A Regional Analysis of Mortgage Possessions

Published on July 2016 | Categories: Documents | Downloads: 21 | Comments: 0 | Views: 93
of 57
Download PDF   Embed   Report

Comments

Content

A REGIONAL ANALYSIS OF MORTGAGE POSSESSIONS: CAUSES, TRENDS AND FUTURE PROSPECTS1

By John Muellbauer, Professor of Economics and Official Fellow and Gavin Cameron, Research Officer, at Nuffield College, Oxford

Executive Summary • This paper analyses the variation in court possession actions and orders over regions and of total UK possessions rates, in order to understand better the causes of the high rates of mortgage possessions in the 1990s. It untangles the web of economic forces and

administrative responses, and forms a view about future prospects. • Data on court orders given actions and on suspended court orders provide evidence of a shift in behaviour by the Courts. Data on court actions brought by lenders provide evidence of a shift in behaviour by mortgage lenders. UK data on possessions rates reflect both shifts. • Evidence suggests that there was a temporary softening of policy in the County Courts beginning in 1991 but that Court policy had returned to normal by 1995. • Shifts in policy by lenders are probably a mix of intended changes in behaviour and initial delays in setting up systems to deal with possessions. The evidence suggests that this policy softening effect is gradually decaying. • The outlook for the rate of possessions depends partly on how rapidly this effect decays, on the hard-to-quantify effects of the tightening of DSS rules from October 1995 for claimants, as well as on the macroeconomic fundamentals.

1



The analysis suggests that the falls in the rate of possessions seen at the end of 1996 will be followed by significant further falls in 1997-98 provided a significant rise in interest rates can be avoided.

1.

INTRODUCTION

The years 1990 to 1995 saw a record number of households, around 345,000, containing perhaps one million individuals2 suffering the misfortune of mortgage possession. An important question concerns the institutional response to this crisis: was there a shift in policy by the courts and by the mortgage lenders? The purpose of this paper is to analyse the variation over regions, in order to understand better the causes of the high rates of mortgage possessions in the 1990s, untangling the web of economic forces and administrative responses, and to form a view about future prospects. This paper argues that there was a temporary softening of policy in the County Courts beginning in 1991 but that Court policy had returned to normal by 1995. There is evidence, however, that mortgage lenders are still a little “softer'' than they were before 1991.

We examine four sources of empirical evidence. First, aggregate UK data on the relationship between rates of mortgage arrears and possessions indicate a fall in possessions beginning in 1991 while arrears continued to rise for two more years. This suggests a shift in policy or procedures by the courts and/or mortgage lenders. Second, regional data from County Courts for England and Wales show a rise in the ratio of suspended court orders beginning in 1991 and suggesting a softening of policy. Thirdly, the ratio of court orders to court actions was smaller in 1991-94 than would have been expected given the economic fundamentals. These last two sets of evidence point to a shift in behaviour by courts. Fourthly, the rate of court actions was lower in
2

1992-96 than would have been expected given the economic fundamentals, pointing to a shift in behaviour by mortgage lenders.

By pooling time series/cross section data across regions, the methodology is to identify national policy shifts by courts and by lenders in terms of common time effects across regions, distinguishable from economic fundamentals varying by region and by time. The importance of these time effects are confirmed in aggregate time series econometric models for the UK rate of mortgage possession. The paper concludes by considering prospects for the rate of court action for mortgage possession. The paper also throws interesting light on regional variations in rates of court actions and orders. For example, it destroys the myth that the possessions "crisis" was largely a Southern phenomenon. Fuller details are set out in the technical appendices to Sections 4, 5 and 7 and in a data appendix.

2. (a)

COUNTY COURT DATA AND THE HISTORICAL BACKGROUND The Relevance of County Court Data

In most cases, mortgage possession involves court proceedings. Ford et al (1995) report that even in cases where households voluntarily handed the keys of their property to their mortgage lender, evidence of court proceedings was often a requirement to be eligible for rehousing in the social rented sector. Thus, data on court proceedings are of more general interest for understanding the phenomenon of mortgage possession. Data on County Court actions and orders for mortgage possession have been published for 1986 to the present and on a quarterly basis since 1987. They are the only source of regularly available regional data to map possessions.

3

To be more specific about the nature of the data, the Lord Chancellor's Department defines terms as follows.

Actions Entered: A plaintiff begins an action for an order for possession of residential property by way of a summons in a county court.

Orders Made: The court, following a judicial hearing, may grant an order for possession immediately. This entitles the plaintiff to apply for a warrant to have the defendant evicted. However, even where a warrant for possession is issued, the parties can still negotiate a compromise to prevent eviction.

Suspended Orders: Frequently, the court grants the mortgage lender possession but suspends the operation of the order. Provided the defendant complies with the terms of the suspension, which usually requires the defendant to pay the current mortgage instalments plus some of the accrued arrears, the possession order cannot be enforced.

(b)

Historical Background

Between 1980 and 1990 average mortgage debt in the UK more than doubled relative to income. The house price boom of the mid to late 1980s3 came to an end amid sharp rises in interest rates in 1988-1990, the response of the policy makers to the deterioration in the UK's balance of payments and inflation. House prices fell in nominal terms for the first time since the late 1950s, particularly in the Southern regions of the UK. This left many households, particularly those who had bought near the top of the boom, with negative equity. The combination of negative equity, and the drain on cash flows from the high interest rates of the 1988-1992 period and from income loss or business failure, was a particularly critical mix of circumstances. When households have
4

positive equity, the opportunity to trade down is available as a response to cash flow problems. When households have negative equity, this is unavailable unless mortgage lenders and their regulators have a particularly tolerant attitude and allow households to trade down while retaining negative equity.4 Initially, negative equity products were not offered by most mortgage lenders, though these did become available increasingly over time (see the interview evidence cited by Ford et al (1995), ch 4). It appears that practices by the County Courts also altered in the 1990s, with longer repayment periods for households in payment arrears being permitted, see Ford (1994) and Ford et al (1995), ch 5.

To appreciate the context in which the policy of mortgage lenders and of courts is likely to have shifted, recall that in 1991 mortgage possessions were frequently headline news. Heightened public concern was reflected in the implicit contract agreed between the Government and mortgage lenders in November-December 1991, 6 months before the 1992 General Election, to reduce possessions. On the Government’s side this included the commitment to pay DSS mortgage support payments direct to lenders rather than to mortgagors and to stimulate the housing market by raising the Stamp Duty ceiling for a year and by giving ear marked grants to housing associations to buy up properties originally intended for owner occupation, see Stephens (1996) for fuller details. By the time of the November 1994 Budget, however, public concern about possessions had faded sufficiently that the Government felt able to announce major reductions in the DSS safety net for new mortgagors and some reductions for existing mortgagors to take effect from October 1995, see Dale (1995 ) and Stephens (1996).

The Courts are unlikely to have been immune from these shifts in public and official concern. It can also be argued that the Courts may have placed more of the responsibility for mortgage default in the early 1990's on lending practices in the 1980s of some mortgage lenders and on
5

macro policy mismanagement by the Government. This would also have led to a softening of court policies. The passage of time and a declining proportion of cases coming to court of mortgages originating in the 1980's would in due course cause a return to normal practices.

It is also possible that what we interpret as a purposive temporary softening of policy may have been partly the result of the inability of existing court facilities to cope with the flood of possessions cases. This would imply an apparent softening of policy as the number of cases peaked, followed by an apparent hardening as the number of cases declined again. Mortgage lenders would have been subject to the same set of influences as defaults increased, including lags in training staff and setting up systems for dealing with cases. Interview evidence in Ford et al (1995) suggests this was the case.

This view contrasts with another interpretation of a pattern of high possessions rates followed by lower rates. This emphasizes the efficiency and speed of reaction of mortgage lenders in possessing the most disastrous first, implying a later reduction in possessions rates. The latter would result from the change in the composition of the population at risk with the removal of many of the highest risk cases. As we will see, there are several important pieces of evidence against this interpretation.

In addition, two further arguments are often put forward. The first is that an apparent softening of policy could have been the result of the difficulty lenders had, particularly in 1991-3, in selling houses which had been taken into possession. A related argument concerns the fact that mortgage indemnity policies typically insured the top 25% of the value of a property. It has been argued that mortgage lenders therefore had incentives for taking possession rapidly when losses were small. Typically, homes in possession appear to sell at a discount below similar regular
6

homes. Thus, when nominal house prices in Southern regions in 1991-3 fell to levels 10% or more below their 1988-90 levels, the argument is that mortgage lenders would have begun to bear a significant part of the losses and so had reduced incentives to exercise the possession option. Note that this argument would imply a significantly greater softening of policy in the Southern regions where the biggest nominal falls occurred. We test for such an effect later.

There is a third, less controversial aspect to this line of argument. If, as Breedon and Joyce (1992) argue, taking homes into possession weakens house prices further, such action by one mortgage lender imposes negative externalities on the others by bringing about further defaults. Thus, there is scope for collective action. The implicit contract between mortgage lenders and the Government in November-December 1991 can be seen as providing just the required impetus for such collective action to slow possessions rates.

To understand the variations that have occurred in rates of mortgage possession, such alleged shifts in behaviour need to be taken into account along with the influence of variations in economic conditions, such as the debt/equity ratio (i.e., the ratio of mortgage debt to the value of the home) and debt service ratios (i.e., the ratio of mortgage interest payments to income), unemployment shocks, small business failure rates, and house price developments. Indeed, econometric studies of aggregate mortgage possession data such as Breedon and Joyce (1992), Brookes et al (1994), and Allen and Milne (1994) estimated on data up to 1990 or 1991 break down badly on later data. These studies use mortgage arrears to help explain mortgage possessions. But from 1991, possessions started to fall while arrears continued to rise for some time, introducing a break in their relationship, see Chart 1. Though this is not conclusive evidence of a structural shift in behaviour, it is very much consistent with this hypothesis.
7

Incidentally, this shift in behaviour appears not to be due to distortions in the month in arrear data, see Muellbauer (1996), the source for the estimated 10% in arrear rate shown in Chart 1.

CHART 1: RATES OF POSSESSION, 12 MONTH AND 10% IN ARREAR
Note: The rates are log possessions rate, the log 12 month in arrear rate and the fitted log 10% balance in arrear rate

3.

REGIONAL CONTRASTS

Regional data offer more scope than aggregate data for detecting such shifts in behaviour. It seems likely that shifts in behaviour by County Courts and mortgage lenders would be approximately uniform across the country, though short-term variations could be region specific: there is pressure on the courts to operate a national system of justice and the major mortgage lenders operate in all the regions. The housing market experience of different regions has been more diverse: nominal and real house prices have moved very differently in different regions and the timing of the build-up of mortgage debt differed across regions. Debt service ratios differ across regions with varying ratios of debt relative to income and the differential impact of the mortgage interest tax relief ceiling. Unemployment rates and small business failure rates have also varied substantially across regions. In these various

dimensions, a broad North-South grouping of regions is helpful in summarizing information: within each broad group there were more shared experiences than between groups.

To give some background to the regional data from the courts for 1986-1996, note that the Southern regions led the strong UK upswing in house prices in 1986-88. After interest rates rose from the summer of 1988 to a peak in 1990, house prices in the Southern regions slowed sharply and then declined in 1990-93 by 17 to 23%, even though interest rates were by then
8

falling. Table 1 shows the dates of peaks and troughs in nominal mix-adjusted house price indices and the peak-trough decline by region. Furthermore, Southern unemployment rose sharply and much more than in Northern regions. In the latter, house prices continued to rise quite strongly in 1988-90 and since then have fallen only marginally if at all. In 1995-6, the Southern regions were again leading the recovery in house prices, though unemployment rates had hardly begun to diverge again by the end of 1996.

TABLE 1 HOUSE PRICE PEAKS AND TROUGHS BY ENGLISH REGIONS AND WALES, 1986-1996 Period Peak Trough % Fall North 1994 1995 5.1 Yorks & Humber 1991 1994 4.6 East Mid 1990 1993 7.2 East Anglia 1989 1993 22.0 South East 1989 1993 19.4 South West 1989 1993 20.6 West Mid 1990/1 1993 4.2 North West 1992 1993 3.9 Wales 1990 1992 2.4

Source: Table B3, Compendium of Housing Finance Statistics, Council of Mortgage Lenders Notes: 1. The SE figure is an average of a 16.7% fall in London and a 22.7% fall in the Rest of the South East. Note that nominal house prices in Scotland and Northern Ireland showed continuous rises throughout the 1986-95 period.

CHART 2A: RATE OF COURT ACTIONS, DEBT/EQUITY RATIO AND DEBT SERVICE RATIO (SOUTHERN REGIONS)
Note: The Southern regions are defined as: East Anglia, East Midlands, South East including Greater

London, and South West

9

CHART 2B: RATE OF COURT ACTIONS, DEBT/EQUITY RATIO AND DEBT SERVICE RATIO (NORTHERN REGIONS)
Note: Wales The Northern regions are defined as: North, North West, Yorkshire & Humberside, West Midlands and

To compare court actions across regions it is necessary to scale by the number of outstanding mortgages in each region. Charts 2A and 2B show basic data on court action rates, debt/equity ratios and debt service ratios, respectively averaged across Southern and Northern regions. These charts show that in 1986-1988 court action rates were higher in Northern than in Southern regions but that Southern rates rose more sharply to the 1991 peak before falling back to similar rates in 1995. These charts explode the myth that the possessions "crisis" was largely a Southern phenomenon.

The charts also show that debt/equity ratios started lower in Southern regions, since historically loan-to-value ratios have been lower there, see Table 2 below. With the fall in nominal house prices after 1989, however, debt/equity ratios rose much more strongly there. Debt service ratios (defined in the note to Table 2) have throughout been higher in Southern regions. This is consistent with the systematically higher ratio of house prices to earnings found in these regions. The rising and falling pattern shows clearly the influence of the rises in interest rates in 1988-90 and the subsequent declines. Not surprisingly, our empirical results confirm the strong association between court action rates and debt service ratios visually apparent in these charts.

10

TABLE 2 AVERAGE LOAN TO VALUE RATIOS IN 1970-95 FOR FIRST-TIME BUYERS BY REGIONS AND 1986 VALUES OF DEBT/EQUITY AND DEBT SERVICE RATIOS Period Yorks & East East South South West North North Humber Mid Anglia East West Mid West Wales LVR% 84.3 83.9 83.0 80.4 79.1 79.9 82.6 84.0 83.1 Debt/Equity % 40.9 38.6 35.3 36.2 34.2 37.1 38.4 37.0 39.3 Debt/Service % 28.7 27.4 28.4 36.0 44.5 39.3 31.3 27.8 28.6
Source: LVR from DOE 5% sample. Debt/Equity = average mortgage/average 1985 second-hand house price indexed by mix-adjusted indices. Debt Service = (average mortgage) (tax adjusted mortgage interest)/personal disposable income per capita indexed to regional earnings index, see forthcoming CML Discussion Paper for further details.

It is instructive to look at court orders relative to court actions, see Charts 3A and 3B. Court orders, including suspended orders, rest on specific decisions by courts whereas court actions are largely the consequence of cases brought by mortgage lenders. There are some notable differences in movements of the ratio of orders to actions. In particular, there is little sign of a fall since 1992 in the South East, South West and East Anglia, in contrast to the other regions. This can be explained partly in terms of a worse underlying debt/equity position in the Southern regions, see Chart 2A. However, because of delays in court procedures, there is also a lag of orders behind actions.

CHART 3A: RATIO OF COURT ORDERS TO ACTIONS, IMPLEMENTED ORDERS TO COURT ORDERS AND CHANGE IN NOMINAL HOUSE PRICES (SOUTHERN REGIONS)

Note:

The Southern regions are defined as: East Anglia, East Midlands, South East including Greater

London, and South West

11

CHART 3B: RATIO OF COURT ORDERS TO ACTIONS, IMPLEMENTED ORDERS TO COURT ORDERS AND CHANGE IN NOMINAL HOUSE PRICES (NORTHERN REGIONS)

Note: Wales

The Northern regions are defined as: North, North West, Yorkshire & Humberside, West Midlands and

Our econometric model is used to argue that, taking into account the deterioration in the economic environment and the lags, there appears to have been a fall in the ratio of orders to actions, particularly in 1991-4. This is interpreted as a relaxation of court policy. However, this feature of the data is not obvious to the naked eye examining Chart 3A and 3B: it only becomes apparent when controlling for the other forces acting on court orders. The empirical evidence is set out in Section 5 and its appendix.

That there has been a shift in court policy is visually more obvious in examining the ratio of implemented orders to total orders since 1990, the period over which the data have been available, also shown in Charts 3A and 3B. This generally shows a decline since 1990, despite the deterioration in the economic environment. The increasing use of suspended court orders over this period can be taken to be a sign of a more lenient attitude by the courts particularly in 1991 and 1992 when house price falls and rises in unemployment were at their most severe. Charts 3A and 3B also show the rate of change of nominal house prices, clearly revealing the negative shocks occurring in 1991-2 in the Southern regions.

The unemployment rate data throw further light both on the timing and the extent of the recent recession in different regions. Chart 4 shows that in Southern regions, unemployment
12

rates reached their low point in 1988, while in Northern regions this occurred in 1989. Furthermore, while in most of the Southern regions, the rise by 1993 led to unemployment levels similar to or higher than in 1986, in all Northern regions, the 1993 peak was still substantially below 1986 levels. Thus, regional differentials in unemployment rates narrowed sharply in the 1990s.

CHART 4: UNEMPLOYMENT AND VAT DEREGISTRATION RATES

Note:

The Southern regions are defined as: East Anglia, East Midlands, South East including Greater

London, and South West and the Northern regions are defined as: North, North West, Yorkshire & Humberside, West Midlands and Wales

The peak year for business failures, as measured by de-registrations of firms from the VAT register shown in Chart 4, came one year earlier, in 1992, in all regions, though the peak was a little higher in Southern regions.

4.

ECONOMETRIC MODELS OF MORTGAGE POSSESSIONS

In our view, the probability of possession is the result of the simultaneous occurrence of two factors: a vulnerable debt/equity position and a trigger factor including elements such as an unfavourable cash-flow position and unfavourable expectations for an improvement. A rational borrower would not default on a mortgage just because of cash-flow problems if the equity cushion relative to debt was sufficient to allow trading down or out.

