Scoring for loans, or: the Matthew effect in finance
Last year, we moved to a lovely but not particularly well-off area in Frankfurt. If we applied for a loan, this means that we might have to pay higher interest rates. Why? Because banks use scoring technologies in order to determine the credit-worthiness of individuals. The data used for scoring include not only individual credit histories, but also data such as one’s postal code, which can be used as a proxy for socio-economic status. This raises serious issues of justice.
Sociologists Marion Foucarde and Kieran Healy have recently argued that in the US credit market scoring technologies, while having broadened access, exacerbate social stratification. In Germany, a court decided that bank clients do not have a right to receive information about the formula used by the largest scoring agency, because it is considered a trade secret.
This issue raises a plethora of normative questions. These would not matter so much if most individuals, most of the time, could get by without having to take out loans. But for large parts of the population of Western countries, especially for individuals from lower social strata, this is impossible, since labour income and welfare payments often do not suffice to cover essential costs. Given the ways in which financial services can be connected to existential crises and situations of duress, this topic deserves scrutiny from a normative perspective. Of course there are deeper questions behind it, the most obvious one being the degree of economic inequality and insecurity that a just society can admit in the first place. I will bracket it here, and focus directly on two questions about scoring technologies.
1) Is the use of scoring technologies as such justified? The standard answer is that scoring expands access to formal financial services, which can be a good thing, for example for low-income households who would otherwise have to rely on loan sharks. Banks have a legitimate interest in determining the credit-worthiness of loan applicants, and in order to do so cheaply, scoring seems a welcome innovation. The problem is, however, that scoring technologies use not only individual data, but also aggregative data that reflect group characteristics. These are obviously not true for each individual within the group. The danger of such statistical evaluations is that individuals who are already privileged (e.g. living in a rich area or having a “good” job) are treated better than individuals who are already disadvantaged. Also, advantaged individuals are usually better able, because of greater “financial literacy”, to get advice on how they need to behave in order to develop a good credit history, or on how to game the system (insofar as this is possible). The use of such data thus leads to a Matthew effect: the have’s profit, the have-not’s lose out.
There are thus normative reasons for and against the use of scoring technologies, and I have to admit that I don’t have a clear answer at the moment (one might need more empirical data to arrive at one). One possible solution might to reduce the overall dependence on profit-maximing banks, for example by having a banking system in which there are also public and co-operative banks. But this is, admittedly, more a circumvention of the problem than an answer to the question of whether scoring as such can be justified.
2) Is secrecy with regard to credit scores justified? Here, I think the answer must be a clear “no”. Financial products have become too important for the lives of many individuals to think that the property rights of private scoring companies (and hence their right to have trade secrets) would outweigh the interest citizens have in understanding the mechanisms behind them, and in seeing how their data are used for calculating their score. In addition, social scientists who explore social inequality have a legitimate interest in understanding these mechanisms in detail. It must be possible to have public debates about these issues. Right now, the only control mechanisms for scoring agencies seems to be the market mechanism, i.e. whether or not banks are willing to buy information from them. But one can think of all kinds of market failures in this area, from monopolies and quasi-monopolies to herding behaviour among banks.
One might object that without trade secrecy there would be no scoring agencies at all, and hence one could not use scoring technologies at all (note that this only matters if one’s answer to the first question is positive). But it seems simply wrong that transparent scoring mechanisms could not work. After all, there is patent law for protecting intellectual property, and in case this really doesn’t work, one might consider public subsidies for scoring agencies. The only objection I would be worried about would be a scenario in which transparency with regard to scoring agencies would reinforce stigmatization and social exclusion. But the problem is precisely that this seems to be already going on – behind closed doors. We cannot change it unless we open these doors.
