Apologies for adding another post, but I’m concerned about the conclusions drawn from the data presented and would like to raise the importance of benchmarking. Most of what I see presented, especially in the initial post, are some delinquency rates for a few batches of loans with no market context or benchmarking. mizzmonika noted the apples-to-oranges comparison and this merits further comment.
For a fixed income product, like an Upstart loan which is packaged into a securitized product, the key credit risk attributes for each batch of loans are the yield (determined by the price paid for the loans by the investor) and the expected default rate (estimated by proxy via the credit rating) over the lifetime (term) of the instrument. If the expected default rate is higher, then the yield should be higher (lower price) to compensate (as noted by many: mizzmonika, Saul, ar3ar3, FoolishJE, chang88).
Most importantly, the performance of these loans, and their associated securitizations, should be measured relative to a representative benchmark.
If anyone is going to make performance statements (like delinquency rate) about a batch of loans issued at time S with X credit rating yielding Y% over term T, then this performance needs to be referenced against a comparable set of loans, like the market averages, or competitor averages, for loans with similar issue date, rating, yield, and term (as well as other attributes specific to the securitization structure).
For example, why do I care if Upstart’s delinquency rate is 6% if the competitor’s rate is 10% for the same types of loans/securitization? This gets to the core value proposition of Upstart: potentially better loan performance than the existing market. But no one has presented data to address this core value proposition one way or the other. In fact, on page 5, KBRA states “While the Company’s model has not been tested through a full economic cycle, default and delinquency rates have not shown meaningful and sustained degradation since the onset of the COVID-19 pandemic.” (also noted by philiproth with caveats)
Can anyone show data that supports the idea that Upstart’s loans (and hence AI model), in an apples-to-apples comparison, are performing better/worse/differently in terms of delinquency (or better yet, realized yield after term) than some other group of comparable loans, ideally a market benchmark for this class of asset-backed securities?
I have not seen a benchmarked comparison presented, and hence the delinquency rates presented so far are almost meaningless. If I have missed something, please let me know, I am not claiming perfect or expert knowledge, just challenging what has been presented so far.
Additional Comments on Use of Data
It is unfortunate that delinquency rates are a lagging indicator - only time can tell how their loans are actually performing.
This is incorrect, delinquency rate is a leading indicator of loan performance. Loans will, by definition, go delinquent at a point in time before they are determined to be in default.
But I think you made a mistake in what I would call your “second derivative” research - ie digging through search numbers and all the other stuff. It was a fun “what if” thing to do, but a lot of people made serious decisions with their money based on it, including you, and it proved to be incorrect.
I agree here. We need to be very careful with statistics like online reviews, keyword search trends, and estimating total addressable market. Unless the underlying data can be shown to be somewhat representative, then these metrics are likely riddled with biases that we cannot estimate which leads to conclusions that are random guesses at best and completely misguided at worst.
Best of luck to all of us.