13

Similarly, someone able to meet mortgage payments is unlikely to seek possession: UK borrowers who default face a high probability of being pursued for their unpaid debt in the future and of being denied access to credit for at least some years. background to this is set out in more detail in the Appendix to this section. The theoretical

This is a more general view of mortgage possession than the option pricing approach popular in the US literature. (see Kau et al (1992), applied to UK data by Ncube and Satchell, (1994) and discussed by Dale (1995).) The Box below summarizes the key points of the option pricing approach and objections to applying this approach in the UK.

14

THE OPTION PRICING VIEW OF MORTGAGE POSSESSION Assumptions (a) (b) (c) (d) (e) as in many US states, the borrower's liability ends when he or she gives up the property deeds to the lender. The possession decision is in the hands of the borrower. There are no transactions costs. There are no restrictions in access to credit: thus no risk of acquiring bad credit record. rationality and good information.

Conclusions • • Given assumptions (a) to (e), the rational borrower calculates the present value of mortgage payments and defaults if this exceeds the value of the house by some margin. This margin is not zero because by defaulting now, the borrower would give up the option of recovery or of defaulting in the future which has some value. Option pricing theory, assuming that house prices and interest rates follow simple statistical processes, is used to compute this margin.

Objections • Vandell (1995) reviews this US literature and the empirical evidence and argues that this approach is defective in a number of ways since it omits transactions costs, credit market imperfections, trigger events such as divorce or shocks to cash flows such as unemployment, and also omits the behaviour of mortgage lenders. Moreover, assumption (a) is violated in the UK: borrowers remain liable for their debt even after possession and lenders or their insurance companies can pursue borrowers for the negative equity and costs that remain after the possessed house has been sold off.



Nevertheless, a precarious net equity position is undoubtedly an important element in the probability of possession and can be defined by the mortgage debt/equity ratio exceeding some threshold, where mortgage debt includes arrears. However, the probability of possession also depends on a trigger function which depends on current cash flows and shocks to income and house prices. Specifically, we assume the trigger rises with the debt
15

service ratio and the change in the unemployment rate and falls with positive house price shocks as measured by the rate of change or the rate of acceleration in house prices.

One can think of default arising from the intersection of the events `debt/equity ratio exceeds some threshold' and `trigger exceeds another threshold'. We use standard probability arguments to arrive at an expression for individual default probabilities and hence aggregate default rates by region. The economic fundamentals are represented by debt/equity ratios and the trigger effects mentioned above. See the appendix to this section for a fuller explanation of these points.

We model three aspects of the legal process dealing with mortgage default using County Court data by region: the ratio of implemented to total court orders, the ratio of total court orders to actions and the rate of actions itself. In each case, the outcome should depend on the economic fundamentals varying by region and by time, as discussed above, and on time effects for 1991 to 1996, the same across regions, capturing potential shifts in policy and the delayed reaction of courts and lenders to the possessions crisis. In addition, the equation for every region includes a factor specific to that region as a 'fixed effect' to capture long-run differences between regions such as differences in age and occupational structure, the ownership of financial assets and inequality within regions. See the appendix to this section, eq(5), for an illustration of the actions equation. Worse economic fundamentals should raise the court actions rate, decided by lenders, but also the ratio of implemented orders and the rate of court orders, determined by the courts, given actions brought. With worse economic fundamentals, the probability of a household avoiding eventual default falls so that, ceteris paribus, a higher proportion of actions brought results in court orders granted and implemented (rather than suspended) ones at that.
16

To the list of economic fundamentals discussed above, we also added proxies for lending quality. For example, one of these was derived from changes in the market share of centralized mortgage lenders in the previous five years. Ford et al (1995) provide evidence on the higher default rates of mortgages from this source. Poor lending quality in the past should be associated with higher default rates. This effect shows up only in the equation for the rate of court actions. The previous year's rate of VAT deregistration as a proxy for small business failure was also found to affect the rate of court actions, but not the ratio of implemented to total court orders or the ratio of orders to actions.

17

Appendix to Section 4: Theoretical Background A precarious net equity position is undoubtedly an important element in the probability of possession and, when c is small, can be defined by ln (mortgage debt/equity) > c (1)

where mortgage debt includes arrears. However, the probability of possession also depends on a trigger function which depends on current cash flows and shocks to income and house prices. Suppose trigger = f(debt service ratio, ∆ur, ∆lnhp) and the debt service ratio, (2)

dsr =

(mortgage debt)r -------------------y

(3)

where r is the tax adjusted mortgage interest rate and y is personal disposable income, ur is the unemployment rate and hp is an index of house prices.5 The change in the unemployment rate ∆ur is taken as a measure of income shortfalls and the rate of change of house prices ∆lnhp as a proxy for shocks in the equity position. The debt service ratio and the change in the

unemployment rate are expected to have a positive effect on the trigger, while the rate of change of house prices should have a negative effect.

18

Formally speaking, one can think of default rising from the intersection of the events ln(mortgage debt/equity) > c and trigger > c0. Then: Prob(default) = Prob(bad debt/equity) x Prob(bad trigger given bad debt/equity) (4) If the two events, 'bad debt/equity' and 'bad trigger' were independent, Prob (default) = Prob (bad debt/equity) x Prob (bad trigger)

In practice, the two events will be positively correlated. In any event, we will be taking an approximation of (4) using the log mortgage debt/equity ratio and the variables in the trigger function by translating these probabilities for individual households into relative frequencies of regional populations of households.

We have regional data on average house prices and average mortgages but not on the distribution of debt/equity ratios from which the vulnerable tail of the distribution could be analysed. Instead we rely on the existence of a relatively stable relationship between the vulnerable tail and the mean of the distribution.

To make this more concrete, consider the distribution of log debt/equity illustrated in Figure 1. Define the vulnerable tail as being the part of the distribution where log debt/equity exceeds c. Note that the point 0 marks the point where debt equals equity. The area under the vulnerable tail of the distribution with mean M1 is shaded in grey. Now, suppose there is a fall in average house prices which shifts the distribution, denoted by the dotted line, to the right, so that the new mean is M2. The area under the vulnerable tail is now the bigger hatched area. Generally, as the mean M moves to the right, the vulnerable tail area increases. We now suppose that the log of
19

the vulnerable tail area can be approximated as a linear function of log mean debt/equity6, on which we actually have data.

The theory above has been developed for mortgage possession considered broadly. In practice, possession can be initiated by mortgage borrowers or by lenders. Given information

asymmetries between them and different objectives the possession probability given by (4) will not have exactly the same relationship with economic fundamentals for borrower or lender initiated possessions. However, one would still expect the same set of economic fundamentals to be operative in both cases. Most court actions fall into the latter category, but not all. As Ford et al (1995) observe, for households in possession to obtain access to local authority housing, possession has typically to be the result of court proceedings. In some cases, therefore, court actions may follow at the initial request of the borrower. However, it is safe to assume that the number of actions is not under the direct influence of the courts, though it may depend on the perceived probability of success, which may depend on court policies in the recent past.

Thus we hypothesize that in the ith region the log percentage of court actions can be expressed by the following equation which can be stated in words as follows:

log(court actions rate)it = constant + regional fixed effecti + time effectt + a1log (debt equity)it-1 + a2 debt service ratioit + a8 change in unemployment rateit + a4 VAT deregistration rateit-1 - a5 rate of change of house pricesit - a6 rate of change of house pricesit-1 + a7 indicator of poor lending qualityt-1

20

In symbols, this translates

logpait

= a0 + biDregioni + ctDyeart + a1 ldeit-1 + a2dsrit + a3∆urit + a4 VATit-1 - a5 ∆lhpit -a6∆lhpit-1+ m5dsclt-1 (5)

where Dregioni is a dummy variable which is 1 for region i and 0 for other regions Dyeart is a dummy variable which is 1 for year t from 1991 onwards and 0 for other years ldeit-1 is last year's value of the log (deb/equity) ratio dsrit is the debt service ratio in region i in year t ∆urit is the change in the unemployment rate VATit-1 is last year's VAT deregistration rate in region i, a proxy for the small business failure rate ∆lhpit is the change in the log house price index m5dsclt-1 is the 5 year moving average of the lagged change in the share of centralized mortgage lenders in total mortgages outstanding, a proxy for lending quality.7 Equation (5) was arrived at by testing down from a more general specification including current and lagged values of the debt equity ratio, up to 2 lags in the debt service ratio, up to 2 lags in the unemployment rate and in the rate of change of house prices. This should be a general enough specification to be able to proxy reasonably well expectations by lenders of house prices, interest rates and incomes which could be relevant to the decision to bring a possession action.

21

As noted in Section 2(b) above, the time effects in eq.(5) are likely to represent a mix of influences: policy shifts by lenders, adjustment lags by lenders and a 'discouraged actions' effect as a result of a policy shift by the courts, resulting in a lower success rate for actions brought.

