|
|
Post by solicitorious on Apr 5, 2016 13:19:12 GMT
All very interesting, but for practical purposes completely meaningless, as the results are entirely dependent on the values given to the base parameters, which is complete guess work and will vary significantly depending on the specifics of each individual loan and external macro factors. About as constructive as posting "This is a comment"! It adds nothing than what has already been discussed. It's a model, and will have inputs, which could be anything from 0-100%. The point of the model is to discover "What if" the inputs are D% and L%, which are readily understandable numbers already used in the P2P industry. That is, what does the histogram of losses look like for the loanbook (and more importantly, for my personal portfolio) for well-chosen input values. Well-chosen being the likely worst case scenarios. You can decide what that should be, but for the sake of example we're trying D=50%, L=50%. The model can also provide answers for the limiting case. Every loan defaults (D=100%), so what is the loanbook overall loss for any value of L%?  Losses begin when L% > 32.85% Max loss of 98% when L%=100% A capital loss of 12% (equivalent to a year's simple interest) is incurred when L% is 44.8%
|
|
|
|
Post by solicitorious on Apr 6, 2016 4:15:01 GMT
So then, would this be a fair summation? 'Despite a recent increase in theoretical risk [owing to savingstream taking on several significantly larger loans], the Monte Carlo analysis demonstrates that the overall risk of investing in SS, under quite bleak hypothetical scenarios, remains quite tolerable.'Would be genuinely interested to hear whether Y/N - with reasoning, of course... FWIW, the same analysis could be applied to other platforms and IMHO could become the industry-standard benchmark or metric for assessing and comparing platform loan-risk. In the above graph, the further to the "right" a platform's slope appears, the better. I suspect, although have not proved, that SS is superior to AC, FS and MT in this regard, and suspect the differences would be quite marked for at least one of these platforms. I can't do this analysis single-handedly of course... I would need numerate helpers to collate the data, and I would then apply the MC analysis. As an aside - I am surprised I am the only person - seemingly - who is attempting, or who has ever attempted, this sort of quantitative analysis of P2P... |
|
ben
Posts: 2,020 
Likes: 589
|
Post by ben on Apr 6, 2016 7:41:16 GMT
All very interesting, but for practical purposes completely meaningless, as the results are entirely dependent on the values given to the base parameters, which is complete guess work and will vary significantly depending on the specifics of each individual loan and external macro factors. any financial model/theory in any subject would be pointless doing with that opinion, nobody saying it is going to be right but it shows what could happen if x happens, nobody can calculate for all variables that why there is a risk attatched to it.
|
|
|
|
Post by Deleted on Apr 6, 2016 15:07:05 GMT
Each to their own, but I stand by my comments. ![]()  |
|
|
|
Post by solicitorious on Apr 6, 2016 17:06:05 GMT
Each to their own, but I stand by my comments. ![]() Your cut-and-paste skills do you credit, but... you have not even attempted to demonstrate how either the data or the model is "garbage"... 
|
|
|
|
Post by solicitorious on Apr 6, 2016 17:26:36 GMT
All very interesting, but for practical purposes completely meaningless, as the results are entirely dependent on the values given to the base parameters, which is complete guess work and will vary significantly depending on the specifics of each individual loan and external macro factors. I haven't looked at the approach taken by solicitorious and I don't understand why he would think SS is somehow "superior" to other platforms. I suspect that if curves were drawn for the other platforms, similar as in the last graph I posted, they would all be to the left of the SS curve, i.e. losses would start occurring at a lower L% (i.e. the across the board fall in asset value in the event of default) than SS's 32.85%, and would indicate higher loanbook losses for each point on the L% scale. Reasons would be:- no PF, higher LTVs, more 2nd charges, etc. In this particular aspect, I am confident SS would be demonstrated as "superior", perhaps markedly so against another platform.
|
|
mikes1531
Member of DD Central
Posts: 6,453
Likes: 2,320
|
Post by mikes1531 on Apr 7, 2016 16:42:29 GMT
Anyhow, for the current loan book* for D=50%, L=50% (uniform), T=0.10% (1 in a 1000 chance of a total loss) Model says: <snip> overall loss 8.61%, including times when there's no loss average loss 8.61%, if there is a loss Comparison with when there's no PF <snip> overall loss 10.61%, including times when there's no loss average loss 10.61%, if there is a loss Is the difference between the two cases -- with and without PF -- exactly 2% because that's the amount of the PF? Or is that just a coincidence?
|
|
|
|
Post by solicitorious on Apr 7, 2016 17:36:56 GMT
Anyhow, for the current loan book* for D=50%, L=50% (uniform), T=0.10% (1 in a 1000 chance of a total loss) Model says: <snip> overall loss 8.61%, including times when there's no loss average loss 8.61%, if there is a loss Comparison with when there's no PF <snip> overall loss 10.61%, including times when there's no loss average loss 10.61%, if there is a loss Is the difference between the two cases -- with and without PF -- exactly 2% because that's the amount of the PF? Or is that just a coincidence? Yes, when as in this example, with large and numerous losses which for each sample the loss total always exceeds the PF, the With PF loss will always result in the full PF or 2% being deducted from the loss. Hence, when averaged, there will be a 2% difference in the overall figures. e.g. (for simplicity assume PF is £2 mil) Sample With PF loss loss 4 2 6 4 5 3 3 1 7 5 .. .. When you take the averages of both columns, there is a difference of 2 between the averages. This obviously won't be the case for lower input parameters which result in a fewer/lower losses such that the PF is not always used in its entirety.
