

9

11

13

15

17

OVERDRAFT RATECREDIT CARDPERSONAL LOAN '(£2500, £3500 from Jan '99)PERSONAL LOAN '(£10,000 tier)

36

Chart 5: sample weights by age, income, eduacation, labour status, housing tenure and financial wealth

0%

2%

4%

6%

8%

10%

12%

14%

16-2

020

-25

25-3

030

-35

35-4

040

-45

45-5

050

-55

55-6

060

+

age

%sample

% sample, 1995

% sample, 2000

0%

2%

4%

6%

8%

10%

12%

14%

1 2 3 4 5 6 7 8 9 10

income deciles

% sample% sample,1995

% sample, 2000

0%

10%

20%

30%

40%

50%

High qualification Mediumqualification

No qualifications

%sample % sample, 1995

% sample, 2000

0%

15%

30%

45%

Employed Unemployed F.T.student

Retired

%sample % sample, 1995

% sample, 2000

Self-employed

Other

0%

15%

30%

45%

Not owner-occupier

Living withmortage-debtors

Living withowner

occupiers

Owners-occupiers

Owners-occupiers

withmortgage

%sample% sample, 1995

% sample, 2000

0%

10%

20%

30%

40%

No financial wealth Fcial wealth belowmedian

Fcial wealth abovemedian

%sample% sample, 1995

% sample, 2000

37

Chart 6: predicted unsecured-debt by age and income

5.5

6.0

6.5

7.0

7.5

16 to 20 20 to 30 30 to 45 45 to 60 60+

age

2000 (Model 1) 1995

2000 (Model 2)

5.5

6.0

6.5

7.0

7.5

8.0

y < perc 10th 10th<y<30thperc

30th <y< 50thperc

70th <y< 90thperc

y< 90th perc

income

1995 2000 (Model 1)

2000 (Model 2)