Forgive what might be an obvious question, but though I see there is a concept of trying to determine the risk of a loan, it isn't clear to me how that is actually done.
What is used as a source of identity of the borrower? IE, how do you punish bad borrowers on future loans? Or does that fall outside of this?
Credit-risk assessment is indeed the most brittle aspect of what I'm building.
In short -- borrowers have to get a cryptographically signed attestation from what's called a Risk Assessment Attestor in order to request a loan in Dharma. RAAs use whatever means they have to assess a borrower's identity and creditworthiness -- be that through social media logins, uploaded identification documents, authenticated phone numbers, their loan history both on and off chain, etc. They are compensated for this by a fee that is allotted to them in the loan contract. Ostensibly, borrowers are deterred from defaulting on their loans insofar as their future creditworthiness on the platform will be marred.
For the time being, Dharma Labs is the sole RAA of the protocol, but, as time goes on, we hope to build mechanisms for allowing other trusted third parties to enter the ecosystem as RAAs.
Risk assessment is literally the entire product is it not? Loans are not a technical problem to solve, and there's no inherent usefulness in it being a smart contract. A guy in a office is not a risk assessment strategy.
Not necessarily. I think a great example of something that smart contracts enable in a lending scheme is built-in secondary market functionality. A lot of marketplace lending platforms have built in proprietary secondary markets, but none of them are anywhere near as liquid as crypto markets are in general, and the vast majority are not interoperable, to my knowledge.
Your ownership in a loan on Dharma is denominated by a cryptographic token like Bitcoin, Ether, or any other, meaning its just as easy to trade your stake in a loan as it is to trade a cryptocurrency.
Assets built on open protocols are fundamentally easier to build products and technologies around -- insofar as they lower the barrier to entry for developing financial applications, their openness is extremely valuable.
> whatever means they have to assess a borrower's identity [...] uploaded identification documents
Whee, racial profiling!
> deterred from defaulting on their loans insofar as their future creditworthiness on the platform will be marred
Oooh, scary :-) You're not even threatening to half-assedly try to involve authorities if someone just runs with the money?
I think this is a very interesting engineering feat, but it does seem to suffer from the basic problem that there is no real connection between the real world and this fantasy world called "the blockchain".
Also, a small comment on the README:
> submitting to the contract any subset of the investors' bids whose total cash amount is equal to the desired principal
If you really mean "equal" here, you are looking at the NP-complete subset sum problem in general. Also, there might just not be solutions at all. Maybe you don't actually mean "equal", it should become more practical if you allowed some deviations from the borrower's initially specified amount.
I think the idea is that RAAs can bridge the reality divide by tying the pseudo-anonymous contract identity to a real life identity. You can, for example, require the user upload a video showing them, any government IDs, their house number, their last utility bills, pay stubs, etc. You might require a mobile app that could look at ongoing phone telemetry/contact info to see someone has a network and is regularly going to a job. The goal isn't mathematical certainty, just to make the effort/cost of theft < the value of the loan.
Some startups do similar things to this (Tala) to lend in countries without credit bureaus and IIRC they can achieve loan repayment rates >= 'prime' US borrowers.
Exactly. Tala's a great analogy insofar as developing world microfinance faces a lot of same challenges Dharma will likely encounter -- namely the lack the of legal recourse for defaults.
I heard a guy talking on Bloomberg radio this morning about this in regards to emerging markets bond investing. The interviewer was commenting that the legal system didn't provide good recourse in the event of default. The guy said like you that the credit analyst would need to focus more on the likelihood of default than the recoverability of the collateral.
It reminded me of my time working on credit models for subprime mortgages. We forecasted 2 variables: frequency and severity of default. The total loss prediction was the product of the two. If your severity is always 100% at least it makes your model simpler.
What is used as a source of identity of the borrower? IE, how do you punish bad borrowers on future loans? Or does that fall outside of this?