Modelling households’ financial vulnerability with consumer credit and mortgage renegotiations

In Italy, the financial stability risks stemming from households’ debt are limited. Italian households are less indebted than European ones, though their debt-to-disposable income ratio has grown significantly since the beginning of the last decade, reaching 61.3 per cent in 2017 (37.2 per cent in 2002).¹ While mortgage indebtedness has remained quite low by international standards (Figure 1a), consumer credit has strongly expanded in recent years (from −1.0 per cent in 2013 to 9.0 per cent in 2018; Figure 1b) and, as a share of disposable income, it has become very similar to that of other countries in the euro area (Magri et al., 2019).

Figure 1

Download asset Open asset

Despite the significant expansion in loans to households, the debt-service to income ratio (DSR) has not risen much and has remained at around 10 per cent (Figure A.1 in the Appendix A). In fact, the impact of higher debt on loan instalments has been mitigated in recent years by exceptionally low interest rates. Households have benefited from low rates also by renegotiating the terms of their mortgages, particularly in 2015 and 2016 (Figure 2a). Low rates and renegotiations have contributed to improving debt sustainability. The higher selectivity of banks in granting loans with respect to the pre-crisis levels has also contributed to the decrease in the new non-performing loan ratio, which has fallen from the peak of 2009 both for mortgages and for consumer loans (Figure 2b).

Figure 2

Download asset Open asset

This evidence suggests that a timely monitoring of the risks associated with the household sector requires a model that projects the evolution of financially vulnerable households, accounting for both the dynamics of consumer credit and the possibility of revising the contract terms through mortgage renegotiation.

Several indicators of financial vulnerability have been proposed in the related literature (eg, D’Alessio and Iezzi, 2013). A first indicator that has been greatly employed by central banks to assess households’ ability to meet their debt obligations is the debt service-to-income ratio (Bank of Canada, 2011; Bank of England, 2014; Bank of Italy, 2011; European Central Bank, 2004, among others). Some studies have been carried out to assess how debt delinquency varies for different threshold values (May and Tudela, 2005; Tiongson et al., 2009; Dey et al., 2008; Beer and Schurz, 2007; Karasulu, 2008). Vacca et al. (2013) confirmed that the risk of arrears increases at a 30 per cent threshold and for households whose income net of debt payments is below the relative poverty line. A second indicator that has been used to identify financial vulnerable households is the financial margin, defined as the net income available to households after debt payments and basic living expenses (Martin and Persson, 2006; Vatne, 2006; Zajączkowski and Żochowski, 2007; Albacete and Fessler, 2010; Ampudia et al., 2016; Bettocchi et al., 2018, among others). Michelangeli and Rampazzi (2016) show that the shares of households classified as vulnerable according to both indicators (the first based on the debt service-to-income ratio and the second based on the financial margin) are similar. However, the first indicator is much simpler to compute and does not suffer from arbitrariness in the definition of some of its components, such as basic living costs. Following Vacca et al. (2013) and Michelangeli and Rampazzi (2016), we define a household as vulnerable if its debt service-to-income ratio exceeds 30 per cent and its income is below the median of the population.

To account for consumer credit growth and for the possibility of renegotiating a mortgage, which could affect the debt service-to-income ratio, we propose an extension of the microsimulation model developed by Michelangeli and Pietrunti (2014). This latter model includes the dynamics of income and mortgages (without renegotiations), while the outstanding consumer credit is assumed to be constant over time (the growth rate equals zero). These assumptions were adequate overall in the early 2010s, when mortgage renegotiations were very rare and consumer credit was growing at a relatively slow pace and up from very low levels in Italy. The recent development in household loans now means we need to accurately model the evolution over time of consumer credit and mortgage renegotiations, whose combined effect on households’ vulnerability is ex-ante unknown. The growth in consumer loans may increase vulnerability because households would face a larger loan instalment; at the same time, renegotiations act in the opposite direction by reducing periodic loan repayments. Not accounting for these dynamics may lead to a bias of an unknown sign in the projection of households’ vulnerability.

Building on the information on households’ characteristics and loan types provided in the Bank of Italy’s Survey on Household Income and Wealth (SHIW), we allow the amount and the cost of consumer loans to change over time according to the higher-frequency macroeconomic data. This implies, for instance, that a rapid increase in consumer loans can be promptly taken into account in the microsimulation model. More in detail, the projection of consumer credit is achieved in three steps: estimation of household participation, forecast of the total amount of consumer credit and computation of the instalment paid by each household. While we use an empirical approach for the first two steps, we impose some structure on the third one, assuming a standard amortization scheme. For mortgage renegotiations, we introduce simple heuristics, according to which households revise their contract terms when they pay an interest rate that is significantly above current market rates. This becomes a share of households renegotiating their mortgage that is consistent with the evidence from the SHIW.

