IMF working paper suggests using browsing history in credit score calculations

The working paper doesn't really contain any original research (at least on this topic); they cite a single paper by Tobias Berg, Valentin Burg, Ana Gombović, Manju Puri [3]. The IMF working paper/blog post concludes, on the basis of this paper, that a system "powered by artificial intelligence and machine learning" (ugh) can improve credit assessments using features such as "the type of browser and hardware used to access the internet".

I feel kind of depressed that this needs to be said, but...

1. There is no causal link between user-agent and loan repayment.

2. Attaching high-value signals such as credit scores to such easily spoofed data as user-agent and browsing history has some amazingly obvious problems that are never mentioned.

The level of technological and statistical incompetence on display here is rather astounding, and it's difficult for me to take any of this seriously even before the ethical questions begin.

This is part of an ongoing trend I've noticed: the quality of human capital in financial services seems to be in rapid decline. The cream of the crop was scraped off by huge tech companies on the industry side. On the academic side, all the top students are choosing CS PhD programs instead of econ/finance and top-tier faculty are leaving core econ/business/finance/etc. departments for specialized institutes/departments. And even within finance, already degraded, the under-60 talent that does remain isn't doing stuff like loan origination or credit risk assessment for obvious reasons.

I think it's time to start strictly regulating things like credit scores. If not for reasons of privacy/equality, at least merely because the entire industry seems to be incompetent and getting worse.

--

[1] https://blogs.imf.org/2020/12/17/what-is-really-new-in-finte...

[2] https://www.imf.org/en/Publications/WP/Issues/2020/08/07/Fin...

[3] https://academic.oup.com/rfs/article/33/7/2845/5568311

a) how do they acquire users' web history? Do they purchase it from ISP?

b) how do they know what ISP to ask?

c) how does the ISP know my web history? Last time I checked, almost all websites I visit use https, so the only thing they can see is domain, not /pathname/

d) how does it feel to poke your nose where it doesn't belong?

In financial analysis of a company, the number one item you tend to look at is "free cash flow" - the "cash available for the company to repay creditors or pay dividends and interest to investors." [1]

Why then do we not take income into account in credit worthiness? Why do I not periodically upload my W2 or tax return to the credit bureaus in order to show that yes, I'm credit-worthy because my income makes it possible to service my debt?

The entire idea of a credit score based on having had credit and paying past loans off is asinine. To start throwing nonsensical parameters like browsing history into the equation is second derivative asinine.

[1] https://www.investopedia.com/terms/f/freecashflow.asp