Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

IMF researchers: digital footprint yields better credit assessment

blogs.imf.org

Post reply on HN
Print view View original

111–120 of 165 posts

Re: IMF researchers: digital footprint yields better credit assessment

#111

Post by JorgeGT »

alfiedotwtf wrote: post #106
Animats wrote: post #97

Would information about CEO behavior be useful to investors? That's worth considering. Browsing history? Phone call patterns? It's being worked on at the Harvard Business School.[1] "We measure the behavior of 1,114 CEOs in six countries parsing granular CEO diary data through an unsupervised machine learning algorithm. The algorithm uncovers two distinct behavioral types: “leaders” and “managers”. Leaders focus on m…

What will be measured, will be gamed

Goodhart's law: when a measure becomes a target, it ceases to be a good measure

Re: IMF researchers: digital footprint yields better credit assessment

#112

Post by zaptheimpaler »

thegrimmest wrote: post #42
throwaway189262 wrote: post #32

Earlier quoted context omitted.

More accurate isn't always good because creditworthiness is all about stereotyping people. Consider this example: lesbians are statistically more likely to divorce. The ML models decides that lesbians are a higher mortgage risk and raises their interest rates. It does this indirectly by raising interest rates on married couples with different last names and owners of Subaru vehicles. In another, the model sees that p…

A lot to unpack here, let me try. Are machine learning algorythms capable of stereotyping people? How accurate does a stereotype have to be before it's a useful predictor? If these models lead to good data, then what's wrong with them? If people who follow the NBA do in fact default on their loans more, shouldn't they pay higher interest? I agree that the healthy should not be compelled to subsidize the costs of heal…

Have you considered the logical endpoint of your line of reasoning?

Basically, loans should only be given to those who are 99% likely to pay them back. Anyone who is at risk of default, say because they lost their job should be denied, making a permanent descent into poverty and misery even more likely.

Health insurance should only be given to those who are already healthy. Anyone who is sick enough to significantly benefit from insurance should be rejected. Even if they have an easily treatable condition that costs $50 in medicine a month, they should be rejected, guaranteeing their health gets worse.

In a world that worked so hard over centuries to create modern healthcare, we should only use it for people who barely need it. We may have literally figured out how to transplant organs and save lives, but nah its too expensive so why bother?

In your world, the slightest shock in someone's life would mean all safety nets get pulled, guaranteeing they will never recover. Then that data gets fed back to an ML model, confirming that yes indeed, people who need money or healthcare never do well anyways, so why bother with a lost cause?

Anyways, your worldview is fucked. Hope you think much more deeply about it. Or just grow old in the world you wish for, maybe one day you will need surgery, insurance will reject you, banks will not loan you the $200k it costs, and you will die for the sake of the algorithms and profit maximization.

Re: IMF researchers: digital footprint yields better credit assessment

#113

Post by nerdponx »

visarga wrote: post #95

> the type of browser and hardware used to access the internet, the history of online searches and purchases ... once powered by artificial intelligence and machine learning, these alternative data sources are often superior than traditional credit assessment methods, and can advance financial inclusion, by, for example, enabling more credit to informal workers and households and firms in rural areas Here is one of t…

Well said. I would also add that computing such a score requires the same kind of intensive surveillance that many people here on HN (including me) are morally opposed to.

Re: IMF researchers: digital footprint yields better credit assessment

#114

Post by dTal »

insert_coin wrote: post #48
dTal wrote: post #41

Earlier quoted context omitted.

And how do you propose we solve it? If there's a systematic financial disadvantage experienced by black people, possibly as a result of racism, then continuing to financially disenfranchise them will hardly help, will it? Access to liquid capital is pretty critical if you want to make money, isn't it? Judgement is notoriously self-fulfilling. The best way to keep someone poor is to treat them like a poor person. The…

And how did 'give everyone a home loan never mind if they can't afford it' worked out in 2008? Sometimes I do wonder if people who present the argument you just presented realize that that is exactly how we've been doing things, and it just keep making the problem bigger. Maybe I'm too much of a romantic, but I still harbor hope at some point they will realize it and begin to contemplate trying something different.

We're not talking about giving loans to people who can't afford it. We're talking about giving loans to black people.

Nobody is arguing against fairly assessing an individual person's ability to pay off a loan - the issue at question is whether it's okay to use race as a statistical proxy for that.

Re: IMF researchers: digital footprint yields better credit assessment

#115

Post by random5634 »

dredmorbius wrote: post #67
pvg wrote: post #66

Earlier quoted context omitted.

Let's wrap up here but, again, there is a totally trivial remedy for an article being "evil". Just don't post it to HN.

Exposing the evil is the utility here. Letting this proposal float unobserved is the greater evil. Sunshine ... isn't always a great disinfectant, but it is in this case.

Good lord. The logic twists here are crazy.

Anyways - in developing countries models along these lines are already being used - there are no or limited credit scores and identity issues can be big. So people are already getting financing as a result of this evil - usually microfinance

Re: IMF researchers: digital footprint yields better credit assessment

#117

Post by cesarb »

hntrader wrote: post #36
drtillberg wrote: post #28

Earlier quoted context omitted.

This. Slippery slope to denying credit based on a consumer's decision to avoid using Facebook or to search using DuckDuckGo.

DDG is more popular in the tech-literate crowd, most of whom are earning large salaries and therefore we'd expect that DDG usage should be positively (not negatively) correlated with credit worthiness

You are missing the point: DuckDuckGo is famous for its privacy stance, therefore one would expect that one's use of DuckDuckGo would not appear as "DuckDuckGo usage", but instead as "no search engine usage". The end result would be similar to what seems to happen in the USA (if one believes the anecdotes found in the Internet): not having a credit card means one is rejected when doing things like buying or even renting a house or a car, even though that has nothing to do with credit cards.

Re: IMF researchers: digital footprint yields better credit assessment

#118

Post by zajio1am »

aneemzic wrote: post #63
hntrader wrote: post #46

Earlier quoted context omitted.

We must disambiguate between political objectives and the practice of credit risk modelling. Credit risk, f(X), is an unknown population function that needs to be estimated using observed data X. If including tallness into X improves our estimate of f(X), then we've gotten a better model. You've asserted that X should only contain an individual's past actions instead of their inherent traits such as tallness. This ma…

I don't agree that this has anything to do with political objectives. It's a question of ethics. My domain knowledge is irrelevant, the ML Modelling aspect is irrelevant. The discussion was specifically around whether it's ok to include inherent traits when determining the credit worthiness of an individual. If you think it is, that's fine. We might as well just taking the same approach to crime, and start locking in…

> We might as well just taking the same approach to crime, and start locking individuals up or not extending job offers,

The main difference is that people have right to trial and to be considered innocent until proven guilty. But there is no 'right to credit'. Credit is fundamentally two-party contract.

Also there is shared limit to risk, forcing creditors to take more risk with some people means they may not take that risk in other cases (not giving credit to someone who would be marked lower risk with more informed decision) or forcing them to raise credit cost to everybody.

Re: IMF researchers: digital footprint yields better credit assessment

#119

Post by pmichaud »

thegrimmest wrote: post #44
pmichaud wrote: post #25

Earlier quoted context omitted.

When I have this conversation with myself, here's what I come up with: Shouldn't we be allowed to everything into account when determining credit worthiness? Even if it's socially uncomfortable? We get locally more accurate pricing / risk assessment that way, but we also create feedback loops that we as a society have decided we want to not contribute to. The cost for this is the pricing inaccuracy which we collectiv…

Well, I personally haven't consented to socializing the cost of anything. I'm not sure what the ethical grounds are for imposing this on me, and in so doing dilluting available accuracy of data about people. These feedback loops you're referring to, what order are you proposing for your chicken and your egg? What's causing what in your view?

I think the deal with feedback loops is that the elements of the loop are mutually causal, which is what makes them feedback loops at all.

Which one of the causes was first in history seems academic and immaterial if we're at t>0 and wanting to devise a strategy to slow it down or stop it.

Re: IMF researchers: digital footprint yields better credit assessment

#120

Post by zajio1am »

visarga wrote: post #95

> the type of browser and hardware used to access the internet, the history of online searches and purchases ... once powered by artificial intelligence and machine learning, these alternative data sources are often superior than traditional credit assessment methods, and can advance financial inclusion, by, for example, enabling more credit to informal workers and households and firms in rural areas Here is one of t…

> Maybe the model is biased, how do we know?

There are multiple mutually incompatible metrics of fairness, being unbiased in one makes it biased in others. See e.g. https://www.chrisstucchio.com/pubs/slides/crunchconf_2018/sl...

Post reply on HN