It is also conceivable that the time effects partly reflect a composition shift in the population of households at risk of possession. Suppose, for example, that lenders brought actions against the most disastrous cases first. This would suggest high rates of possession at first, followed by some tailing off after the worst cases had been eliminated from the population at risk. One test of this interpretation involves testing the alternative hypothesis that the time effects are homogenous across regions. Since regions differ considerably, particularly in the timing and size of house price changes but debt/equity ratios, these composition effects should occur at different times in different regions but we can reject this. Another test of the change in composition hypothesis comes from noting that it implies a positive time effect in 1990 to compensate for later negative ones. We test this pattern in Section 5 and in Section 7 on UK aggregate data.

Something like a pure court policy shift is likely to be directly measured by the estimated time effects in the equations for the rate of court orders, po, given the rate of actions brought. Court orders are at the discretion of the courts and not of the mortgage lenders bringing the actions. Since court procedures take time, we must build in a lagged reaction of orders to actions. But it is possible to accept the restriction that, in the long run, orders move in proportion to actions, provided one controls for the effect of changes in the economic fundamentals: worse economic fundamentals raise the ratio of orders to actions. The restriction can be imposed via an

'equilibrium correction model' (see Hendry (1995)) linking logpo and logpa:

22

∆logpoit = b1 ∆logpait + b2(logpait-1 - logpoit-1) + effects from region and year dummies and economic fundamentals

(6)

Note that b2 < 1 and b1=0 would indicate partial adjustment of orders to last year's actions rate, while b2 < 1 and b1=b2 would indicate partial adjustment to this year's actions rate. Eq.(6) encompasses both.

The other evidence for a shift in court policy can be found in the data for the ratio of implemented to total court orders, rio ie., 1-ratio of suspended to total court orders. This can be derived from the estimated time effects in an equation similar to (5) above but formulated for the ratio of implemented orders. We turn to the evidence in Section 5 and its appendix.

23

5.

EMPIRICAL FINDINGS FOR REGIONAL DATA

Full empirical evidence is set out in the appendix to this section. A summary now follows:

The Ratio of Implemented to Total Court Orders We find that worse fundamentals, as seen in a higher debt/equity ratio, an increase in the unemployment rate, a high debt/service ratio and negative house price shocks, all significantly raise the ratio of implemented court orders. For example, a rise in the debt service ratio from 0.4 to 0.5, such as might be caused by the tax adjusted mortgage interest rate rising from 8% to 10%, after two years results in a 21% rise in the ratio of implemented to total orders, even ignoring the knock-on effects on unemployment and house prices.

In terms of policy shifts, it appears that in 1991 and 1992, the ratio of implemented court orders was around 19% lower than would have been expected purely on the basis of the economic fundamentals. By 1993, this apparent softening of policy had fallen to 6% and had virtually disappeared by 1994. Though the precision of these findings must be qualified by the fact that these data only begin in 1990, they are consistent with the view that court procedures softened. The regional fixed effects suggest that the Southern regions tend to have slightly lower ratios than one might have expected given the economic fundamentals.

Court Orders Given Court Actions The model incorporates a lagged reaction of court orders to court actions resulting from typical delays in court procedures. The economic fundamentals found significant here are the debt/equity ratio, the change in the unemployment rate, the change in the debt service ratio and the rate of acceleration of house prices. A 10% rise in the debt/equity ratio results in the long run in a rise of around 6% in the ratio of court orders to actions.
24

In terms of policy shifts, there was a drop in court orders relative to actions of around 14 or 17% in 1991-92 compared with what might have been expected given economic fundamentals. The effect then fades in 1993-94 to one fourth of this magnitude and we can accept the hypothesis that it has disappeared by 1995. The fixed effects here suggest little systematic difference between southern and northern regions.

The appendix to this section suggests that, while the general pattern is robust, the precise estimates of the policy shift effects have some sensitivity to the econometric specification.

Modelling the Rate of Court Actions The rate of court actions responds strongly to last year’s debt/equity position (a 10% increase in the latter causing a 5 to 6% increase in the former) and in addition to the current change in nominal house prices (a 10% decrease raising the current court action rate by 14% and to high debt service ratios in the current year (a rise from 0.4 to 0.5 in the debt service ratio results in a 12% rise in the rate of court actions). There are also smaller effects from unemployment shocks and the previous year’s rate of VAT deregistration (for example, if this rises from 12% to 13% there would be a 5 to 6% rise in the court actions rate).8 The proxy for lending quality based on the change in market lenders in the previous five years has a significantly positive effect. It suggests that the aftermath of lower lending quality of the late 1980s involved a rise of about 18% in the court actions rate.

The pattern of the time effects proxying various aspects of a policy shift by lenders and their reaction to the policy shift in the courts suggest an effect beginning at around 30% in 1992 and gradually diminishing to around 8% in 1996. Thus, even in 1996, court action rates were
25

about 8% below what would have been expected given the economic fundamentals. The decline between a 17% effect in 1995 and the 8% effect in 1996 may also reflect the tightening of DSS rules for benefit claimants discussed above. Thus, this evidence is

consistent with a shift in policy by lenders after the November-December 1991 implicit contract between the Government and the lenders, and the view that lenders were generally still sticking with their agreement not to put into possession DSS claimants whose mortgage payments were being made direct to mortgage lenders. Assuming the proportion of such cases in the total of households at risk has declined, one would expect this effect to fade over time, as suggested by our estimates.

This pattern of time effects is broadly similar in an alternative, less well fitting equation, that results when the lending quality is omitted. However, there are then notable readjustments of the effects attributed to the economic fundamentals. For example, the debt service ratio and the change in the unemployment rate become more important while the effect attributed to house price changes falls. The details are discussed in the appendix to this section.

For court action rates, the regional fixed effects suggest strongly that, given economic fundamentals, the southern regions experienced systematically lower court action rates and Wales, in particular, systematically higher court action rates than the (other) English regions. There could be a number of reasons for this, including the greater ownership of financial assets, and the greater preponderance of non-manual workers in the southern regions. It is well-known that such workers tend to be lifted more easily out of cash-flow problems by higher earnings increases than manual workers.

26

Earlier we discussed the hypothesis that lenders may have held off from possession actions when house prices fell more than 10% below the 1988-90 average, since given selling costs and discounts, the lenders would then be sharing losses with the insurers. A variable

measuring the number of quarters in each year when prices were more than 10% below the 1988-90 average was constructed. This would have had a negative effect had the hypothesis been valid. However, its effect is positive in practice implying that economic fundamentals overwhelmed such a strategic response by the lenders. We also tested the hypothesis that policy shifts were significantly different across the North/South divide and were able to reject this hypothesis. This also casts doubt on the ‘change in composition’ hypothesis discussed in Section 2(b) and Section 5 which attributes the decline in possessions to the early weeding out of the most at risk cases. The hypothesis also implies a positive time effect in 1990 (and perhaps 1991) followed by negative ones. This pattern can be rejected.

27

Appendix to Section 5

Our data analysis begins with two sets of data focussed on policy shifts by courts rather than by lenders.

(a) The ratio of implemented orders We begin with the evidence from the data on the ratio of implemented orders, rio. The data here begin only in 1990. With only 7 years of data, the strength of the econometric evidence will be limited. Nevertheless, testing down from a more general specification gives the following, using weighted least squares, where regions with a larger number of mortgages are weighted more heavily: rioit = 0.37 (5.6) -0.09DSE (1.8) -0.21D91 (4.5) -0.27∆lhpit (1.6) -0.07DEA (1.6) -0.15DSW (3.4) -0.17D92 (4.1) +1.08avdsrit (5.9) -0.05DEM (2.1) +0.11DWW (0.4) -0.06D93 (2.4) -0.06DNN (2.7) -0.04DWM (2.0) +0.21ldeit-1 (1.9) +0.06∆urit (3.9) (7) -0.02DNW (1.1)

s.e. = 0.0353, R2 = 0.925, DW = 1.96 sample 1990-96, 63 observations Key: DEA, DEM etc are regional dummies where EA=East Anglia, EM=East Midlands, NN=North, NW=North West, SE=South East, SW=South West, WW=Wales and WM=West Midlands. D91, D92,D93 are the time dummies, respectively one in the years 1991, 1992 and 1993 and zero otherwise. lde=log (debt/equity), ∆ur=change in the unemployment rate, ∆lhp=change in log house price index, avdsr=two year moving average of debt service ratio. Note: Throughout the paper, Durbin-Watson statistics are corrected for regional breaks, absolute values of t-ratios in parentheses, and R2 is adjusted R2.
28

Worse economic fundamentals in the form of the two year average of the debt service ratio, the change in the unemployment rate and the log debt/equity ratio and the rate of change of house prices all raise the ratio of implemented orders. The region dummies do not show very systematic regional variations, though Wales and Yorkshire and Humberside (the reference region) followed by the North West show the highest ratios of implemented orders, while the South West followed by the South East show the lowest.

The time effects for 1991-96 proved significant only for 1991, 1992 and 1993. They suggest that policy was softest w.r.t. suspended orders in 1991 and progressively tightened in 1992 and 1993, returning to 'normal'levels by 1994.

Specification (7) was tested in two ways. First, against a more general specification of the economic fundamentals and time effects for 1994 to 1996. An F-test of (7) against a more general alternative involving 3 more time effects and 4 more parameters for the economic fundamentals (adding ldeit, urit, dlhpit-1 and dsrit-1 effects) gives F7,40 = 0.422 [p=0.88] which is insignificant at the 5% level. A second test involved adding interaction effects between the 3 time effects and a dummy for southern regions. This checks that the claimed homogeneity of time effects across regions is statistically acceptable. Since economic circumstances differed considerably between southern and northern regions, this is quite a powerful specification test. The F-test here gives F3,44 = 0.774, which is also insignificant [p=0.51] at the 5% level.

