ceejay
Posts: 975 
Likes: 1,149
|
Post by ceejay on Apr 25, 2018 22:01:37 GMT
In 2007 the financial system was brought down when mortgages of sound borrowers were diced and sliced with mortgages to unsound borrowers to produce financial instruments that were then sold on in the financial markets. The UK (and USA) spent the previous 11 years clawing their way out of that hole. My perception is this Algorithm system seems to be doing something similar though smaller and more transparently. Rubbish. That is not what happened at all. To be fair, that post was made approaching closing time...
|
|
cb25
Posts: 3,528
Likes: 2,668
|
Post by cb25 on Apr 25, 2018 22:06:02 GMT
In 2007 the financial system was brought down when mortgages of sound borrowers were diced and sliced with mortgages to unsound borrowers to produce financial instruments that were then sold on in the financial markets. The UK (and USA) spent the previous 11 years clawing their way out of that hole. My perception is this Algorithm system seems to be doing something similar though smaller and more transparently. Rubbish. That is not what happened at all. I would accept an element of that as truth, that sound/unsound mortgages were packaged together and sold on, but reasons why it went belly up (imo) were a) credit agencies incorrectly giving products AAA rating, b) risk being incorrectly priced, c) models incorrectly assuming almost zero risk of failure, d) absolutely massive leveraging.
Can't see any overlap between that and AC's algorithm
|
|
bg
Member of DD Central
Posts: 1,368
Likes: 1,929
|
Post by bg on Apr 26, 2018 6:18:35 GMT
Rubbish. That is not what happened at all. I would accept an element of that as truth, that sound/unsound mortgages were packaged together and sold on, but reasons why it went belly up (imo) were a) credit agencies incorrectly giving products AAA rating, b) risk being incorrectly priced, c) models incorrectly assuming almost zero risk of failure, d) absolutely massive leveraging.
Can't see any overlap between that and AC's algorithm
If you want to sum up what happened in 2007/8 then the best/easiest summary would be excess debt and d) absolutely massive leveraging. Northern Rock for example in the UK went under because it overstretched itself. Made loads of mortgages and funded them by borrowing on the short term money markets. When money markets froze they couldn't roll their funding and so became insolvent. They overleveraged, plain and simple. When it comes to CDO's (which I think is what brightspark is referring to), good loans were not being mixed up with bad loans. The loans were all 'bad'. In general, pools of loans were made to people that couldn't afford to repay them (subprime loans). Often with a low teaser rate they could afford that then jumped in a couple of years to a rate that they had no hope in paying. These pools were then sliced into tranches, with the riskiest (junior) tranches taking the first hit of any loss and the safer (senior) tranches taking any deeper losses. Problems then arose when these senior tranches were given AAA ratings by the agencies and sold on as if they had the same risk profile as prime mortgages. Thing is they didn't...when the loans defaulted and the house market tanked houses were being sold for peanuts and senior tranches were being wiped out along with the more junior tranches. The value of some of these AAA loans went close to zero which shocked the financial system and lead to massive losses. I'm summarising a lot here but which ever way you look at it there is no analogy to the GBBA. The GBBA operates on the same model as FC, RS, Zopa - or any of the big platforms slice up loans so lenders are diversified. Diversification is why P2P works, it was why it was set up. Lenders put a small amount into each loan to spread risk and of course it makes sense to automate this process. A closer example of a CDO style loan in my opinion would be any 2nd charge loan or the kind of splitting into tranches of loans the likes of MT have done (for example the Birkenhead loan). AC have never split loans into differing risk tranches.
|
|
|
|
Post by brightspark on Apr 26, 2018 7:42:02 GMT
I can see that opinions vary on my previous comment from I am uttering a load of tosh to perhaps there is a grain of truth in what I am saying. As we saw in 2007 diversification is ok only up to a point in reducing the impact of poor loans. Meanwhile I shall stick to the MLIA having had my fingers mildly singed via GBBA1 with a chunk of dud loan 227.
|
|
oxdoc
Posts: 19
Likes: 20
|
Post by oxdoc on Apr 26, 2018 20:59:12 GMT
Hi chris, thanks for explaining the way the allocation algorithm works, and I appreciate that initially a lender's funds may have to be concentrated in a few contracts to get money lent quickly. I don't understand, though, why loan chunks need to stay so large for so long. If 20% of a person's initial investment is put into one loan, and the average percentage allocated to that loan is 1%, why can't parts of that loan be immediately traded with people who have less than 1% in return for chunks of their loans of the same size, until the loan is about 1% of everyone's balance? It seems to me that unless the initial investment is very large, then that should be possible to do - it is known what each person needs to buy and sell of each loan to have the average allocation, and whilst it may not be possible to give everyone exactly that amount, I don't see why getting it into the low single figure percentages for everybody should take that long - if a number of people invest in a loan that makes up 20% of their account, getting it down to no more than 2% in anyone's account would seem to only take 9 transactions per user. I have not sat down and worked through the maths of it, though, so there could be something I have overlooked.