Very interesting post, Lisa. Speaking a little away from competence, let me try to question three things here. The first regards the claim of a Matthew effect. You say initially that a well-functioning credit system benefits the less advantaged. If that is right, would it not be that if scoring systems helps improve the functioning of the financial system, there may be a net benefit to the less advantaged? Perhaps there are different types of costs and benefits here – better access to credit against being ‘profiled’ in a certain way – affecting individuals differently. But that leads me to ask whether you could say a little more on the *specifics* of how the system might disadvantage the disadvantaged?
The second is to ask you whether there is any epistemic issue involved in the practice? You say that banks have a legitimate interest in credit-worthiness. Perhaps collecting sufficient individual data is very costly or intrusive, or even unobtainable (worries you have voiced yourself in theorising about justice). In that case, would using aggregate data have any more plausibility? It strikes me that some cases of using aggregate data is straightforwardly outlot – e.g., racial profiling. But, could it be that the indicators you mention – such as post code – are not ‘aggregates’ that we have such strong reason to reject if they serve some legitimate end (as per the above)?
Third, you mention at the very end that the issues of stigmatisation and exclusion cut both ways. I am not sure here whether there might be some equivocation. I presume that these phenomena in a transparent policy would be like the interpersonal effect of, e.g., using food stamps in public places – that is, things that turn on the information being in the open. Those effects, it is usually argued, would be, at least, lessened by calculating and distributing opaquely (e.g., through direct bank transfers). Whatever the other problems, it sounds like the scoring techniques you mention are of the latter kind. They may have other ways in which they can cause stigmatisation and exclusion, but I am not sure whether it is of the same kind. (Not sure how much would turn on this point – perhaps the kinds involved here are worse. But it seems important to spell out what exactly is the kind of social effect in question in order to know how they compare.)
Andrew, thanks for giving me a chance to clarify these points.
1) You are right: a good credit system can (!) benefit the least advantaged. This is precisely why I'm torn about whether or not it should be used at all: there are also points on the positive side! The way in which there might be a Matthew effect is that if you have "bad" data, you will have to pay more for your loan, or maybe not get one at all. You might say that having an opportunity to be considered for credit is still better than not having a chance at all.
2) Yes, getting access to data otherwise is often difficult. Most of all, it is costly for banks. I am not so sure there is unobtainability – there are certainly things you cannot predict, but then these are a general problem for financial institutions, nothing in particular about scoring. Intrusiveness can be an issue, but, again, it seems to work in both ways, as banks also receive quite a lot of information about you in scoring.
3) Yes, scoring is not taking place in the open, so in that sense there is no stigmatization (the neighbors don't see it, as they would see the use of food stamps). But still, some individuals are treated worse than others for the simple fact of belonging to a different group, and it might make it more difficult for them to get out of this group (because of higher interest rates).
I found an interesting paper on a related issue yesterday: http://www.rmm-journal.de/downloads/abstracts/Abstract_Scholes.pdf. It's on statistical discrimination in hiring. Scholes distinguishes between cases (such as insurance) in which being treated as an instance of a risk class is unavoidable, and cases (such as in hiring) in which it is avoidable, so the question would be whether access to loans is more like the former or more like the latter. She holds that the problem of (legal forms of) statistical discrimination in hiring is that the data in question (her example is smoking), while statistically related to productivity, are not a directly related to the individual's ability to be productive, and that such practices put pressure onto individuals to change private habits or to be discriminated against. This seems also relevant for finance – although it is hard to tell which data enter into the scores, because of secrecy. So a thorough normative evaluation might not even be possible without knowing more details!
Lisa, thanks for the further thoughts. From what you say on 1 and 2, my sense is that you are right much turns on a complex empirical calculation. I suspect it also turns on your starting point. For instance, when you say ‘might have to pay more for your loan’, much would depend on against what baseline you are judging:
‘Worse than an ideal baseline’ (for which a theory of background justice is needed).
‘Worse than an individuated scoring system’ (which, connecting to 2, would require a debate about what is feasible).
‘Worse than without a scoring system’ (where the Matthew effect claim seems less plausible)
Perhaps a case for or against the idea needs to be carefully placed against one of these (or other possible) options.