|
|
|
|
Post by solicitorious on Apr 7, 2016 20:39:54 GMT
FWIW, the same analysis could be applied to other platforms and IMHO could become the industry-standard benchmark or metric for assessing and comparing platform loan-risk. For example, here is a different platform (no PF). I won't name them until I have triple-checked the figures.  Losses begin when L% > 20.0% Max loss of 100% when L%=100% A capital loss of 12% (equivalent to a year's simple interest) is incurred when L% is 44.3% The Monte Carlo histogram for D=50%, L=50%, T=0.10% 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% 99.93% loss >3% 99.56% loss >5% 94.78% loss >10% 53.46% loss >15% 9.21% loss >20% 0.13% loss >30% 0.03% loss >40% 0.00% loss >50% 0.00% overall loss 10.38%, including times when there's no loss average loss 10.38%, if there is a loss
|
|
|
|
Post by solicitorious on Nov 22, 2016 18:25:41 GMT
Bi-annual update, for the current loan book for D=50%, L=50% (uniform), T=0.10% (1 in a 1000 chance of a total loss) Model says: 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% 99.92% loss >3% 98.91% loss >5% 76.95% loss >10% 0.66% loss >15% 0.01% loss >20% 0.00% loss >30% 0.00% loss >40% 0.00% loss >50% 0.00% overall loss 6.04%, including times when there's no loss average loss 6.04%, 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% 100.00% loss >5% 98.91% loss >10% 7.96% loss >15% 0.04% loss >20% 0.00% loss >30% 0.00% loss >40% 0.00% loss >50% 0.00% overall loss 8.04%, including times when there's no loss average loss 8.04%, if there is a loss A significant positive change since 7 months ago. I would put this down to the repayment of some 2nd charge loans, so now only three remain, a general lowering of LTVs (weighted average down from about 61% to 56%) and the ballooning of the PF to a shade below £3 million. The risk of a significant capital loss on a fully-diversified portfolio has all-but been eliminated (according to this model, its assumptions and parameters...) Congratulations savingstream !
|
|
jamesc
Member of DD Central
Posts: 447
Likes: 253
|
Post by jamesc on Nov 22, 2016 19:39:57 GMT
Bi-annual update, for the current loan book for D=50%, L=50% (uniform), T=0.10% (1 in a 1000 chance of a total loss) Model says: 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% 99.92% loss >3% 98.91% loss >5% 76.95% loss >10% 0.66% loss >15% 0.01% loss >20% 0.00% loss >30% 0.00% loss >40% 0.00% loss >50% 0.00% overall loss 6.04%, including times when there's no loss average loss 6.04%, 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% 100.00% loss >5% 98.91% loss >10% 7.96% loss >15% 0.04% loss >20% 0.00% loss >30% 0.00% loss >40% 0.00% loss >50% 0.00% overall loss 8.04%, including times when there's no loss average loss 8.04%, if there is a loss A significant positive change since 7 months ago. I would put this down to the repayment of some 2nd charge loans, so now only three remain, a general lowering of LTVs (weighted average down from about 61% to 56%) and the ballooning of the PF to a shade below £3 million. The risk of a significant capital loss on a fully-diversified portfolio has all-but been eliminated (according to this model, its assumptions and parameters...) I think that is interesting but IMHO I think some of your reasoning is flawed, I don't think LTVs have lowered e.g. J** was 63% I suspect it maybe because you are using the headline LTVs for some of the DFLs e.g. DFL009 has a headline LTV of 35% but that is against GDV whereas currently LTV is closer to 100%. Although I do agree that the increased PF must help.
|
|
|
|
Post by solicitorious on Nov 23, 2016 9:22:43 GMT
OK, a slight fly in the ointment I had overlooked.  If you could list the loans that are affected by GDV calculations, with their "true" LTVs, I will run it again. Someone might like to create a sticky list of the GDV-based loans, so lenders are always aware of the difference... |
|
cooling_dude
Bye Bye's for the PPI
Posts: 2,853
Likes: 4,298
|
Post by cooling_dude on Nov 23, 2016 11:42:29 GMT
OK, a slight fly in the ointment I had overlooked. If you could list the loans that are affected by GDV calculations, with their "true" LTVs, I will run it again. Someone might like to create a sticky list of the GDV-based loans, so lenders are always aware of the difference... All the LTGV loans are on my index in bright purple ( Here) - The OP of all the DFLs threads indicate the true LTV if available. However, for DFLs, a true LTV isn't always available, as SS don't provide a new valuation doc at each tranche; we only have the original VR, and may be lucky to get the odd VRs during the development. Even when we do have a VR when a Tranche is released, as soon as there are any immediate works after that VR, the property will either increase or decrease in value (as is the case with DFL001, which when from being valued @ £4,000,000 when the loan went live to £3,500,000 5 months late - subsequently the LTV went from 80% to 118%). Maybe it is best just to omit the DFL loans for the simulation, as for these loans there too many variables as noted above. "A Monte Carlo Simulation of Losses for PBL loans".
|
|