The backtest exercises, run over the periods 2012–14 and 2014–16, show that with the new features, the model is better able to replicate (out of sample) the drop in vulnerability observed in the survey data. In particular, adding the possibility of renegotiating the mortgage terms turns out to be crucial for reproducing the downward trend observed between 2012 and 2016. Second, the new model allows us to assess how a sustained growth in consumer credit would affect households’ debt sustainability. Third, more flexible stress tests can be run.²

Finally, we provide a brief analysis on how the share of vulnerable households and the debt at risk would change if we were to define all households with a DSR of above 30 per cent as vulnerable (ie, not only those with an income below the median of the population). While households with an income below the median are the ones at a higher risk of default, this broader definition is more in line with some international studies (see, for instance, Beer and Schurz, 2007; Djoudad, 2010; International Monetary Fund, 2011; Bankowska et al., 2015).

We contribute to the literature on microsimulation models that aim at evaluating the vulnerability of households. This strand of the literature exploits detailed microeconomic data to identify which type of households are more exposed to the risk of default. By combining microeconomic and macroeconomic data, these models provide updated measures of the risks for financial stability stemming from the household sector and may even contribute to forecasting or early warning indicators. The first papers on this topic were those of Martin and Persson (2006), Vatne (2006) and Zajączkowski and Żochowski (2007) for Sweden, Norway and Poland, respectively. Among recent papers evaluating households’ vulnerability under normal conditions and scenarios of stress, we find Djoudad (2010) for Canada, Albacete and Fessler (2010) and Albacete and Lindner (2013) for Austria, Galuščák et al. (2014) for the Czech Republic, Michelangeli and Pietrunti (2014) and Bettocchi et al. (2018) for Italy, and Ampudia et al. (2016) for several European countries. Like the previous works, our paper exploits survey data, which are subject to measurement error.³ To our knowledge, our paper is the first to develop a methodology for the projections of consumer credit and mortgage renegotiations. By integrating micro and macro data, we preserve heterogeneity while exploiting the higher frequency of the macro data.

Our paper also relates to the literature on consumer credit. In recent years, many central banks have begun to publish data on consumer credit aggregates to show the main trends of this kind of debt and the emerging risks (Magri et al., 2019; European Central Bank, 2017; Bank of Italy, 2018a; 2018b). Furthermore, microeconomic data from household surveys have been exploited to capture the heterogeneity in the access to consumer credit market. For instance, Magri (2002) uses the SHIW to show that participation in consumer credit is positively correlated with income; a similar result is obtained by del Río and Young (2006) who, by focusing on unsecured debt for the years 1995 and 2000, show that the increase in participation and in the amounts borrowed are explained by the growth in household income. Magri et al. (2011) use Eurostat’s EU-SILC survey for nine European countries in the period 2005–08 to highlight that, regardless of the broad heterogeneity in participation across countries, borrowers tend to have a medium-high income, which they believe could be associated with the preference of lenders for granting loans to safer borrowers. Several papers have provided evidence of the procyclicality of debt and tried to identify the drivers of both demand and supply (Kiyotaki and Moore, 1997; Nakajima and Ríos-Rull, 2014). A robust finding is that durable consumption expenditures is procyclical as well as consumer debt (Bertola et al., 2006; Iacoviello, 2008; for a review, see Claus and Claus, 2016). Our paper proposes a novel approach for replicating the dynamics of consumer credit over time, and, in particular, for capturing the heterogeneity in the groups of households that hold this type of debt.

The paper is organized as follows: Section two describes recent developments in the Italian credit market, distinguishing between consumer credit and mortgages. Section three provides a description of households’ financial vulnerability based on the SHIW. Section four presents the model of households’ financial vulnerability that includes consumer credit dynamics and mortgage renegotiations. Section five shows the backtest exercises and the results for the baseline scenario. Section six provides a robustness analysis, focusing on all households with a DSR above 30 per cent and Section seven concludes.

In this section, we describe how we built our projections of households’ vulnerability over time. Specifically, we present the approach employed to model both consumer credit dynamics and mortgage renegotiations.¹⁰ For the projections of income and mortgage instalments, we rely on the modelling approach described in Michelangeli and Pietrunti (2014).¹¹

With respect to consumer credit, we took a three-step approach. First, we modelled households’ participation in the consumer credit market; second, we projected households’ loan amounts, replicating the macro growth rate of consumer credit; and finally, we computed the instalments using a standard French amortization schedule. For mortgage renegotiations, we assume that households revise their contract terms if their mortgage rate is significantly above current market rates.

