|
|
Post by solicitorious on Sept 19, 2015 15:35:48 GMT
Only 7%, I'm surprised, I bet most have already made more than 7% in interest to date. Although your recoveries could take years to materialise and a lot of opportunity cost. Yes, but a non-negligible chance of losses exceeding 10%, and that is assuming full application of the PF. Also, as I stated at the beginning, these numbers are based on an idealised investment profile, where you own the entire loanbook, or are at least invested in perfect proportion to it. Your actual investment profile will be different, with potentially very different loss probabilities, perhaps better, perhaps worse... For example, if you were only invested in the Superyacht, your loss probability would be zero for all the numbers we have discussed so far. Whereas, if you were only invested in PBL058, while you might only have a 50% chance of a loss, such loss would almost certainly be around 20%, giving an expected loss of 10.3%. And just running the (D=50%, L=50%) model again for equal amounts in each loan I find an overall expected loss of 8.01% instead of 6.96% and almost twice the chance (16.09%) of a loss in excess of 10%.
|
|
|
|
Post by marek63 on Oct 7, 2015 14:59:09 GMT
Did you play with any more on these ideas??
|
|
|
|
Post by solicitorious on Oct 7, 2015 18:10:37 GMT
I'm playing all the time.  Was hoping for some more ideas to maybe refine the model. I'll do another model run shortly to take account of recent loans. Need to optimize the code too, as it takes about 25 mins to run! |
|
registerme
Member of DD Central
Posts: 6,624 
Likes: 6,437
|
Post by registerme on Oct 7, 2015 18:35:03 GMT
Do you try to model correlation risk and jump risk?
|
|
|
|
Post by solicitorious on Oct 7, 2015 20:46:47 GMT
Do you try to model correlation risk and jump risk? 'fraid not, as I don't really know what they are, or whether they are within the capabilities of Excel. If you can give me a steer, I'll look into it.  Based on the current 47 live loans. for D=50%, L=50% (uniform), Model says: Chance of any loss 100.00%, no loss 0.00% loss <0.5% 0.03% loss >0.5% 99.97% loss >1% 99.95% loss >2% 99.58% loss >3% 98.20% loss >5% 85.01% loss >10% 7.03% loss >20% 0.01% loss >30% 0.00% loss >40% 0.00% loss >50% 0.00% overall loss 7.08%, including times when there's no loss average loss 7.08%, if there is a loss Comparison with when there's no PF Chance of any loss 100.00%, no loss 0.00% loss <0.5% 0.00% loss >0.5% 100.00% loss >1% 100.00% loss >2% 100.00% loss >3% 99.95% loss >5% 98.20% loss >10% 32.50% loss >20% 0.01% loss >30% 0.01% loss >40% 0.00% loss >50% 0.00% overall loss 9.08%, including times when there's no loss average loss 9.08%, if there is a loss
|
|
registerme
Member of DD Central
Posts: 6,624
Likes: 6,437
|
Post by registerme on Oct 7, 2015 21:21:04 GMT
Correlation risk - www.investopedia.com/terms/c/correlation.aspJump risk - www.investopedia.com/terms/e/eventrisk.aspThey can be modelled in Excel, and they can be used with MC simulations, but you may need additional maths libraries / to write your own. I'm no quant, but to my non-maths way of thinking about such things correlation risk is that defaults aren't independent, ie if we see a default in a property bridging loan then we should really think that other property bridging loans may be more likely to default - they are not discrete. Jump risk is "on default, the price won't change by 1.5%, it will change by 30+%". Think Switzerland putting a floor on their currency or sub-prime CDOs going from AAA rated 100p in the pound to 15p in the pound overnight - there's no nice default curve (or currency pair value, or equity price etc) but a BANG. |
|
|
|
Post by solicitorious on Oct 7, 2015 21:55:39 GMT
Correlation risk - www.investopedia.com/terms/c/correlation.aspJump risk - www.investopedia.com/terms/e/eventrisk.aspThey can be modelled in Excel, and they can be used with MC simulations, but you may need additional maths libraries / to write your own. I'm no quant, but to my non-maths way of thinking about such things correlation risk is that defaults aren't independent, ie if we see a default in a property bridging loan then we should really think that other property bridging loans may be more likely to default - they are not discrete. Jump risk is "on default, the price won't change by 1.5%, it will change by 30+%". Think Switzerland putting a floor on their currency or sub-prime CDOs going from AAA rated 100p in the pound to 15p in the pound overnight - there's no nice default curve (or currency pair value, or equity price etc) but a BANG. OK, thanks. I am generally familiar with these ideas, and have tried to simplify the model with the D and L parameters, also generating the curve for D=100%, which is obviously the worst-case scenario.