It is conceivable that ∆lhpit is endogenous in that a shock to rioit might feed back within the year to depress house prices further. A simple check on this is to instrument ∆lhpit using the fitted values from a parsimonious system of regional house price equations. The potential OLS bias is
29

in a negative direction. The IV estimate of the coefficient on ∆lhpit is marginally more negative still, suggesting the absence of any bias.

b.

The orders rate conditional on the actions rate

As discussed in the appendix to Section 5, the specification allows for lags in the response of court orders to court actions. Here we can estimate for the period 1987 to 1996. Testing down from a more general specification gave the following results: ∆logpoit = 0.23 (3.1) + 0.43 ∆logpait (6.1) +0.02DEM (0.5) +0.04DSE (1.5) +0.003DWM (0.01) +0.51∆2dsrit (2.1) -0.14D92 (1.9) R2 = 0.953 90 observations +0.71(logpait-1 - logpoit-1) (10.0) +0.10DNN (2.7) -0.04SW (1.5) 0.41ldeit-1 (4.4) -1.67av∆2lhpit (8.2) -0.03D93 (0.5) DW = 2.28 -0.04D94 (1.2) (8)

-0.003DEA (0.01) +0.05DNW (1.9) +0.02DWW (0.7) +0.06∆urit (3.0) -0.17D91 (2.8) s.e. = 0.06 sample 1987-1996

Note: Durbin-Watson statistic corrected for regional breaks.

where av∆2lhp is the average rate of acceleration of house prices in the current and the past year and ∆2dsrit is the two year change in the debt service ratio.

30

Testing this specification against an alternative also including year dummies for 1995 and 1996, the current debt equity ratio, the current rate of change of house prices, the unemployment rate, the level of the debt service ratio, the lagged VAT deregistration rate and the average growth of market share of centralized mortgage lenders over the previous 5 years as a proxy for lending quality gives an F-test of F8,63 = 1.31[p=0.26], so that specification (7) cannot be rejected. Specification (7) can also be accepted against an alternative which includes the interactions of a southern regions dummy with the year dummies for 1991 to 1994. This gives an F-test of F4,64 = 0.76[p=0.56].

Eq(8) potentially suffers from the two endogeneity biases, one from the potential endogeneity of ∆logpait and the other from the potential endogeneity of ∆lhpit, a component of av∆2lhpit. As a check on this, eq(8) was re-estimated instrumenting ∆logpait, using eq(10) below where ∆lhpit was replaced by its fitted value, and instrumenting ∆lhpit in eq(8) by its fitted value from a parsimonious system of regional house price equations. Little change in the estimated parameter values results. Indeed the estimated effect of av∆2lhpit becomes a little more negative still and the estimated effect of ∆logpait becomes slightly more positive, both in the opposite direction of the anticipated direction of the bias.

To make the point that estimates for these time effects do have some sensitivity to variations in the specification, consider the effect of omitting the change in the debt service ratio in (8): we then find

31

∆logpoit =

0.16 (2.3)

+0.45∆ logpait +0.74(logpait-1 - logpoit-1) (6.5) (10.3) +0.02DEM (0.6) -0.05DSW (1.4) +0.045 ∆urit (2.5) -0.21D92 (3.0) DW = 2.32 +0.10DNN (2.7) +0.02DWW (0.4) -1.85av∆2lhpit (9.7) -0.10D93 (2.4) -0.08D94 (2.6) (9) +0.05DNW (1.7) -0.01DWM (0.4)

+0.04DEA (0.01) +0.03DSE (1.4) +0.33ldeit-1 (3.8) -0.20D91 (3.3) s.e. = 0.0624 R2 = 0.951,

The most interesting difference between eq(8) and eq(9) is in the pattern of time effects reflecting the softening of policy and the decay of this policy shift. Eq(9) suggests a peak in 1992 and then a monotonic decay, which looks marginally more plausible than the pattern of point estimates in eq(8). Given the limitations of the evidence on court policy shifts from the ratio of implemented orders in subsection (a) above, the evidence in (8) and (9) is fairly consistent with that in (7). Both specifications suggest little systematic regional variation in the ratio of orders to actions between regions, though the North stands out slightly as having a high ratio where the South West has a low ratio of orders to actions.

c.

The court actions rate

The specification was discussed in the appendix to Section 4, see eq(5). As noted there, the time effects here reflect a mix of policy shifts and adjustment lags by lenders, in part reacting to court policy shifts. The results are as follows: logpait = 2.00 (7.2)
32

-0.19DEA (3.9) -0.20DSE (3.3) +0.55ldeit-1 (5.9) -1.54 ∆lhpit (10.2) -0.27D92 (8.4)

-0.04DEM (1.3) -0.20DSW (4.6) +0.03 Durit (2.3) -0.53 ∆lhpit-1 (3.9) -0.27D93 (5.3)

-0.06DNN (1.7) +0.30DWW (6.7) +1.23dsrit (4.8)

+0.13DNW (3.2) +0.15DWM (5.2)

+5.02VATit-1 +16.2m5dsclit-1 (2.6) (3.1) -0.25D94 (5.4) -0.19D95 (3.5) -0.08D96 (1.1) (10)

s.e. = 0.0583, R2 = 0.969, DW = 1.88, sample 1987-1996, 90 observations

This specification of the three has the most comprehensive set of significant economic fundamentals including VAT deregistration and the proxy for past lending quality. Testing this specification against an even more comprehensive one containing also the current log debt/equity ratio, the unemployment rate, the lagged debt service ratio, a lagged dependent variable and a 1991 dummy gives an F-test, F5,62 = 1.00 [p=0.43] which is not significant at the 5% level.9 Testing against an alternative that permits the time effects to be different for southern regions gives an F-test F5,64 = 1.56 [p=0.18] which is also not significant at the 5% level.

Again, a check on the endogeneity of ∆lhpit contradicts the hypothesis of an endogeneity bias. The point estimate of the instrumented coefficient is marginally more negative and a Hausman test confirms the absence of bias [p=0.77].

The regional dummies suggest a systematic North/South pattern of differences: compared with the reference region, of Yorkshire and Humberside, East Anglia, the South East and the South
33

West all have substantially lower court actions rates, while Wales, the West Midlands and the North West have substantially higher court actions rates than economic fundamentals would have predicted.

These regional fixed effects capture long-run differences between regions, for example, in age and occupational structure, income-age profiles, income riskiness, and the ownership of financial assets and hence in the relationships between the average and the size of the vulnerable tail for the distributions of debt/equity and debt service ratios. Thus, for example, the lower court action rates in the southern regions may reflect the greater ownership of financial assets there, see Regional Trends, the higher proportion of white collar workers and the somewhat younger age structure. It is well known that income-age profiles peak later for non-manual workers than for manual workers. Since younger households typically have greater debt exposure, the fact that in southern regions, many of these households can expect substantial age related income increases means that for given average debt/equity and debt service ratios, it is likely that a higher proportion of southern households would find themselves lifted away from the mortgage default margin. This would help to account for the lower court action rates in southern regions, given the measured economic fundamentals included in our equation.

This is all still consistent, however, with a higher 1991 peak for the court actions rate in southern regions than in the rest of the economy: that, of course, is explained by the greater deterioration in economic fundamentals in the southern regions.

As noted earlier, the time effects for court actions are likely to reflect in part the impact that 'softer' court policies would have had on the propensity of mortgage lenders to bring actions. The lower the probability of success, the less likely is an action to be brought. This 'discouraged
34

action' effect is likely to lag behind the shift in court policies. Indeed, we find no detectable effect in 1991 of a softening of lenders policies as reflected in the court actions rate. Recall also that the implicit contract between the mortgage lenders and the government to reduce possessions rates in return for various government concessions was negotiated in November/December of 1991 so that hardly any of the consequences would have shown up in the 1991 figures.

An alternative interpretation of the negative time effects in 1992-5 discussed in Section 2(b) suggested that they may have been the result of a shift in the composition of borrowers at risk in which lenders dealt with the most disastrous cases first. This would then have led to a tailing off of the proportion of actions brought with the worst cases eliminated from the population at risk. This hypothesis, however, implies given the late 1989 peak in house prices, a corresponding increase in 1990 and/or 1991 above levels warranted by observed economic fundamentals, in the rate of court actions brought. To test this we included a 1990 and a 1991 year dummy with respective coefficients -0.01(t=0.2), -0.07(t=0.6). The hypothesis can therefore be rejected in favour of our policy softening alternative.

These estimates suggest a steady decline in the softening of lenders policies, though the effect had still not quite faded out in 1996, with the point estimate suggesting an effect about one quarter of the 1992 effect. This is consistent with the hypothesis that those who bought in 198891, the cohort likely to have been most subject to mortgage default, are making up a lower and lower proportion of new cases.