4.1 Modelling consumer credit

4.1.1 Participation in the consumer credit market

For each household i, participation in the consumer credit market in period t depends on participation in the previous period in both the consumer credit market and in the real estate market, on the income quartile, and on the purchases of consumer durables:

(1) $$\begin{gathered} {\displaystyle D_{cc}(i,t)={\alpha }_0+{\alpha }_1{D}_{cc}(i,t-1)+{\alpha }_2{D}_{imm}(i,t-1)+{\alpha }_3{D}_y(i,t)+¿ \\ +{\alpha }_4{D}_{durables}\left(i,t\right)+\epsilon \left(i,t\right)\mathrm{f}\mathrm{o}\mathrm{r}i=1,\dots ,I;t=1,\dots ,T} \end{gathered}$$

where $D_{cc}(i,t)$ is a dummy variable equal to one if household i has a consumer loan in year t and zero otherwise, $D_{imm}\left(i,t-1\right)$ is a dummy variable equal to one if household i has a mortgage loan in the previous period and zero otherwise.¹² is a vector of income quartile dummies. $D_{durables}\left(i,t\right)$ is a vector of household durable consumption dummies. We prefer to use a vector of dummies for the amount of durable consumption, rather than a continuous variable, to minimize the errors associated with the projections of this variable. We consider five intervals, which were chosen starting from its sample distribution. We thus have five different dummy variables, $D_{durables}\left(i,t\right)$ , each of which takes the value 1 if the continuous consumption falls in the respective interval and zero otherwise.¹³ We assume that the errors are normally distributed.

The regression was estimated using a linear regression model.¹⁴ The regression coefficients imply that households with higher income and belonging to the higher classes of durable consumption are more likely to have a consumer loan. Furthermore, having a consumer debt or a mortgage in the previous period is associated with a higher probability of participating in the consumer credit market in the current period (Table A.1). The process is thus characterized by high persistency, which could be due to true state dependence or to individual-specific permanent unobserved heterogeneity. For instance, an individual may participate in the consumer credit market because of some unobserved characteristics, such as the need for money, despite other observed characteristics. In our approach, we set all individual effects equal to zero. Richiardi and Poggi (2014) show that this approach is optimal when the goal is forecasting mean values and the model is linear.¹⁵

The coefficients for each regressor $(\widehat{{\mathrm{\alpha }}_0},\dots ,\widehat{{\mathrm{\alpha }}_4})$, as well as the mean $\widehat{\mathrm{\mu }}$ and the standard deviation $\widehat{\mathrm{\sigma }}$ of the error term, were then used to simulate each household’s participation in the consumer credit market. We ran 50 different simulations to have some variability in the estimated error term.¹⁶

The estimated probability of entering the consumer credit market for each household in each period $\widehat{D_{cc}}\left(i,t\right)$ is given by:

(2) ${\displaystyle \begin{array}{rl}\widehat{D_{cc}}(i,t)& =\widehat{{\alpha }_0}+\widehat{{\alpha }_1}{D}_{cc}(i,t-1)+\widehat{{\alpha }_2}{D}_{imm}(i,t-1)+\widehat{{\alpha }_3}\hat{D}(i,t)\\ & +\widehat{{\alpha }_4}\widehat{D_{durables}}(i,t)+\widehat{ϵ}(i,t)fori=1,\dots I;t=1,\dots ,T\end{array}}$

In our setup, $\widehat{D_{cc}}\left(i,t\right)$ always takes values of between zero and 1. However, for simulation purposes, we are not interested in a continuous probability, but in identifying which households participate in the consumer credit market and which households do not. In other words, we need a variable equal to one for participants and equal to zero for non-participants. Thus, for each year, we first ordered our households based on their probability of participation $\widehat{{\mathrm{D}}_{\mathrm{c}\mathrm{c}}}\left(i,t\right)$ ,, from the highest to the lowest. Then, we selected a threshold $\widehat{{\mathrm{D}}_{\mathrm{c}\mathrm{c}}}\left(i,t\right)$ , above which all households are assigned for participating in the consumer credit market. The threshold is chosen to capture recent trends in participation.¹⁷

The simulated value of the participation in the consumer credit market for household i is:

(3) ${\displaystyle D_{cc}^{simul}\left(i,t\right)=\{\begin{array}{l}1if\widehat{D_{cc}\left(i,t\right)}>\widehat{D_{cc}^{Threshold}\left(i,t\right)}\\ 0if\widehat{D_{cc}\left(i,t\right)}\le \widehat{D_{cc}^{Threshold}\left(i,t\right)}\end{array}\mathrm{f}\mathrm{o}\mathrm{r}i=1,..,I;t=1,..,T}$

4.1.2 Projection of the consumer loan amount

After having established whether a household participates in the consumer credit market, we needed to assign a value for the consumer loan amount. We assumed that changes in consumer credit depend on household income quartile, on household consumption of durables, and on the aggregate dynamics of consumer credit.

Therefore, for each household i that had already participated in the consumer credit market in the previous period, we computed the change in the consumer credit amount

${\displaystyle \mathrm{△}{L}_{cc}\left(i,t\right)=L_{cc}\left(i,t\right)-L_{cc}\left(i,t-1\right)}$

and then we ran the following regression:

(4) ${\displaystyle \mathrm{△}{L}_{cc}\left(i,t\right)={\alpha }_{1}t{c}_{cc}\left(t\right)+{\alpha }_2{D}_y\left(i,t\right)+{\alpha }_3{D}_{durables}\left(i,t\right)fori=1,..,I;t=1,..,T}$

where ${tc}_{cc}\left(t\right)$ is the growth rate of consumer loans to households in the Italian economy.

The results of equation (4) are reported in Table (A.2) in Appendix A. As expected, ${\mathrm{\alpha }}_1$ has a positive sign, reflecting the fact that a stronger aggregate credit growth translates into higher average household consumer credit. Households with higher consumer durable expenses, ie, those belonging to the fifth class, tend to have larger positive changes in consumer credit.

For each household i already participating in the consumer credit market, the estimated change in consumer credit at time t is computed using the coefficient of the regression (4):

(5) $$\begin{gathered} {\displaystyle \widehat{\mathrm{△}{L}_{cc}(i,t)}=\widehat{{\alpha }_0}+\widehat{{\alpha }_1}t{c}_{cc}(t)+\widehat{{\alpha }_2}\widehat{D_Y}(i,t)+\widehat{{\alpha }_2}\widehat{D_{durables}}(i,t) \\ fori=1,..I;t=1,..T} \end{gathered}$$

The projection of the consumer loan amount over time is given by the previous period loan amount plus the estimated change:

(6) $$\begin{gathered} {\displaystyle \widehat{L{}_{cc}^{simul}(i,t)}=L_{cc}(i,t-1)+\widehat{\mathrm{△}L{}_{cc}(i,t)}fort=1 \\ \widehat{L{}_{cc}^{simul}(i,t)}=L_{cc}^{simul}(i,t-1)+\widehat{\mathrm{△}L{}_{cc}(i,t)}fort\S gt;1 \\ andi=1,..,I} \end{gathered}$$

For the first period of the simulation, we have actual data on $L_{cc}\left(i,t\right)$ from the SHIW; in the following periods, ${\displaystyle \widehat{L_{cc}^{simul}\left(i,t-\text{1}\right)}}$ is derived from the model.

For households that are estimated to have participated in the consumer credit market at time (t+1) for the first time, we do not have the previous period’s consumer credit, thus we estimated the loan amount by means of a pseudo-panel. First, we created G groups of similar households based on their age class, job type, durable consumption class, and mortgage tenure. Second, we estimated equation (4) for the median group change in the consumer loan amount, $∆{\mathrm{L}}_{\mathrm{c}\mathrm{c}}\left(\mathrm{g},\mathrm{t}\right)$ . Third, a household i, that belongs to group g and enters the consumer credit market at time t, was assigned the loan amount estimated on the pseudo-panel regression:

(7) $$\begin{gathered} {\displaystyle \widehat{L{}_{cc}^{simul}(i,g,t)}=L_{cc}(g,t-1)+\widehat{\mathrm{△}L(g,t)}fort=1 \\ \widehat{L{}_{cc}^{simul}(i,g,t)}=L_{cc}^{simul}(i,g,t-1)+\widehat{\mathrm{△}L{}_{cc}(g,t)}fort\S gt;1 \\ andi=1,..,I;g=1,...,G} \end{gathered}$$

After having computed the loan amount for each household that participates in the consumer credit market, we constructed the total consumer debt in the simulated economy:

(8) ${\displaystyle T{L}_{cc}^{simul}\left(t\right)=\sum _{i=1}^I\widehat{L_{cc}^{simul}\left(i,t\right)}fort=1,\dots ,T\wedge i=1,..,I}$

Then, we calculated the growth rate of the consumer credit loan in our simulation and we compared it with the actual aggregated data. As our goal was to have similar growth rates, we introduced a multiplicative adjustment factor $adj\left(t\right)$ for each household loan amount, which then became equal to ${\displaystyle \mathrm{a}dj\left(t\right)\widehat{L_{cc}^{simul}\left(i,t\right)}}$ . For instance, in the 2016–18 simulation, it equals 1.08 for the first year and 1.09 for the second year. These adjustment factors are not high and appear reasonable, especially if we account for the fact that the survey data could have problems of misreporting and that households are not selected to be representative with respect to their consumer credit debt.

4.1.3 Annuitization of consumer loans

Once we have the projected consumer debt we can perform an annuitization to calculate (annual) instalments. Assuming that consumer debt is repaid by a French amortization schedule, we need to make assumptions on maturity and interest rates.

For households having already consumer debt in the survey data, we use the maturity and interest rates available in the survey. For the households having more than one personal loan, we calculate an average maturity and an average interest rate, and we aggregate all debts in a single one (only loans for household needs other than property purchase, credit card receivables, and bank overdrafts are considered).

The theoretical debt that would be consistent with the observed instalments, under the assumptions just mentioned (remember that in t=0 we observe both instalments and debt), is very close to the actual declared debt (Figure 8).

Figure 8

Download asset Open asset

Depending on whether the change in consumer credit ($∆L_{cc}\left(i,t\right)$) is negative or positive we have two different approaches. In the first case, $∆L_{cc}\left(i,t\right)\S amp;lt;0$, we assume (partial) reimbursement of the outstanding debt. In the second case, $∆L_{cc}\S amp;gt;0$, we assume that a new contract of consumer credit is signed, and its instalments are added to those of the old (already outstanding) debt. The new loan has the same maturity as the old one at origination and the same type of interest rate (fixed vs variable). We therefore assume that a household’s preferences for maturity and type of interest rate remain constant over time.

For those that do not already have consumer debt at t=0, but are projected to participate in the consumer credit market in the following years, we assume that they choose a variable interest rate, sign their contract at the current rate, and, based on the observed data, set the debt maturity at 5 years.¹⁸

In this section, we present the results obtained by simulating the model under different specifications. We define the model described in Michelangeli and Pietrunti (2014) as ‘old’ and the model augmented with both mortgage renegotiations and consumer credit dynamics as ‘new’. The macroeconomic inputs to the model are described in Appendix B.

5.1 Backtesting

To evaluate the model’s performance in terms of the accuracy of its projections, we carried out backtest exercises on two waves of the SHIW. Starting from either the 2012 or the 2014 wave, we present here the two-year out-of-sample predictions of the share of vulnerable households in the population and the share of total debt held by them (Figure 9).

Figure 9

Download asset Open asset

For the period 2012–14, characterized by positive economic growth, the low interest rates and the contraction in consumer credit contributed to the reduction in the share of vulnerable households. Instead, with respect to the debt at risk, the differences between the two models are negligible reflecting the fact that a contraction in the share of vulnerable households with consumer credit has a limited impact on the total debt at risk. For the period 2014–16, in a context of very low interest rates, accounting for the possibility of revising the contract terms through renegotiations is crucial for a better prediction of both indicators of vulnerability.

Overall, both backtesting exercises show that the ‘new model’ has superior out-of-sample predictions overall and is better equipped to capture the downward trend of vulnerability observed in the last few years.

5.2 Baseline simulation over the period 2016-19

Figure 10 shows the prediction of the share of vulnerable households (Panel a) and the share of total debt held by these households (Panel b) in 2019 starting from 2016, the most recent SHIW wave. According to all model specifications, household vulnerability is projected to increase. The ‘new model’ indicates that between 2016 and 2019, the percentage of vulnerable households may have increased from 1.6 to 2.3 per cent, while the share of debt at risk may have only changed slightly, from 10.4 to 12.0 per cent.²³

Figure 10

Download asset Open asset

In the first year of the simulation, the projections of the ‘old model’ indicate that the share of vulnerable households remained essentially stable as the increase in income growth, coupled with the low interest rate, was sufficient to compensate for the effects of the increase in the growth rate of mortgages. The ‘new model instead projects an increase in the share of vulnerable households, which is entirely driven by the high growth rate of consumer credit. In the second year of the simulation, a rise in interest rates, which affects both the loan payments of households holding a variable interest rate mortgage and those associated with new originations, and a positive credit growth (for both mortgages and consumer credit) contributed to the increase in the share of vulnerable households despite a positive income growth. The total effect is more pronounced in the ‘new model’, when consumer credit growth is also considered. In the last year of the simulation, 2019, higher interest rates and growth rate of consumer credit are expected to drive a further increase in the percentage of vulnerable households.

In terms of debt at risk, the differences between the ‘old’ and the ‘new’ models are negligible, since the negative effects on vulnerability associated with the expansion of consumer credit are compensated by the positive effects associated with mortgage renegotiations.

We have so far focused on households with a DSR above 30 per cent and an income below the median of the population. This choice is because highly indebted low-income households are the most fragile and display the highest default risks (Vacca et al., 2013).

However, as a robustness check and in line with other studies in the related literature (Djoudad R., 2010, among others), we also monitor highly indebted households with a high income. Specifically, we analyse the risks relating to households with a DSR above 30 per cent and an income above the median of the population. As shown in Table 4, even though this group of households represents a small share of the population, it holds about 20 per cent of the total household debt. Interestingly, the share of households without any debt is very high and similar for high-income households with a DSR ≤30 per cent and for other low-income households (respectively, 82 and 88 per cent for each group).

The importance of this group for financial stability analysis is shown by the probability of being in arrears by more than 90 days, which is about 4.1 per cent, whereas it is only 1.4 per cent for high-income low-DSR households. Moreover, the former group of households in arrears is considerably more indebted than all the other groups (Figure 11).