What you say on point 3 seems to suggest that the issue is less stigmatisation and exclusion than the problem of ‘being profiled’ and how one feels about that. I think I share the intuition that the latter is a problem. But if that is the driving force, transparency does not seem a great response. It does not remove the profiling. It simply makes it more public. That suggests the transparency approach would be no better, and perhaps it would be worse because it adds the possibility of public stigmatisation and exclusion to it. If that is right, my hunch is that the case against transparency rests more on some of the long-term considerations – like the importance of public debate about it – but does it also seem to you that such a case does seem potentially in conflict with (some) concerns of individuals affected by the system?
Apologies, the last lines should have read '…the case for transparency rests…'
Andrew, you are right about the importance of data. I'm less interested in an "ideal baseline" in this context, I think the more relevant points of comparison are a) no credit at all (seems worse, in many cases) and b) an individuated system (which might be more expensive). In this non-ideal scenario, and as long as banks are profit-maximizing and hence under pressure to minimize costs, a scoring system might be the best we can get. But there could be a scenario in which non-profit-maximizing banks could maybe offer b). The question is whether such a bank would be feasible, and would be better than a scoring scenario, all things considered.
On your second point: the feeling of "being profiled" is going to be there whether or not there is public knowledge about the factors that enter the calculation. There might be specific cases in which revealing specific data points (e.g. which suburbs get how many positive or negative points) could be harmful, but what I'd be interest in is which data in general enter the scores, and that's where the long-term arguments apply.
Lisa, sorry, one more query: why do you think a not-for-profit (maximising) bank could/would be better positioned? They would also be subject to limitations in acquiring information, perhaps especially that which requires extra outlay. Have some of the places you have researched had any practice of it?
Yes, they also have to acquire information, but they might be able to „afford“ to do so on an individual basis, building a relationship of trust with clients and getting to know their situation. This is the model I have seen in one the banks I did research on. They are operating in contexts in which scoring was not developed, at least not at the time when they started there, and they have an explicit goal to support economic development in these countries. So it’s a special case. They are now confronted with the fact that they could also rely on scoring, in some countries at least, and as far as I know they are quite torn about whether to use it or not.
Thanks for the post, Lisa. The issues that you raise are interesting and ones that I've not thought about at all previously.
My comment is aimed at making more precise exactly your question. You mention that part of what gives these questions their interest is that citizens rely upon loans. It seems therefore that, if citizens had greater incomes such that they did not have to rely upon loans, then lots of the normative questions would fall out of the picture. That is, there looks to be much stronger case for scoring, say, is loans were optional and not relied upon. With this in mind, I'm tempted to say that what is principally objectionable is not to do with the intricacies of loans, but with the fact that citizens rely upon them. That is where we should be focusing our attention. You could respond to this in agreement but nonetheless maintain that there remains the following question: In light of the fact that citizens rely on these loans, is scoring justified? Is this the question that you have in mind?
Tom, yes, exactly. One might also wonder whether scoring is a good idea if loans were a purely optional item (such as one might wonder about the justice or injustice of things going on in certain practices do in their spare time, such as reading clubs or whatever you have). But loans are a part of many people’s lives, and this as such does not have to be a bad thing (it increases your possibilities if you have some flexibility in the temporal dynamics of earning and spending). What seems bad is that many people need to rely on loans not only for, say, major investments with a long term horizon, but also for financing consumption goods at the end of the month. This is what makes the question about scoring more than some unimportant side-track of social justice debates.
interesting discussion, looking from Banking eye; Bank need some kind of mechanism to judge the payback capability of a customer. Even if a bank will do this at individual level, how one can assure that the person who is working on a customer will not discriminate based on the out look of a person instead of payback capabilities.
Without having much knowledge about scoring system postal code might play lesser role than income of the person.
Yes, discrimination can also go on at an individual level. Of course, there are also other factors such as income, income of spouse, etc., that go into the bank's calculation. I don't know if they also go into the scoring values.