How Basel-Stadt wants to prevent over-indebtedness with a tool from the BFH

Thanks to digitalisation, data from existing administrative registers can increasingly be used in a fruitful way to monitor social issues such as poverty or debt. The potential of algorithm-based pattern recognition is also increasingly becoming the focus of applied research. This can be used to make forecasts for the early detection of critical life situations. Against this backdrop, the Bern University of Applied Sciences focuses on a digital transformation that places people at the centre in the strategic topic area of “Human Digital Transformation”. Based on this, we are developing the foundations for a data-based monitoring system for the canton of Basel-Stadt in order to advise people in precarious situations and to prevent over-indebtedness.

The data from the tax administration is an interesting starting point because it shows the financial situation of the entire population and provides comprehensive information about income and assets. Because debts can be claimed as deductions on tax returns, it can be assumed that tax data also provide extensive insight into the debt situation of individual households and the population as a whole. Tax declarations contain a list of debts with the amount owed, interest and creditors. Mortgage debts, personal loans and small loans are well recorded. In addition, a distinction is made between private and business debts (including the associated interest). However, it is also known that many people in poverty with high debts do not declare them conclusively to the tax authorities. This sets initial limits to a tax data-based debt monitoring system. Nevertheless, we think that it is already possible with the help of tax data to identify risky forms of debt that can be used in the sense of an early warning system. This paper examines the first steps of such an approach.

In order to develop a data-based debt monitoring system, we rely on tax data for the years 2016-2019. This data is linked to the population register and the social benefits data of the canton (premium reductions, social assistance, family allowances). This makes it possible to determine the household composition and the connection to the social benefit system. For the following evaluations, the data are limited to the population of working age (incl. children). They include data for 117,658 persons.

Private debt and various forms of risky debt, 2019

230501 Diagramme Tabellen online9

Indicators for the early detection of over-indebtedness

When does a debt become problematic? This is the central question. Debts are not problematic in themselves, but they become a burden when they cannot be paid and over-indebtedness occurs. From a logic of early detection, it would be important to identify critical forms of indebtedness that point to cases from which over-indebtedness may result. Can we measure risky forms of indebtedness? The preceding table shows a series of evaluations of the existing tax data on the debt situation of households in Basel-Stadt. In the following, we explain whether and to what extent the evaluations mentioned are suitable as early indicators of possible over-indebtedness.

Private debt

The first two indicators show how many people declare private debts in their tax returns (indicator 1) and how many of them do not own real estate (indicator 2). With the latter, it can be ruled out that the private debts are mortgage debts. Overall, 37% of persons of working age live in households with private debts. However, these consist largely of mortgage debt. Indicator 1 is therefore not suitable as a risk indicator. Significantly fewer people have private debts without mortgages – namely 7.2%. In order to exclude trivial debts, we additionally set a limit for a relevant debt amount at CHF 10,000. However, this indicator also seems to us to be of little significance because the income situation and thus the possibility of paying off debts is disregarded.

Risky debt

The indicators in the area of risky debt seem more interesting to us. Here, the income situation is included in a differentiated way by comparing the disposable income with the total amount of debt (indicator 3) and with the interest on debt (indicator 4). The disposable income refers to the income that remains for a household after deducting fixed costs (especially rent, health insurance premiums) for a minimum subsistence level according to social assistance. accordingly, 4.2% of people in Basel-City have debts that are higher than their annual disposable income, and for 1.6% of people the interest on debt is even higher than their annual disposable income. As the figure below shows, this mainly affects people with low incomes – but not exclusively. From the middle income bracket onwards, there are hardly any people who have trouble paying their debt interest.

Risky debt interest (excluding mortgage interest) by income class

230501 Diagramme Tabellen online10

Source: Linked tax data of the Canton of Basel-Stadt, 2019, calculations BFH 2023

The income classes (1st-10th decile) are formed according to a so-called decile division. This means that all households are sorted according to their household income and grouped into ten equally sized groups. The first decile comprises the lowest-income group in the canton, the tenth decile the highest-income group.

Mortgage interest rates

Other risk indicators ask whether there could be problems paying mortgage interest. Over many years, mortgage interest rates have fallen steadily. For example, the average of fixed mortgages over ten years was still 2.75% in June 2013 and just under 1% in August 2019. However, inflation and the National Bank’s interest rate decisions in response to it led to a rapid rise in mortgage interest rates for long-term mortgages in 2022. Within a few months, they more than doubled. The peak was 3.9% in October 2022. Owners with expiring mortgages may thus have to renew mortgages at conditions that exceed their financial circumstances.

Following this logic, we have formed three indicators that test whether the maximum disposable income for a household is sufficient to pay mortgage interest rates of 1%, 2.5% or even 5%. We offset these hypothetical interest rates against the mortgage debt recorded. According to our calculations, there are few households that would have trouble paying interest in an interest rate environment of 1% to 2.5%. At mortgage interest rates of 5%, however, 6.5% of owners and 1.7% of the working population would no longer have sufficient declared income to pay the instalments on their mortgages. As can be seen from the following figure for mortgage interest rates of 2.5%, people with low incomes would have more difficulties, but not exclusively.