Figure 11

Download asset Open asset

We construct a liquidity index, equal to the ratio of liquid financial assets (deposits, certificates of deposits, repos and postal accounts) to loan instalments. This index measures the number of years during which a household could service total debt only with its liquidity. High- and low-income households with a DSR above 30 per cent have a similar liquidity index (Table 5), suggesting that all households with a DSR above 30 per cent emerge as particularly fragile since they can only cover their debt instalments for less than one year with their most liquid assets. Moreover, households with a DSR above 30 per cent are very similar also with respect to other stock indicators that are considered relevant for identifying households’ fragility, such as the debt-to-income ratio and the debt-to-assets ratio.

Figure 12 shows the baseline projections for the period 2016–19.²⁴ With this broader definition, the share of vulnerable households in the population in 2016 is about 1 percentage point higher (2.8 per cent) and the share of debt at risk is about 14 percentage points higher (24.3 per cent) than in the case where only households with a DSR above 30 per cent and an income below the median are considered. According to the ‘new model’, the share of vulnerable households is projected to have increased to 3.4 in 2019, in response to the significant consumer credit growth. The share of debt held by all vulnerable households instead would fall from 24.3 to 19.9 per cent between 2016 and 2019. In this case, the reduction would be greater than that estimated only for low-income vulnerable households. This is because the effect of mortgage renegotiations is more relevant for high-income highly indebted households.

Figure 12

Download asset Open asset

This paper presents a novel modelling approach for the evolution of consumer credit and mortgage renegotiations within a microsimulation model of households’ vulnerability. In particular, an extension of the model by Michelangeli and Pietrunti (2014) is provided to account for the recent increase in consumer credit and its progressive concentration among the high-income households, and to account for the wide recourse to mortgage renegotiations occurring in a context of very low interest rates.

These new features of the microsimulation model have proved to be crucial for carefully assessing the financial stability risks stemming from the household sector. The model with consumer credit and mortgage renegotiations outperforms Michelangeli and Pietrunti (2014) in capturing the trends of vulnerability over the last few years.

Both mortgage renegotiations and consumer credit affect vulnerability, though in opposite directions. By reducing loan instalments, mortgage renegotiations contribute to decreasing vulnerability. Consumer credit instead drives an increase in the share of vulnerable households, though with small effects on the overall debt at risk. A model accounting separately for these two forces can be particularly useful, in particular at a time when they impact household financial vulnerability in opposite directions and it is not possible to establish ex-ante the overall effect. The model indicates that between 2016 and 2019, the percentage of vulnerable households increased from 1.6 to 2.3 per cent, while the share of debt at risk only changed slightly, from 10.4 to 12.0 per cent. In terms of the number of vulnerable household heads, this model generates a larger share of vulnerable households than in Michelangeli and Pietrunti (2014), which is mostly driven by the high growth rate of consumer credit. Instead, in terms of debt at risk, the differences between the two models are negligible, since the negative effects on vulnerability associated with the expansion of consumer credit are compensated by the positive effects associated with mortgage renegotiations.

The model is very useful for stress test analysis and, specifically, for evaluating scenarios of stress for the interest rates of both mortgages and consumer credit. For instance, with respect to the baseline scenario, in a very adverse scenario for 2019, characterized by an increase of 200 basis points in the 3-month Euribor rate, in the 10-year IRS rate and in the consumer credit rate, combined with a decrease of 4 percentage points in the growth rate of nominal income, the share of vulnerable households would be higher by about 0.3 percentage points and their debt by about 2.2 percentage points (Appendix A). The results of this stress test suggest that the financial conditions of Italian households would remain overall quite sound even if more hostile events were to occur, since vulnerability indicators would always remain below the levels reached in 2012

The model has been built to evaluate the dynamics of those households that would have the biggest difficulties in meeting their debt payments if negative shocks occurred, ie, those with a DSR above 30 per cent and an income below the median of the population. Nevertheless, the model is also suitable for analysing the dynamics of vulnerability for all households with a DSR above 30 per cent, including those with an income above the median. The model’s projections indicate that between 2016 and 2019, the share of fragile households would increase, but the share of debt at risk would decrease, indicating that the effect of mortgage renegotiations is more significant for high-income highly indebted households.

1. 1
   Financial Stability Report

   1. N Albacete
   2. P Fessler

   (2010)

   Stress testing Austrian households, Financial Stability Report, Austrian Central Bank.

   • Google Scholar
2. 2
   Financial Stability Report

   1. N Albacete
   2. P Lindner

   (2013)

   Household Vulnerability in Austria - A Microeconomic Analysis Based on the Household Finance and Consumption Survey, Financial Stability Report, Austrian Central Bank.