35

If we omit the proxy for poor lending quality, the lagged dependent variable becomes significant and we obtain a more gradual fading out, with the 1996 effect only just under one half of the 1992 effect, see eq.(11):

logpait =

1.06 (2.8) -0.18DEA (3.3) -0.31DSE (5.8) +0.40ldeit-1 (3.1) +1.83dsrit (8.4) -0.31D92 (8.1)

+

0.30logpait-1 (4.5) -0.04DEM (1.2) -0.24DSW (5.2) -0.02DNN (0.4) +0.26DWW (5.3) +0.07DNW (1.5) +0.09DWM (2.5)

+ -

0.06 ∆urit (5.3) 1.09 lhpit (7.5) -0.23D93 (4.2) +5.20VATit-1 (2.5) -0.29D94 (5.8) -0.18D95 (3.1) (11) -0.13D96 (2.1)

s.e. = 0.0640, R = 0.962, DW = 2.34 sample 1987-1996 90 observations

2

Whether one accepts the implications of eq(11) - slow fade out - or of eq(10) - fast fade out of policy softening - depends on whether one regards the lending quality proxy in eq(10) as plausible. Certainly eq(10) fits better.

36

6.

FORECAST - SCENARIOS FOR 1997

This section considers some forecast scenarios for 1997. An analysis of aggregate quarterly data suggests that the UK possessions rate is forecast better from data on the court actions rate (for England and Wales) in the previous four quarters, together with some economic fundamentals, than from data on the court orders rate combined with economic fundamentals. Since our earlier discussion suggests that court orders depend on court actions, forecasting court actions is the key. We now consider three forecast scenarios applied to each of our respective specifications (10) and (11) detailed in the Appendix to Section 5. These are included to make the point that differences in the econometric specification do have forecast implications. Specification (10) has somewhat weaker interest rate and unemployment effects and somewhat stronger house price effects than specification (11). Specification (11) also suggests that it takes a little longer for these effects to feed through.

The common features of the scenarios set out here are the assumption of a 1.2% fall in the unemployment rates in South East, the South West and East Anglia and a 1% fall elsewhere, nominal personal disposable income per head rising by 7.4% in these Southern regions and 6.4% elsewhere, the same change in the VAT deregistration rate as in the previous year, and the assumption, as far as policy by the lenders is concerned, that the court actions rate is 5% below the rate implied by economic fundamentals (as opposed to 8% in 1996). We have assumed the tightening of the DSS rules is already reflected in the 1996 time effect so that there is no additional effect in 1997.

Scenario I is a moderately optimistic one with stable interest rates and a significant house price recovery. Scenario II is a pessimistic one. Scenario III is the most optimistic both about interest rates and house prices. Table 3 summarizes the key points.
37

Scenario I

Scenario II

Scenario III

Adjusted mortgage interest rate in 1997

same as 1996

1 percentage point higher

same as 1996

House price increase in South East, East Anglia, South West

12%

10%

17%

House price increase in rest of England and Wales

7%

6%

12%

38

Table A.1 Regional Court Action Rates for Mortgage Possession: Forecasts and Historical Data (Percentages) Scenario Scenario Scenario I II I II III I II III 1991 1995 1996 1997 1997 1997

III

East Anglia East Midlands South East South West North North West Wales West Midlands Yorks & Humberside

1.76 1.86 2.45 2.05 1.47 1.98 2.21 2.08 1.74

0.77 0.90 0.93 0.85 0.70 1.10 1.14 0.94 0.86

0.76 0.82 0.85 0.76 0.79 1.05 1.04 0.86 0.87

0.65 0.81 0.71 0.68 0.67 0.92 0.82 0.87 0.81

0.71 0.86 0.78 0.73 0.70 0.97 0.86 0.92 0.85

0.61 0.75 0.66 0.63 0.62 0.85 0.76 0.81 0.75

Southern Regions Northern Regions England & Wales

2.24 1.90 2.10

0.90 0.95 0.92

0.82 0.93 0.87

0.71 0.84 0.77

0.78 0.88 0.82

0.66 0.78 0.71

Source: see text

39

Note: results based on equation 10. These results are marginally different from Table 4 in the Housing Finance paper for reasons discussed in footnote 9.

Table A.2 Regional Court Action Rates for Mortgage Possession Forecasts and Historical Data (Percentages) Scenario Scenario Scenario I 1991 1995 1996 1997 II 1997 III 1997

East Anglia East Midlands South East South West North North West Wales West Midlands Yorks & Humberside

1.76 1.86 2.45 2.05 1.47 1.98 2.21 2.08 1.74

0.77 0.90 0.93 0.85 0.70 1.10 1.14 0.94 0.86

0.76 0.82 0.85 0.76 0.79 1.05 1.04 0.86 0.87

0.70 0.87 0.71 0.70 0.74 0.99 0.92 0.91 0.88

0.76 0.94 0.79 0.77 0.79 1.06 0.98 0.97 0.95

0.66 0.82 0.68 0.66 0.70 0.94 0.87 0.86 0.84

Southern Regions Northern Regions England & Wales

2.24 1.90 2.10

0.90 0.95 0.92

0.82 0.93 0.87

0.73 0.90 0.81

0.81 0.97 0.88

0.69 0.86 0.77

Source: see text Note: results based on equation 11.

40

The differences in specification of equations 10 and 11 noted above most relevant for forecasting into 1997 are two: eq(10) displays smaller effects from the debt service ratio while eq(11) displays more persistence, via the role of the lagged dependent variable. The latter is the main reason why the fall in court action rates between 1996 and 1997 in Scenario I is larger in Table A.1 (from eq(10)) than in Table A.2 (from eq(11)).

With a larger debt service ratio effect in eq(11), Table A.2 then shows a sharper rise in court action rates going from Scenario I to Scenario II than does Table A.1. Note that Table A.2 suggest that even a 1 percentage point rise in the average mortgage rate for 1997 compared with 1996, would be sufficient to undo the benefits operating via higher incomes, lower unemployment and the upward momentum in house prices: for England and Wales,

particularly for northern regions, the court actions rate is predicted to rise. Although Table A.1 has less negative implications, even it suggests that the magnitude of interest rate rises which the July 2nd Budget has put in prospect may well negate the effects of other improvements in economic conditions, particularly in northern Britain. However, much will depend on the size of the ripple effect that rises in house prices in London and the more affluent parts of the South East will have on other regions. Scenario III, which assumes the strongest house price rises, shows the differences both at the regional level and for England and Wales in comparison with Scenario I.

41

7.

UK EVIDENCE ON THE POSSESSIONS-ARREARS RELATIONSHIP

Chart 1 above demonstrated the shift that took place in the 1990’s in the relationship between CML data on the rate of possessions and the proportion of mortgages 12 months or more in arrear. Previous work by Breedon and Joyce (1992), Brookes, Dicks and Pradhan (1994) and Allen and Milne (1994) has modelled the CML (Council of Mortgage Lenders) data on possessions in terms of arrears and other determinants. These authors impose the constraint that in the long run, the flow of possessions moves in proportion to the number of households in long term arrears, defined variously as over 6 months or over 12 months. Chart 1 which plots the rate of possession, pp, plotted against the ratio, p12m, of households 12 months or more in arrear to the number of mortgages, shows that in 1991 possessions peaked and then declined while the stock of arrears continued to climb before turning down also. As discussed below, this cannot be explained by the measurement error in the arrears data which arises when interest rates change. Chart 1 also plots log p10f, the fitted value of the proportion of mortgages over 10% in arrear against the log of the possession rate.

Econometric analysis confirms this visual impression. Using a quarterly interpolation of the biannual CML data, we were able to find a simple relationship between the rate of possessions, the arrears rate, the debt service ratio, the unemployment rate and the change in house prices once time dummies for 1991-1997, reflecting policy shifts, are included. In the long run, the ratio of possessions to long-term arrears cases rises by around 50% for an increase in the debt service ratio of 25%, equivalent to a rise in the after tax interest rate from 8% to 10%. An increase in the unemployment rate from 8% to 10% in the long run also implies an increase in the ratio of possessions to arrears of around 50%.

42

The time effects show a clear hump shaped pattern beginning in 1991, peaking in 1993 and declining steadily to 1997. Their magnitude, particularly if they were given a steady state interpretation (see appendix), is clearly larger than of the corresponding time effects in regional court orders and court actions equations, respectively reflecting court policy shifts and policy shifts by mortgage lenders. To some degree this is likely to reflect the fact that policy shifts by both sets of agents are reflected in possessions outcomes. But a bigger factor is that the softening of lenders’ policy that occurred after the end-1991 implicit contract with the government, not only reduced possessions but increased the count of numbers in arrears. Indeed, the latter would be the logical consequence of the former. Econometric models of the arrears rate confirm a pattern of positive time effects between 1992 and 1996, also with a hump-shaped time profile. This must mean a double element in the gap between possessions and arrears.

The combination of the aggregate time series evidence on possessions and arrears also fails to support the hypothesis that the decline in the possessions rate from 1991 was caused by the early weeding out by lenders of the most at risk cases. This would have resulted in respectively positive and negative time effects in 1990 or even 1991 in the possessions and arrears equations followed by the reversal of these effects in subsequent years. But these 1990-1991 effects are absent.

