But in the biggest one that ever teaches the realities of home, economists Laura Blattner at Stanford University and Scott Nelson at the University of Chicago point out that the difference in creditworthiness between the small and large is not just biased, but that small and small groups with low income have little to do with their credit history.
This means that if this data is used to calculate the fee and the loan reward is used to predict repayment, then the statement may not be accurate. This lack of accuracy leads to inequality, not just bias.
The implications are complex: good algorithms will not solve the problem.
“It’s a very good result,” says Ashesh Rambachan, who studies mechanics and economics at Harvard University, but did not participate in the study. Prejudice and small-scale writing have been tedious for some time, but this is the first major experiment that focuses on lending to millions of real people.
Financial statistics override the amount of financial data, such as employment history, financial history, and purchasing habits, in the same number. In addition to making loans, loans are now being used to make many life-changing decisions, including insurance options, employment, and housing.
To find out why fewer people are being helped differently by lenders, Blattner and Nelson collected debt reports from an unknown 50 million in the US, and tied each of them to economic and financial factors taken from a business list, property to sell mortgages, and information about lenders who gave them credit.
One reason for this is that the first type of research is because these datasets are proprietary and are not publicly available to researchers. “We went to the credit bureau and we had to pay them a lot of money to do this,” Blattner said.
Lots of noise
He then experimented with various ways to show that interest rates were not only biased but also “noisy,” data analysis numbers that could not be used to accurately predict. Take a few loan applicants to get a 620 loan. In the bias, we can expect these scores to increase the applicant’s risk and for the most accurate score to be 625, e.g. Ideally, this bias can be calculated through other legitimate means, such as lowering the rate of approval for small jobs.
Source link
