Removing the Human Factor from DeFi Risk Managemen / Yaron Velner2 Velner

Transcript

foreign [Music] aspects of defy less about regulations more about actually how to make you know design platforms even more independent and self-executable and removing a human factor out of them in the most General sense so the different risk you know kind of Encompass many aspects in this talk and this is also my main expertise I'm only covering economic risk management and also economic risk management is a very

big topic so we'll even narrow it down to something which is kind of the most active risk management that is currently done in defy which is risk management of lending markets and stable coins where basically in the landing Market you give a loan to someone and your main hope and only hope is to eventually get the loan back when you want it of course everything is done in a permissionless manner so someone just you know give you a collateral in return to the loan this is the only guarantee for the user to eventually return the loan uh and and therefore uh the relevant

parameters to define a loan is mostly belong to Value which means how much you let a person borrow against this collateral for example let's say is two thousand dollars and I want to get to borrow us the say for example and I'm giving you one is collateral then obviously if you give me too much credit there at certain point is price might go down and I will just turn away without paying the debt and everything is permissionless here so basically your funds are lost or partially lost in the general case

uh yeah and right now these are done you know all these risk models and assessments are done mostly by humans okay So eventually you have some doubt that vote and approves them so maybe unlike a tradeify everything is completely transparent and at the end of the day and I think even in this conference there are at least like three talks about it uh how to actually do this risk management uh and you know people try to automate it and of course there's a lot of machinery and back end behind it but at the end of the day there is a human uh or a group of people uh that is sort of in charge of it and run these analysis

is daily and then as of now I'm also a part of this human factor so with B protocol which I founded we try to mitigate the human factor and kind of promote a more automated risk management but I'm also contributed to the risk Dao in the risk Dao we just give a risk analysis as a service to protocols which you know by definition services are done by human with human bias and everything and the way I see this area and you know like uh the way lending markets and stable

coins operate that are kind of in sharp contrast to what we see now in decentralized exchanges for example and previously I was CEO of kyber network library network was one of the earliest decentralized exchanges back in the day you know we did a lot of liquidity provision not not by fixed formulas but actually by on-chain algorithms and then at Southern State you know uni swap emerged even bancor before it uh and the star nut kind of became you know let's replace humans with a formula You know despite the formula might uh you know humans might over perform a formula but still it has a lot of advantages

you know even if it's less optimal could be potentially even less efficient there is a lot of Merit into having you know completely self-executed and permissionless protocols and yeah and risk management you know as kind of notorious history already from traditional Finance you know I think kind of they're all Bitcoin uh I don't know initiative or you know Bitcoin started because of the 2008 uh depth crisis which some argued you know was kind of fueled at least uh by a credit rating agencies uh

uh of course you know in in crypto it's even monitorious uh you know we all witnessed how three error how FTX you know all gave bad loans did very you know so with FTX it was outright fraud maybe but you know we are Genesis real or Celsius uh doing you know something which is at best not optimal risk management but also in traditional Finance you know we say this tweet about the London Metal Exchange you know fully regulated reputable traditional uh financial institute so the certain point you know they decided to freeze the liquidation

mechanism which is you know like at least conceptually uh not very different from you know FTX freezing liquidations to Alameda of course you know the intent was completely different uh you know they did it for a good cause but you know accidentally by chance or Not by chance you know it also helped one of the big whales there not to get liquidated and you know we can argue forever whether it was the right decision whether it's a a objective decision Fair decision uh and here what I'm trying to say is that you know we shouldn't necessarily optimize for the best decisions but you

know if I should be more about also transparency and fairness even if it is on the expense of efficiency by the way the same way that blockchain themselves are much less efficient than you know any Amazon cloud or cloud service that you can have out there okay so this is crypto but of course you know in in defy this would never happen right so the device fully transparent everything you know we know what is going on so yes certificate is transparent but right now most landing markets a really kind of on-chain Finance other than a decentralized finance or maybe I should say difference differently so

it's very decentralized very transparent but you know even in a decentralized setup you still have politics so this is a recent event uh where com where average community and we're discussed in whether they should phrase liquidations during the USD CD packing uh so it's important you know kind of to emphasize uh that you know other community is great specifically here Gauntlet are great in this case there's no evidence you know they had any hidden agenda or anything and it's you know eventually they didn't take this decision decision uh you know things could have turned out differently and maybe it was very

justifiable to take this decision but still it doesn't you know it does make the decision better but it doesn't change the concept uh that you know we have like actually most of D5 TV right now is actually subject to you know human or political decisions so yes what is it you know we are trying to promote and this is very much still a work in progress So eventually deciding on you know long long conditions is very much biased and you know I would even say rightfully so biased you know some domain expertise

and maybe some political agenda uh and and and typically this risk analysis you know embeds a kind of objective uh quantitative metrics which we all like to present uh but then there's a lot of Blackmagic and Voodoo about you know kind of the main specific knowledge and you know also again here I'm not arguing it's not justifiable it's very you know justifiable if you look at Luna case USD there are a lot of things there you couldn't have seen uh you know just by looking at the cold quantitative metrics but of course you know it's then

becoming sort of a political then okay maybe UST is outright you know along what about usdt there's a lot of human input there you try to assess you know like the risk risk position of a user whether it will get liquidated I know some teams have some knowledge about you know kind of degree or physical location of the user and can potent because of the transaction history for example and see you know like the time zones that is mostly active on and then uh you know you can try to optimize and maybe even fit into machine learning and kind of say you know users

from Korea might be a riskier and of course eventually you have to balance some risky world uh between you know like the risk The Landing Market take and the profit they generate so all of these decisions you know are very rightful decisions uh of course uh you know if you can optimize your risk management by kind of classifying users according to their I don't know ethnicity or origin you know it might be even maybe your duty as a risk manager but then you know kind of the question if it's really you know if you want to do something different in defy even on the expense of making it less optimal

so what we are trying to promote and this is still a work in progress uh we won't eventually you know and this is this will have to be gradual in this on our best effort uh basis but eventually we want smart contracts to make all these decisions okay rather than Dow votes or others uh and I I daily at the end of the day you know I don't know 10 years from now we want it to be all based on some quantitative uh objective metrics but of course you know the way there is still very long then as and as I mentioned you know a lot of the domain expertise so it's very justifiable to use it in

Risk analysis so in addition to this quantitative data we also initially allow for some confidence factor which is basically some kind of a magic number that's supposed to embed you know like all the human specific assumptions about this asset asset and then given you know these two inputs uh smart contact can just decide on along to Value ratio yeah so recently last week you know uh published like the first candidate for such a formula and I'm not sure you know this is actually the place to discuss uh all the

different parameters but I just want to kind of demonstrate that you know like these parameters are sort of a market objective parameters and here we have this magic number which give give a room to some user input to sorry to some human input in the process yeah so again this is very much still a work in progress and if you want to read you know read more about the formula it will give some references later but other than you know so so one kind of front uh is to extend the research about what kind of formulas could

support such activity but then another front and this is where we at the B protocol mostly focus is how to actually put this data on chain because right now a key indicators such as asset liquidity a market Dex liquidity Etc sorry asset volatility are not really available for example by chain link a oracles and down the road as you know the as the formulas are getting more complex additional input data will be needed so this is like our focus on B protocol and of course this part is not live yet and still work in progress

but even just by you know having this formula uh and and New York the kind of division eventually is to have Smart contracts control everything but just just by having the formula we can still uh we are still able to introduce some you know higher transparency level across different platforms uh so if you remember the formula kind of gets you know a market conditions plus a magic number and then outputs you know a loan to value ratio but we can kind of reverse the process right we can take existing loan to value ratios from learning Market we can take

the market condition and then output the magic number which is this C value here and basically the higher uh the magic number is it sort of means that the platform is taking lower risk so this is a very you know kind of brief research we didn't compound these three markets on usdc and and you could say that for up Bitcoin uni and Link we are sold off at the six and you know like the units here do not really matter it's only for comparison but like these three assets have sort of the same value we see and this is kind of justifiable probably

that you know is has a lower value which means that you know allegedly they take more risk but you know alternatively you can just argue that they see is is a safer asset which is of course true and and then finally if you look at the comp token so you see it as the lowest uh long to Value but also the lowest C which means that you know according to this formula at least despite having the uh kind of the most conservative setup for it they're actually taking the biggest risk with comp and also there you know we can speculate why they decide to do it and you know it's kind of maybe natural

for the compound Community to have more trust in comp uh so this is kind of what this formula can offer now but of course the future uh we see it is is much bigger yeah so so yeah so in in the future you know we definitely Envision fully permissionless Landing markets uh where all the loan to value and risk parameters uh are done uh fully automated but automated with a smart contact not automated by just some back end uh it can still you know allow the Dow to keep a lot of the sort

of power they currently have the Dao for example can configure this confidence level they can still potentially Blacklist certain assets but but yeah but like and they can also decide actually on the formulas either they want to apply for the risk management but then you know just by having this data layer inside so yeah so that's that's kind of the missing block towards a fully permissionless Landing Market which will again potentially be much less efficient might even be a bit rescue but you know we'll have the upsides of kind of being

more loyal to the defy ethos and all the other potential advantages that you know can come with it it can also you know do some more basic stuff for example it can Blacklist certain assets uh to not allow you know like the market manager to list them Etc so you know this device it was you know other than just you know ideals and uh I don't know coding law statements Etc at the end of the day you know there's a lot of competitive barrier right now if you want to start your own Landing Market

because eventually you know you have to compete let's say if you want to compete with uni swap then you know you can kind of compete on the kind of hard objectives such as uh you know user interface marketing uh but it's kind of you know like an immutable product contract that if you folk people can just use it allegedly with the same risk level as they use the uni swap but of course if you want to Fork Ave or compound I don't know I mean you'll have right now have to Fork all of their team uh and you know and of course it's not uh very likely that a smart contract

will outperform you know other or make your team if they're in the audience uh but still having some you know self-executable smart contract can kind of plane the level uh and allow for competitive Advantage uh to be less of a burial uh you know towards more innovation in this space you know we drill offer more neutrality automation because also a lot of the risk management require dark votes which is very tedious process you know much more transparency and decentralization uh so yeah so is that I'm conclude and happy to take any questions if you have

any questions [Applause] oh hi um thanks for the presentation um I wonder with your proposed solution where do you see the risk of being manipulated as well as being accurately reflecting on the loan to value ratio um

sort of being that how do we deal with that yeah so so even before manipulation you know if you look at the USD the lunar stablecoin it wasn't even a manipulation but you know it was the world was a lot Beyond you know a lot more there Beyond you know kind of dry data right I mean like USD you know it's kind of the Thanksgiving turkey paradox so you know you feed the turkey for a month and you know kind of the confidence that the turkey has on its owner only keeps increasing so you know Luna was stable for years and allegedly if you look at the data you know confidence has to be kind of up only but

of course we all knew how it ended ended in terms of you know kind of blunt manipulations so this is something you know we hope to kind of handle in the in the data layer right so instead of just taking like the current market liquidity we can average it you know on the longer durations and then manipulation will be costly uh this is why the data in the Oracle layer is not a trivial part and eventually uh it will take you know kind of a key role in preventing these manipulations sorry yeah if you do like an average out sort of value like a top sort of value

but that it's not going to be very accurate though I don't know yeah yeah yeah yeah no okay but but here you know like let's say even if you take like a day average or a week average so yeah it's it's not accurate but of course you know if you compare to the current status quo so you know you have like making an oven compound maybe maybe voting on parameters once a month so actually in that regard you know even if you take like a weak average it's still a big Improvement comparison in comparison to now still replacing human is eventually less accurate you know humans humans are smart so so so I think it's mostly about that I mean like manipulating the data will be fairly easy to mitigate

uh but you know kind of embedding domain specific knowledge so this is still an out but again we do hope uh you know like along the way idea to figure that out but also that you know people will be willing to kind of do the compromise and to be less precise but more permissionless thank you anyone else okay yeah thank you very much Aaron thank you

[Applause] [Music] foreign [Music]