   • Google Scholar
3. 3
   Financial fragility of Euro area households

   1. M Ampudia
   2. H van Vlokhoven
   3. D Żochowski

   (2016)

   Journal of Financial Stability 27:250–262.

   https://doi.org/10.1016/j.jfs.2016.02.003

   • Google Scholar
4. 4
   Financial stability review, December

   1. Bank of Canada

   (2011)

   Financial stability review, December.

   • Google Scholar
5. 5
   Quarterly Bulletin 2014 Q4

   1. Bank of England

   (2014)

   Quarterly Bulletin 2014 Q4.

   • Google Scholar
6. 6
   Financial Stability Report

   1. Bank of Italy

   (2011)

   Financial Stability Report, 2.

   • Google Scholar
7. 7
   Annual Report for 2017

   1. Bank of Italy

   (2018)

   Annual Report for 2017.

   • Google Scholar
8. 8
   Financial Stability Report 2

   1. Bank of Italy

   (2018)

   Financial Stability Report 2.

   • Google Scholar
9. 9
   Indicators to support monetary and financial stability analysis: data sources and statistical methodologies of

   1. K Bankowska
   2. P Lamarche
   3. G Osier
   4. S Pérez-Duarte

   (2015)

   Measuring household debt vulnerability in the euro area: Evidence from the Eurosystem Household Finances and Consumption Survey, Indicators to support monetary and financial stability analysis: data sources and statistical methodologies of, 39, IFC Bulletin, Irving Fisher Committee on Central Bank Statistics: Bank for International Settlements.

   • Google Scholar
10. 10
    Monetary Policy and the Economy, 2nd Quarter

    1. C Beer
    2. M Schurz

    (2007)

    58–79, Characteristics of Household Debt in Austria: Does Household Debt Pose a Threat to Financial Stability?, Monetary Policy and the Economy, 2nd Quarter.

    • Google Scholar
11. 11
    The Economics of Consumer Credit

    1. G Bertola
    2. R Disney
    3. C Grant

    (2006)

    Massachusetts, London, England: The MIT Press Cambridge.

    • Google Scholar
12. 12
    Assessing and predicting financial vulnerability of Italian households: a micro-macro approach

    1. A Bettocchi
    2. E Giarda
    3. C Moriconi
    4. F Orsini
    5. R Romeo

    (2018)

    Empirica 45:587–605.

    https://doi.org/10.1007/s10663-017-9378-2

    • Google Scholar
13. 13
    Measurement error in the bank of Italy’s survey of household income and wealth

    1. C Biancotti
    2. G D’Alessio
    3. A Neri

    (2008)

    Review of Income and Wealth 54:466–493.

    https://doi.org/10.1111/j.1475-4991.2008.00283.x

    • Google Scholar
14. 14
    Savings and wealth accumulation: measurement, influences and institutions

    1. E Claus
    2. I Claus

    (2016)

    A Collection of Reviews on Savings and Wealth Accumulation pp. 1–8.

    • Google Scholar
15. 15
    The determinants of unsecured borrowing: evidence from the BHPS

    1. A del Río
    2. G Young

    (2006)

    Applied Financial Economics 16:1119–1144.

    https://doi.org/10.1080/09603100500438791

    • Google Scholar
16. 16
    Bank of Canada Review

    1. S Dey
    2. R Djoudad
    3. Y Terajima

    (2008)

    45–54, A Tool for Assessing Financial Vulnerabilities in the Household Sector, Bank of Canada Review, Summer.

    • Google Scholar
17. 17
    The Bank of Canada’s analytic framework for assessing the vulnerability of the household sector

    1. R Djoudad

    (2010)

    Financial System Review pp. 57–62.

    • Google Scholar
18. 18
    Bank of Italy Occasional Paper

    1. G D’Alessio
    2. S Iezzi

    (2013)

    149, Household over-indebtedness: definition and measurement with Italian data, Bank of Italy Occasional Paper.

    • Google Scholar
19. 19
    Financial Stability Review

    1. European Central Bank

    (2004)

    Financial Stability Review, December 2004.

    • Google Scholar
20. 20
    Economic Bulletin, Issue 7 / 2017 – Boxes Recent trends in consumer credit in the euro

    1. European Central Bank

    (2017)

    • Google Scholar
21. 21
    Stress testing the private household sector using microdata

    1. K Galuščák
    2. P Hlaváč
    3. P Jakubík

    (2014)

    Czech National Bank.

    • Google Scholar
22. 22
    Household debt and income inequality, 1963–2003

    1. M Iacoviello

    (2008)

    Journal of Money, Credit and Banking 40:929–965.

    https://doi.org/10.1111/j.1538-4616.2008.00142.x

    • Google Scholar
23. 23
    United Kingdom: vulnerabilities of private sector balance sheets and risks to the financial sector technical notes

    1. International Monetary Fund

    (2011)

    United Kingdom: vulnerabilities of private sector balance sheets and risks to the financial sector technical notes, IMF Country Report, 11/229, July 2011.