|
|
bigfoot12
Member of DD Central
Posts: 1,817
Likes: 816
|
Post by bigfoot12 on Oct 7, 2015 23:48:35 GMT
Correlation risk - www.investopedia.com/terms/c/correlation.aspJump risk - www.investopedia.com/terms/e/eventrisk.aspThey can be modelled in Excel, and they can be used with MC simulations, but you may need additional maths libraries / to write your own. I'm no quant, but to my non-maths way of thinking about such things correlation risk is that defaults aren't independent, ie if we see a default in a property bridging loan then we should really think that other property bridging loans may be more likely to default - they are not discrete. Jump risk is "on default, the price won't change by 1.5%, it will change by 30+%". Think Switzerland putting a floor on their currency or sub-prime CDOs going from AAA rated 100p in the pound to 15p in the pound overnight - there's no nice default curve (or currency pair value, or equity price etc) but a BANG. OK, thanks. I am generally familiar with these ideas, and have tried to simplify the model with the D and L parameters, also generating the curve for D=100%, which is obviously the worst-case scenario. You could probably get an easy first step to modelling the correlated risk of default (and correlated falls in asset prices) by introducing an extra random variable which you add to all the others. So for loan 1 the default random variable is X0 + X1, and for loan 2 it is X0 + X2 and so on, for each simulation X0 is the same for all loans. Also I think that you should include the risk of fraud or other losses by having a non zero probability of total loss, independently from the LTV. You say that the superyacht has a very low LTV, but they sink, catch fire, get stolen, and I assume it is insured, but insurance companies do their best not to pay out. If the borrower is struggling he might have failed to insure it, or only insure it for pleasure use and not for business use. Might it be seized in another territory in lieu of some other debts? I am very new to SS so I am not familiar with this loan, I haven't read any of the details. EDIT:- The SD of X0 would be lower than that for X1..Xn
|
|
|
|
Post by solicitorious on Oct 8, 2015 0:54:50 GMT
OK, thanks. I am generally familiar with these ideas, and have tried to simplify the model with the D and L parameters, also generating the curve for D=100%, which is obviously the worst-case scenario. You could probably get an easy first step to modelling the correlated risk of default (and correlated falls in asset prices) by introducing an extra random variable which you add to all the others. So for loan 1 the default random variable is X0 + X1, and for loan 2 it is X0 + X2 and so on, for each simulation X0 is the same for all loans. Also I think that you should include the risk of fraud or other losses by having a non zero probability of total loss, independently from the LTV. You say that the superyacht has a very low LTV, but they sink, catch fire, get stolen, and I assume it is insured, but insurance companies do their best not to pay out. If the borrower is struggling he might have failed to insure it, or only insure it for pleasure use and not for business use. Might it be seized in another territory in lieu of some other debts? I am very new to SS so I am not familiar with this loan, I haven't read any of the details. EDIT:- The SD of X0 would be lower than that for X1..Xn Not sure these can be modelled additively. Surely the (remote) chance of a total loss must be considered first, and only if not true then a "normal" loss considered, given a default, otherwise the total modelled loss could exceed the asset value? What probability would you place on a total loss? 1 in a thousand?
|
|
bigfoot12
Member of DD Central
Posts: 1,817
Likes: 816
|
Post by bigfoot12 on Oct 8, 2015 8:32:10 GMT
You could probably get an easy first step to modelling the correlated risk of default (and correlated falls in asset prices) by introducing an extra random variable which you add to all the others. So for loan 1 the default random variable is X0 + X1, and for loan 2 it is X0 + X2 and so on, for each simulation X0 is the same for all loans. Also I think that you should include the risk of fraud or other losses by having a non zero probability of total loss, independently from the LTV. You say that the superyacht has a very low LTV, but they sink, catch fire, get stolen, and I assume it is insured, but insurance companies do their best not to pay out. If the borrower is struggling he might have failed to insure it, or only insure it for pleasure use and not for business use. Might it be seized in another territory in lieu of some other debts? I am very new to SS so I am not familiar with this loan, I haven't read any of the details. EDIT:- The SD of X0 would be lower than that for X1..Xn Not sure these can be modelled additively. Surely the (remote) chance of a total loss must be considered first, and only if not true then a "normal" loss considered, given a default, otherwise the total modelled loss could exceed the asset value? What probability would you place on a total loss? 1 in a thousand? 1 in a thousand seems like a sensible starting point. The interesting exercise is to see how sensitive the model would be to a change in this parameter. I agree that this isn't additive and I would model it as you suggest. My bivariate suggestion was to introduce correlation into the probability of default and possibly that of loss given default.