|
|
|
|
Post by chris on Apr 26, 2018 21:40:23 GMT
Hi chris , thanks for explaining the way the allocation algorithm works, and I appreciate that initially a lender's funds may have to be concentrated in a few contracts to get money lent quickly. I don't understand, though, why loan chunks need to stay so large for so long. If 20% of a person's initial investment is put into one loan, and the average percentage allocated to that loan is 1%, why can't parts of that loan be immediately traded with people who have less than 1% in return for chunks of their loans of the same size, until the loan is about 1% of everyone's balance? It seems to me that unless the initial investment is very large, then that should be possible to do - it is known what each person needs to buy and sell of each loan to have the average allocation, and whilst it may not be possible to give everyone exactly that amount, I don't see why getting it into the low single figure percentages for everybody should take that long - if a number of people invest in a loan that makes up 20% of their account, getting it down to no more than 2% in anyone's account would seem to only take 9 transactions per user. I have not sat down and worked through the maths of it, though, so there could be something I have overlooked. There's a couple of factors. With your system every time anyone invested or withdrew or made a loan purchase you would need to rebalance everyone's accounts, many of which would end up being tiny. Imagine a lender receives a 10p interest payment which is then reinvested into a loan. Everyone else then needs to rebalance all their holdings to average out that 10p. Then you have everyone else's trades, deposits, withdrawals, and repayments all generating transactions which need further rebalancing. That can quickly cause it to spiral out of control. We jump through several hoops to make that happen with the access accounts but they wouldn't work with the bespoke mandates so we've decided to retain this separate system. The solution we've chosen is to accept asymmetric holdings and work around that as best we can - which is an exponentially difficult problem if you look for the perfect solution (i.e. computationally impossible to solve at scale). Instead the system searches through random combinations of lenders and loans to find matches where it can to move each lender toward the average. It would be more effective to move each lender toward the average of the two lenders and then iterate quickly to match more and more lenders but this would generate many more transactions. We've also put in place a limit so that lenders will not be moved to too concentrated a position in a single loan even if the average dictates they should - this may end up being temporarily disabled as we tune the accounts and make sure everyone ends up in fair position. With our solution it works best when there are imbalances to work with. If the majority of lenders are broadly balanced then the tolerances that are designed to minimise the otherwise constant rebalancing mean that there are few trades that can occur. This is the current state of play for many lenders, so it'll take new flow of funds and new investments to create imbalances that the system can match. Because of the tolerances and our desire to minimise the otherwise constant rebalancing this will continue to take some tweaking and adjustment to find settings that work for everyone. I've just made one change to try and shuffle things a little more, there are more I can make as we see better how it responds to real world usage.
|
|
|
|
Post by Deleted on Apr 27, 2018 11:56:03 GMT
As excellent as chris is at explaining the inner workings of the "ALGORITHM" it still appears to be an elusive and peculiar beast to comprehend. Despite holding 30% in #441 I've seen no movement in approaching a week .... however In a single day 5 exchanges of #529 reduced my holding from 2% to literally 0! £0.00 .... although I do still hold it in my PSA which I didn't realise overlapped with the GBBA2 ? |
|
oxdoc
Posts: 19
Likes: 20
|
Post by oxdoc on Apr 27, 2018 21:36:55 GMT
chris Thanks for your explanation of the algorithm. I agree that balancing very small amounts is not worth it. But I still don't see why more can't be done to reduce people's exposures to loans when they are in the tens of percent. Suppose someone has a loan that is 20% of their account, say, and it's desired for loans not to make up more than 3% of anyone's account, and the loan makes up less than 3% of the AC loan book. Then people without that loan could be chosen at random and made to swap loan parts up to 2% of the value of the balance of whoever's account is smaller, and this could be repeated until the loan falls below 2% of the first person's account. The difference between 3% and 2% would give a buffer so that no more rebalancing would be necessary unless a large fraction of the account balance is withdrawn or something. Would this really be impractical? It seems similar to what you say your algorithm is doing, except not picking pairs of borrowers at random, but starting with those that have loan parts that make up a large fraction of their account.
|
|
daveb4
Member of DD Central
Posts: 220
Likes: 116
|
Post by daveb4 on May 3, 2018 16:48:17 GMT
GBBA1 over last few weeks keeps selling my loans at say £1 or £2 and purchasing less than 1p so now have accumulated 15% awaiting investment. I have system set up that repaid interest and principle goes into my GBBA2 which I accept but confused with all my loans earning 7% being slowly sold and not reinvested back?
|
|
|
|
Post by elephantrosie on May 3, 2018 21:55:09 GMT
how exactly does GBBA and provision fund work?
|
|