43

Appendix to Section 7: Econometric Evidence from Aggregate UK Data Following a general to simple modelling strategy for data from 1983Q1 to 1997Q1, the following equation is obtained:

∆logppt = + -

0.83 (3.1)

+

0.27 (logp12mt-1 - logppt-2) (5.0) + 0.050 urt-2 (4.9) 1.36∆lhpit-1 (5.3) 0.47md92 (11.5) 0.19md96 (4.3) 0.50md93 (7.9) 0.11md97 (1.7) (12) 0.46md94 (6.6)

1.27 dsrt (6.3) 2.13∆lhpit (8.0) 0.13md91 (4.5) 0.33md95 (6.7)

s.e. = 0.0398 R2 = 0.882 sample period 1983Q1 - 1997Q1

DW = 1.84 n = 57

where pp p12m dsr ur lhp md91-md97

= = = = = =

rate of possession number of households with arrears over 12 months/number of mortgages outstanding debt service ratio unemployment rate log house price index moving average of time dummies discussed below.

The Lagrange multiplier tests for residual autocorrelation are all satisfactory (chi-squared statistics are respectively 0.21, 3.34, 2.75 and 4.16 for 1st, and up to 2nd, 3rd and 4th order autocorrelation). Note that the debt/equity ratio does not appear in this relationship though unemployment, the debt-service ratio and house price shocks are important as in the relationship
44

between possessions and court orders. In the regional orders and actions equations reported in the appendix to Section 5 we tested and rejected the possibility of within the year feedback from court orders and actions to house prices. In a quarterly context, feedback seems even less likely, and so we do not test for it. The time dummies capturing shifts in policy by courts and by lenders and possibly implementation lags by the latter enter in a moving average form - for example, md91 is equal to 0.25 in 1990Q4, 0.75 in 1991Q1, 1 in both 1992Q2 and 1991Q3, 0.75 in 1991Q4 and 0.25 in 1992Q1. This form has the effect of centering the total annual effect correctly at mid-year. The policy shift effect as estimated takes a hump-shaped form, building up to its strongest (negative) effect in 1993 and declining to 1997. These effects are large and highly significant. A proxy for lending quality, defined as in the appendix to Section 5, based on long moving averages of the change in the market share of centralized mortgage lenders proved insignificant. This suggests that the proportion of 12 month arrears cases sufficiently reflects the decline in lending quality that probably took place in the late 1980s.

In order to examine parameter stability, the model was also estimated up to 1990Q3 (omitting the policy shifts obviously). The estimated parameters are remarkably stable, with a slightly higher feedback term and a slightly lower effect of the lagged house price change.

∆logppt = +

1.00 (3.2)

+

0.33(logp12mt-1 - logppt-2) (4.6) 2.33∆lhpt (6.5) -

+

1.39dsrt (5.8) (13)

0.058urt-2 (4.7)

1.09∆lhpt-1 (3.1)

s.e = 0.0432, R2 = 0.897, DW = 1.68 Sample period 1983Q1 - 1990Q3. The chi-squared statistics for respectively 1st, 2nd, 3rd and 4th order residual autocorrelation are 0.08, 0.66, 0.24, 1.51, which again confirms the lack of residual autocorrelation.
45

This remarkably simple model implies that in the long run, the rate of possessions moves in proportion to the rate of 12 month arrears. However, if the debt service ratio or unemployment are high or if house prices are falling, then possessions will be higher than implied by arrears alone.

The feature of eq(12) that deserves further discussion is the remarkably large coefficients on the time dummies, for example -0.50 for the 1993 dummy. Had this been a permanent effect, it would have implied a long-run reduction of the log possessions rate, given the 12 month arrears rate, of 0.50/0.27 = 1.85, since the coefficient on the equilibrium correction term is 0.27. This is far bigger than the time dummy effects for the log court actions rate estimated in eq(10) or eq(11).

As noted in the main text of Section 7, a key component of the explanation of this apparently surprising finding lies in the nature of what a policy softening by mortgage lenders implies for the number of mortgages classified to be over 12 month in arrear. The crucial part of the implicit contract between the government and the mortgage lenders in November/December 1991 was that the lenders would refrain from exercising possession in cases where DSS mortgage payments were being made directly to mortgage lenders. This would have led to many cases creeping into the 12 months in arrear category without experiencing the possessions proceedings that would previously have ensued.10 Thus, the policy softening should have led to a significant increase in the proportion of mortgages 12 months in arrear. Indeed, an econometric model for the 12 months in arrear rate shows significantly positive time dummy effects in 1992-95. Thus the gap between the possessions rate and the arrears rate widened more than between the possessions rate and measures of economic fundamentals such as the debt/equity ratio, debt
46

service ratio etc., but excluding the arrears rate. The econometric evidence on the arrears rate suggests that about two thirds of the widening gap between the possessions rate and the arrears rate is due to the post-1991 shift in the arrears rate.

It is conceivable that another component of the explanation of why the 1992-5 time effects are so much bigger in eq(12) than eq(10) and eq(11) lies in the well-known measurement biases of the months in arrear data. When interest rates rise, monthly payments rise so that the number of months a mortgage is classified to be in arrear temporarily falls. Thus arrears levels are underestimated when interest rates are high and over-estimated when interest rates are low. Since interest rates fell for much of 1991 to 1995, arrears levels would have been overestimated relative to 1989-90, thus exaggerating the gap between possessions and arrears rates and helping to account for the size of the 1992-95 time dummies.

Since 1993, the Council of Mortgage Lenders have published a count of the number of arrears cases which should be free of this measurement bias based on a percent in arrears concept, eg., 5% in arrear, 10% in arrear. Muellbauer (1996) provides estimates back to 1983 of what the 10% in arrear rates would have been, had they been collected earlier. Imperfect though these estimates probably are, they do permit a check on the measurement bias hypothesis. Running eq(12) with the estimated log 10% in arrear rate replacing the log 12 months in arrear rate results in similar (indeed marginally higher) estimates of the 1992-95 time effects, suggesting that the measurement bias hypothesis is not the explanation of the large time effects in eq(12).

47

CONCLUSIONS AND OUTLOOK We have analysed regional data on court actions, court orders and implemented court orders to investigate empirically the influence of variations in the macroeconomic and the regional economic environment on these indicators of mortgage default. We have found evidence of substantial shifts in these relationships beginning in 1991 and similar across regions, reflecting what we term “policy shifts'' by Courts and mortgage lenders. As our discussion indicates, these shifts are probably a mix of intended changes in behaviour and initial delays in setting up systems to deal with the possessions crisis. Our evidence on the UK relationship between possessions and long-term arrears also shows these shifts in behaviour. The court orders data suggest that by 1995 County Court practices had returned to normal, given the court actions submitted to them. In contrast, the data both on regional court actions and on UK possessions cases suggest that by 1995 the mortgage lenders were still possessing at rates significantly below what one would have expected on the basis of the economic fundamentals. This would be consistent with the implicit promise made in 1991 not to possess in cases where DSS mortgage payments were being made direct to mortgage lenders.

The evidence is that this policy softening effect is gradually decaying. The outlook for the rate of possessions depends partly on how rapidly this effect decays, on the hard-to-quantify effects of the tightening of DSS rules from October 1995 for claimants, as well as on the macroeconomic fundamentals.

At least as far as the latter are concerned, there have been major improvements since 1995 with the further reduction in interest rates, rising house prices, especially in Southern regions and falling unemployment rates. The lags in the response of court actions, court orders and possessions rates to these macrofundamentals are considerable. This will mean that the falls
48

seen at the end of 1996 will be followed by significant further falls in 1997-98 provided a significant rise in interest rates can be avoided.

49

Data Appendix Data for 1986 to the present on court orders and actions for mortgage possessions in English regions and Wales are available from the Lord Chancellor's Department.

We want to scale these data relative to the number of mortgages outstanding per region. These are estimated, following methods pioneered by Anthony Murphy, as follows: from Labour Force Survey (LFS) Housing Trailers for 1971, 1981, 1984, 1988 and 1991-93, we obtain estimates of the fraction of owner-occupiers with mortgages, omi for region i, at the end of each year.11 We can obtain fairly accurate estimates of this fraction for the UK as a whole by dividing, nmUK, the number of mortgages outstanding (Housing Finance, Table 29) by ohsUK, the number of owneroccupied houses (Housing and Construction Statistics, Table 9.3), where all figures are at year end. Let romi be the ratio omi/omUK. By fitting a cubic in time to romi we generate interpolated estimates for 1985 to 1995, fromi. We then define the estimated nmI, the number of mortgages in region i by

nmi (nmUK) (fromi)

ohsi ______ ohsUK

Thus, the share of the number of UK mortgages in the ith region equals the share of owneroccupied houses scaled by fromi.

We define, pa, the percentage of mortgages in the ith region in year t subject to Court actions as (actit/nmit-1) x 100, where actit is the number of actions in region i, year t, and analogously for Court orders, po. In logs, these variables are denoted logpa and logpo, respectively.
50

Among the explanatory variables, several rely on estimates of ami, the average mortgage outstanding in region i. These estimates use information on rmpi, the ratio of average mortgage interest payments per household in each region to that in the UK from the Family Expenditure Survey for 1974-95. Data on the average UK mortgage stock could then be scaled by these ratios to obtain estimates of average mortgage stocks in each region. These ratios are not ideal for this purpose, being subject to large sampling variations, particularly in the less populous regions. We use estimated ratios interpolated from cubic equations in time fitted for each region to overcome sampling variation. Also, at times when many households are increasing their mortgage arrears, actual recorded mortgage payments will understate the size of mortgage debts. Later, when households are paying extra to reduce their arrears, mortgage payments will overstate the size of mortgage debts. To the extent that this occurs uniformly across regions, these biases will cancel out. However, we have reason to believe that mortgage arrears are higher in regions with high mortgage possession rates. Our estimates are necessarily crude, therefore.