|
|
|
|
Post by savingstream on Oct 8, 2015 10:52:44 GMT
Remember the ONLY likelihood of a total loss is a massive fraud committed by the borrower. 1 in a 1000 seems to be very conservative, so we would put it at 1 in 10,000 maybe 1 in 100,000 given our DD and security measures in place and even then, those figures are conservative.
All assets have indemnified valuations.
Our legals are indemnified.
We have a big pot of cash in the Provision Fund which is increasing all the time.
We are the most profitable P2P lender in the UK, probably Europe and most likely the known Universe.
Our overheads are very negligible (for now, as we are building the infrastructure and team to accomodate future growth).
|
|
registerme
Member of DD Central
Posts: 6,624
Likes: 6,437
|
Post by registerme on Oct 8, 2015 10:58:10 GMT
Remember the ONLY likelihood of a total loss is a massive fraud committed by the borrower. 1 in a 1000 seems to be very conservative, so we would put it at 1 in 10,000 maybe 1 in 100,000 given our DD and security measures in place and even then, those figures are conservative. All assets have indemnified valuations. Our legals are indemnified. We have a big pot of cash in the Provision Fund which is increasing all the time. We are the most profitable P2P lender in the UK, probably Europe and most likely the known Universe. Our overheads are very negligible (for now, as we are building the infrastructure and team to accomodate future growth). Love the reply (hence the thumbs up), but I really have to call savingstream out on the "most likely the most profitable P2P lender in the known Universe"  .
|
|
|
|
Post by solicitorious on Oct 8, 2015 11:22:45 GMT
Remember the ONLY likelihood of a total loss is a massive fraud committed by the borrower. 1 in a 1000 seems to be very conservative, so we would put it at 1 in 10,000 maybe 1 in 100,000 given our DD and security measures in place and even then, those figures are conservative. All assets have indemnified valuations. Our legals are indemnified. We have a big pot of cash in the Provision Fund which is increasing all the time. We are the most profitable P2P lender in the UK, probably Europe and most likely the known Universe. Our overheads are very negligible (for now, as we are building the infrastructure and team to accomodate future growth). What comments would you have on bigfoot12 's points about possible insurance failure or extra-jurisdictional seizure of an asset? And could you address a point I raised earlier that, in principle, the Provision Fund would be applied to exhaustion, if (God forbid) necessary to cover losses?
|
|
bigfoot12
Member of DD Central
Posts: 1,817
Likes: 816
|
Post by bigfoot12 on Oct 8, 2015 12:30:36 GMT
I love doing simulations with spreadsheets, so I thought I'd try a Monte Carlo Analysis of the SS loanbook. We run the model say 10,000 times What probability would you place on a total loss? 1 in a thousand? It just occurred to me that if you are including rare events you probably need to increase the number of simulations. (You can probably get a good idea if you are doing enough by running it a few times and seeing how much each run varies from the others.)
|
|
Investor
Member of DD Central
Posts: 662
Likes: 590
|
Post by Investor on Oct 8, 2015 13:09:17 GMT
Remember the ONLY likelihood of a total loss is a massive fraud committed by the borrower. 1 in a 1000 seems to be very conservative, so we would put it at 1 in 10,000 maybe 1 in 100,000 given our DD and security measures in place and even then, those figures are conservative. All assets have indemnified valuations. Our legals are indemnified. We have a big pot of cash in the Provision Fund which is increasing all the time. We are the most profitable P2P lender in the UK, probably Europe and most likely the known Universe. Our overheads are very negligible (for now, as we are building the infrastructure and team to accomodate future growth). Though I can quite categorically state that for those who subscribe to either quantum multiverse theories, or M-Theory (extended), that the chances of this Saving Stream being the most profitable in the known universes is very small, 1:∞ to be precise
|
|
Was hoping for some more ideas to maybe refine the model. 
.