    • Google Scholar
24. 24
    Stress Testing Household Debt in Korea

    1. M Karasulu

    (2008)

    Washington, DC: International Monetary Fund.

    • Google Scholar
25. 25
    Credit cycles

    1. N Kiyotaki
    2. J Moore

    (1997)

    Journal of political economy 105.2:211–248.

    • Google Scholar
26. 26
    Temi di Discussione (Working papers) 454

    1. S Magri

    (2002)

    Italian households’ debt: determinants of demand and supply, Temi di Discussione (Working papers) 454, Bank of Italy, Economic Research and International Relations Directorate.

    • Google Scholar
27. 27
    The expansion of consumer credit in Italy and in the Euro area: what are the drivers and the risks?

    1. S Magri
    2. V Michelangeli
    3. S Pastorelli
    4. R Pico

    (2019)

    SSRN Electronic Journal 500.

    https://doi.org/10.2139/ssrn.3435154

    • Google Scholar
28. 28
    Bank of Italy Occasional Papers

    1. S Magri
    2. R Pico
    3. C Rampazzi

    (2011)

    100, Which households use consumer credit in Europe?, Bank of Italy Occasional Papers, 10.2139/ssrn.1998749.

    • Google Scholar
29. 29
    Swedish households’ indebtedness and ability to pay: a household level study

    1. J Martin
    2. M Persson

    (2006)

    Sveriges Riksbank Econ. Rev 3:24–40.

    • Google Scholar
30. 30
    277, When Is Mortgage Indebtedness a Financial Burden to British Households? A Dynamic Probit Approach, Bank of England Working Paper, 10.2139/ssrn.872688

    1. O May
    2. M Tudela

    (2005)

    277, When Is Mortgage Indebtedness a Financial Burden to British Households? A Dynamic Probit Approach, Bank of England Working Paper, 10.2139/ssrn.872688.

    • Google Scholar
31. 31
    A Microsimulation Model to evaluate Italian Households’ Financial Vulnerability

    1. V Michelangeli
    2. M Pietrunti

    (2014)

    International Journal of Microsimulation 7:53–79.

    https://doi.org/10.34196/ijm.00107

    • Google Scholar
32. 32
    Bank of Italy Occasional Papers

    1. V Michelangeli
    2. C Rampazzi

    (2016)

    369, Indicators of financial vulnerability: a household level study, Bank of Italy Occasional Papers, 10.2139/ssrn.2934248.

    • Google Scholar
33. 33
    National Bureau of Economic Research, Working paper

    1. M Nakajima
    2. JV Ríos-Rull

    (2014)

    20617, Credit, bankruptcy, and aggregate fluctuations, National Bureau of Economic Research, Working paper.

    • Google Scholar
34. 34
    Imputing individual effects in dynamic Microsimulation models an application to household formation and labour market participation in Italy

    1. M Richiardi
    2. A Poggi

    (2014)

    International Journal of Microsimulation 7:3–39.

    • Google Scholar
35. 35
    The Crisis Hits Home: Stress-Testing Households in Europe and Central Asia

    1. E Tiongson
    2. N Sugawara
    3. V Sulla
    4. A Taylor
    5. A Gueorguieva
    6. V Levin
    7. K Subbarao

    (2009)

    Washington, DC: The World Bank.

    • Google Scholar
36. 36
    L’indebitamento e la vulnerabilit finanziaria delle famiglie nelle regioni italiane

    1. V Vacca
    2. D Coin
    3. AM Conti
    4. L Leva
    5. D Liberati
    6. E Manzoli
    7. D Marangoni
    8. S Mocetti
    9. G Saporito
    10. L Sironi

    (2013)

    Bank of Italy, Occasional Papers, 163.

    • Google Scholar
37. 37
    How large are the financial margins of Norwegian households? An analysis of micro data for the period 1987-2004?

    1. BH Vatne

    (2006)

    Norges Bank Eco B, Lxxvii 4:173–180.

    • Google Scholar
38. 38
    Proceedings of the IFC Conference on “Measuring the financial position of the household sector

    1. S Zajączkowski
    2. D Żochowski

    (2007)

    62–74, The distribution and dispersion of debt burden ratios among households in Poland and its implications for financial stability, Proceedings of the IFC Conference on “Measuring the financial position of the household sector, Basel.

    • Google Scholar

Author details

1. Carmela Aurora Attinà

   Directorate General for Economics, Statistics and Research, Bank of Italy, Rome, Italy

   Competing interests

   No competing interests reported

2. Francesco Franceschi

   Directorate General for Economics, Statistics and Research, Bank of Italy, Rome, Italy

   Competing interests

   No competing interests reported

3. Valentina Michelangeli

   Directorate General for Economics, Statistics and Research, Bank of Italy, Rome, Italy

   For correspondence

   valentina.michelangeli@bancaditalia.it

   Competing interests

   No competing interests reported