We estimate average debt/equity ratios for each region by scaling the average mortgage amit-1 by hpit, an estimate of the average second-hand house price. This multiplies the Department of the Environment's (D.O.E.) mix adjusted index for the region by the average dwelling price for `other dwellings' in 1985 from Housing and Construction Statistics, Table 10.9. We denote the log of the debt/equity ratio as ldeit.

The debt-service ratio is computed as the product of the tax-adjusted mortgage interest rate (abmrit) and the average mortgage (amit-1) divided by an estimate of after-tax male earnings (yit) scaled by the ratio of average UK male tax adjusted earnings to UK personal disposable income per head. Note that this definition of the debt-service ratio omits the repayment element in regular mortgage payments. Thus,
51

dsrir = (abmrit) (amit-1)/yit

In turn, abmr is defined as (1-sittrt) (gross Building Society mortgage rate), where trt is the standard rate at which tax relief applies and sit is an estimate of the fraction of mortgages under the tax relief ceiling. sit varies from region to region. A simple estimate would define sit = 1 if the average mortgage is under the tax relief ceiling of £30,000 and sit = 30,000/amit-1 otherwise. But this would neglect the inequality of the distribution of mortgages which would, for example, ensure that some mortgages were over the ceiling even with an average mortgage of £25,000 say. sit incorporates an approximate adjustment12 to reflect this.

The Building Society mortgage rate comes from Housing Finance, Feb. 1996, Table 26. Up to 1987, the figures measure the annual average interest rate by dividing recorded payments by the average mortgage stock outstanding. These figures reflect premiums or discounts as well as delays for some societies between announcements of interest rate changes and their implementation. However, the financial years of some societies did not coincide with the calendar year in all cases. From 1988, the figures are an annual average of end of month rates, which also should reflect discounts and implementation delays. Abbey National and Cheltenham and Gloucester are excluded from July 1989 and August 1995 respectively.

After-tax male earnings by region comes from the New Earnings Survey and are April figures. The tax adjustment also varies across regions, estimating tax rates from Regional Accounts data on personal disposable income and personal income.

52

Regional unemployment rates derive from registration records at Job Centres, refer to all workers and come from Economic Trends.

Regional estimates of the fraction of businesses de-registered for VAT come from Department of Employment Gazette, November 1991 and updates were kindly provided by the Department of Trade and Industry.

A measure of credit quality was constructed from data on the share of centralized mortgage lenders in the value of total mortgages outstanding in the UK published in the Historical Compendium of Housing Finance Statistics. If the change in the share is denoted dscml, the 5 year moving average is denoted m5dscml and we work with the lagged value of this.

Quarterly data for Section 7 were obtained as follows. Data on the biannual flow of mortgage possessions cases and on the end of December and end of June count of cases where arrears were over 12 months are published by the Council of Mortgage Lenders. Interpolating (log-linearly) the arrears data is straightforward as one would expect, for example, the September figure on average to be half way between the June and December figures. Interpolating the biannual flow of possessions data into a quarterly flow is less obvious. Assuming that the two quarters in each half year had the same possessions levels or rates would be unrealistic: for example, if possessions are trending upwards, the earlier quarter should have a lower level than the later quarter. A moving average procedure in which, for example, the quarter 1 figure is given by one third of the first half figure plus one sixth of the last half of the previous year’s figure yields a first set of estimates. These are then scaled to ensure that in each half year the sum of the two estimated quarterly figures adds up to the recorded figures for the half year.
53

The definition of the debt service ratio for quarterly data is similar to that for annual data except that the income measure is personal per capita disposable income. Both it and the tax adjusted mortgage interest rate are defined at annual rates. The quarterly unemployment rate and

quarterly rate of change of house prices come from the sources quoted above.

54

ENDNOTES

1.

This paper is a revised and extended account of the research behind our paper, with the same title in Housing Finance, May 1997. This research was supported by the Council of Mortgage Lenders and the ESRC under grant R000 23 4954. We are grateful for helpful comments to Fionnuala Earley, Janet Ford, David Hendry, Duncan MacLennan, Anthony Murphy, George Speight, Mark Stephens, Peter Williams and seminar participants at the Universities of Glasgow and Oxford, at University College, Dublin, University College, London and at the CML Conference `Modelling Possessions and Arrears', 25 February 1997. The survey evidence of Ford et al (1995) suggests that the average family size of these households was not much smaller than of all households with mortgages. See Muellbauer and Murphy (1997) for an analysis of its causes. That is, formally to permit loan-to-value ratios for new lending to exceed 100% by significant margins. Note that this definition excludes the repayment component of debt service cost. For a given mortgage life, this moves in the opposite direction from the interest rate, stabilizing slightly the true debt service ratio. However, as suggested in Muellbauer (1996), the average mortgage life for borrowers facing possession or in arrears has fallen in the 1990s. This resulted in a substantially smaller reduction in the debt service ratio than would have been expected from the fall in interest rates. Log mean debt/mean equity is the log of a ratio of arithmetic means while mean log debt/equity, as illustrated in Figure 1 is the log of a ratio of geometric means. These are not identical but will move very closely together for the kinds of distributional shifts illustrated in Figure 1. It is widely believed that in aggressively bidding for market share in the latter half of the 1980s, the centralized mortgage lenders, who lacked a High Street base, were forced to accept higher risk customers or to lend at riskier loan to income or loan to value ratios. At any rate, the evidence from Ford et al (1995) is for a higher rate of possessions associated with mortgages from this source. Note this would imply almost a one for one increase in percentage terms. Lest this seem large, note that the trough to peak increase in the VAT deregistration rate was only 2½percentage points in Southern regions. This suggests that a little under one fifth of the rise in the court actions rate in Southern regions can be attributed to the rise in small business failure.

2.

3. 4.

5.

6.

7.

8.

55

9.

An alternative specification with very similar implications returns the lagged dependent variable but drops the lagged house price change. This fits a little worse. It was used to generate the Table 4 forecasts reported in the Housing Finance paper which explains why the scenarios reported below differ marginally.

10.

Note that DSS mortgage payments do not include repayment of arrears or of interest on previous arrears or on additional loans collateralized on housing. This explains why even when DSS mortgage payments are being made, mortgage arrears levels can continue to drift upwards. Though these data are not available annually, they are based on a larger and probably more consistently representative sample than the annual Family Expenditure Survey, from which they show systematic divergencies. Ideally, the adjustment would use the distribution of mortgages in each year in each region to count the fraction of mortgages under the tax relief ceiling. We do not have such data even at the national level. Instead, we take the distribution of national mortgage advances as a proxy for the distribution of mortgages, at least as far as the upper tail is concerned. It is used to fit a relationship between the average mortgage advance and the fraction under the tax relief ceiling. This relationship is then applied at the regional level, given data on the average mortgage in each region, but subject to the constraint that sit cannot exceed unity.

11.

12.

56

BIBLIOGRAPHY

Allen, C and A Milne, (1994),”Mismatch in the Housing Market”, Urban Studies, 31, 14511463. Breedon, F J and M A Joyce, (1992), “House Prices, Arrears and Repossessions: A Three Equation Model for the UK”, Bank of England Quarterly Bulletin, May, 173-179. Brookes, M, M Dicks and M Pradhan, (1994), “An Empirical Model of Mortgage Arrears and Repossessions”, Economic Modelling, 11, 2, 134-144. Dale, Peter, (1995), “Future Mortgage Default”, Housing Finance, 25, February, 14-20. Ford, J, “Problematic Home Ownership”, Joseph Rowntree Foundation, September, 1994.

Ford, J, E Kempson and M Wilson, (1995), QTR Mortgage Arrears and Possessions: Perspectives from Borrowers, Lenders and the Courts, HMSO: London. Hendry, D F, (1995), Dynamic Econometrics, Oxford University Press. Kau, J B, D C Keenan, W J Muller and J F Epperson, (1992), “A Generalized Valuation Model for Fixed-Rate Residential Mortgages”, Journal of Money, Credit and Banking, 24(3), August, pp.279-299. Muellbauer, J, (1996), “Measurement Biases in UK Mortgage Arrears Data”, ms., Nuffield College. Muellbauer, J and A Murphy, (1996), “Booms and Busts in the UK Housing Market”, forthcoming, Economic Journal. Ncube, M and S E Satchell, (1994), “Modelling UK Mortgage Defaults Using a Hazard Approach Based on American Options”, Department of Applied Economics, University of Cambridge Working Paper, No. 9408. Stephens, Mark, (1996), “Institutional Responses to the UK Housing Market Recession”, Urban Studies, 33, 337-351. Vandell, K D, (1995), “How Ruthless is Mortgage Default? A Review and Synthesis of the Evidence”, Journal of Housing Research, vol 6, no 2, pp.245-264.

57

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close