Extensions could provide borrowers some cover in the short-term as well as give lenders time to work out troubled credits and selectively prune their CRE portfolios through strategic sales in the secondary market.
Yeah that shit only works if you have enough assets to “prune” them and everyone else isn’t going to be selling on the secondary market at the same time cause the primary market is going pop.
The random forest model, which estimates the life of each loan in years, had an r-squared of 0.65 and a mean-squared error of 53 years. Given that the test set was roughly 31,900 loans, we think a mean squared error of only 53 is a very favorable result
Bruh, that’s not how this works. Whether or not 53 years of mean square error is acceptable depends entirely on how it affects your estimates for maturity. And that has to be calculated by propagating the error through the formulas until you get an error on credit. It’s not a difficult calculation, I was literally taught this in high-school. And if you can use random forest modles then surely you can propagate the error.
The distributions of the actual and estimated data sets looked similar as well, another encouraging sign.
No it isn’t. You are assuming that loans without maturity should have the same distribution on the time they are paid out as loans with maturity. Why would this be the case? If both types of loans had the same pay out date either way, then why would comercial actors take the one with no maturity?
Hmm, looking at the article again, it may be that most of these mortgages actually have a maturity date but it is just missing from the records (for some reason). I am honestly not knowledgeable enough about the CRE market to know which is the case. 60% of records are incomplete, or 60% of loans don’t have maturity (pr something in between).
Yeah that shit only works if you have enough assets to “prune” them and everyone else isn’t going to be selling on the secondary market at the same time cause the primary market is going pop.
Bruh, that’s not how this works. Whether or not 53 years of mean square error is acceptable depends entirely on how it affects your estimates for maturity. And that has to be calculated by propagating the error through the formulas until you get an error on credit. It’s not a difficult calculation, I was literally taught this in high-school. And if you can use random forest modles then surely you can propagate the error.
No it isn’t. You are assuming that loans without maturity should have the same distribution on the time they are paid out as loans with maturity. Why would this be the case? If both types of loans had the same pay out date either way, then why would comercial actors take the one with no maturity?
yeah the analysis there is pretty questionable
Hmm, looking at the article again, it may be that most of these mortgages actually have a maturity date but it is just missing from the records (for some reason). I am honestly not knowledgeable enough about the CRE market to know which is the case. 60% of records are incomplete, or 60% of loans don’t have maturity (pr something in between).