Risky mortgage interest (2.5%) by income class

230501 Diagramme Tabellen online11

Source: Linked tax data of the Canton of Basel-Stadt, 2019, calculations BFH 2023

Official tax estimates

The last indicator starts with the official tax assessment. If no tax return has been filed, the tax administration derives the assumed income from previous tax returns or from salary statements. However, if the income situation has recently changed, e.g. due to a job loss, taxes may be overestimated, resulting in a tax debt that is difficult to settle with current income. The figure below shows that official tax assessments occur in all income brackets, but are particularly common among people with low incomes. Finally, we define risky official tax assessments (indicator 8) as those that affect people in the two lowest income classes.

Official tax assessments by income class

230501 Diagramme Tabellen online12

Source: Linked tax data of the Canton of Basel-Stadt, 2019, calculations BFH 2023

Identifying neighbourhoods with the most debts and problematic mortgages

In a spatial analysis, the different types of risky debts are shown according to residential quarters. This makes it possible to target counselling services more specifically to the problems of those affected. The following figure shows the neighbourhoods that are conspicuous depending on the debt indicators. Risky debt interest from private debts without mortgages is found disproportionately in the Klybeck and Kleinhüningen neighbourhoods. Households with a particular risk of over-indebtedness due to rising mortgage interest rates are found especially in the Gotthelf neighbourhood.

Risky private debts (excluding mortgage interest) by municipality and neighbourhood

230501 Diagramme Tabellen online13

Source: Linked tax data of the Canton of Basel-Stadt, 2019, BFH 2023 calculations

Risky mortgage debt (with mortgage interest of 2.5%) by communes and quarters

230501 Diagramme Tabellen online14

Source: Linked tax data of the Canton of Basel-Stadt, 2019, calculations BFH 2023

Trials with algorithms for early detection of households at risk

Algorithms offer great potential for recognising complex relationships in large data sets. It was therefore examined whether a system for the early detection of households with a particular risk of over-indebtedness could be developed by analysing anonymised tax data (Basel-Stadt, 2016-2019). To this end, the following concrete steps were taken:

  1. Definition of a list of characteristics that can be relevant for recognising households with risky debts (household characteristics).
  2. Training of an algorithm (“random forest” model) with these household characteristics and anonymised tax data from the years 2016-2018 to identify patterns among households with risky private debt (indicator 3) and risky debt interest (indicator 4).
  3. Testing whether the trained algorithm can detect those households with risky debt in the 2019 tax data without being told which households have effectively declared debt in their taxes.

The experiment shows potential for using a data-driven system to detect households with debt. The algorithm detects 46 people with risky debt and 198 people with risky debt interest without having known that they have actually declared debt in their tax returns. Even in this comparatively simple arrangement, the system would offer a starting point for recognising affected households and contacting them, for example, with information about available counselling services.

However, the evaluation also shows the limitations of the experimental system with the available tax data. A large number of the persons effectively classified as risky cases are not recognised as such by the algorithm. The 198 persons with risky debt interest mentioned correspond specifically to only 10.3% of the effectively existing risk cases. The other 1,726 persons declared as risky cases in the 2019 data remain undetected by the algorithm. In addition, there are 34 and 126 persons, respectively, in whom the algorithm detects a risk of risky debts or debt interest, although these are not considered risk cases according to the defined indicators.

It is also surprising which household characteristics are recognised by the system as particularly important for the prediction of risky debts. Of above-average importance are income classes, neighbourhood, nationality, age groups and household size. Somewhat less important are the household composition, the year of observation, the employment status (employed, self-employed), the claiming of means-tested benefits (premium reductions or rent subsidies), real estate ownership, tumour, discretionary taxation, the size of the household, the number of household members, the number of household members and the number of household members
property, discretionary taxation, gender and social welfare receipt. The importance of the characteristics does not differ much from whether we want to predict risky private debt or risky interest on debt. Similar variables are important for both risk indicators.

Importance of characteristics for debt prediction

The dashed line shows the average forecast strength of the model to illustrate which indicators are of above-average and which are of below-average importance.

230501 Diagramme Tabellen online15

Source: Linked tax data of the Canton of Basel-Stadt, 2019, calculations BFH 2023

Further potential of data-based debt monitoring

The presented attempts to link existing data with a focus on tax data have revealed a promising potential for debt prevention and early detection. In a more comprehensive project, the indicators outlined here could be expanded to address, for example, risky debts of the self-employed. In addition, the experiments with the algorithms could be further developed into a practicable early detection system.

The evaluations of the debt indicators and the results of the early detection system could be greatly improved with more detailed and linked data. Currently, there is the limitation that private debts are not further broken down in the tax data. It is also known that many people in poverty with high debts do not declare them conclusively to the tax authorities. Since they do not pay taxes due to low financial resources, it seems to be unnecessary to comprehensively claim debts. The potential could be further improved, for example, by linking the tax data with the databases of the debt counselling centres.

However, a more in-depth discussion would be needed on the implementation of such a system in accordance with the rule of law. The debt monitoring system would have to be designed in such a way that personal rights, for example with regard to data protection, are not violated. For this, it would have to be determined which official body would operate the system. In addition, the concrete measures to be taken if a person is identified as being at risk in the system would have to be clarified.

The article was also published by the Christoph Merian Foundation.

Creative Commons Licence

AUTHOR: Oliver Hümbelin

Prof. Dr. Oliver Hümbelin is a lecturer at BFH Social Work and conducts research on the interrelationships between poverty and inequality, health and poverty, and the social welfare system in Switzerland.

AUTHOR: Lukas Hobi

Lukas Hobi is a doctoral student at BFH Social Work. His research focuses on poverty and acquisition as well as data analysis and visualisation.

Create PDF

Related Posts

None found

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *