benaj
Member of DD Central
N/A
Posts: 5,608 
Likes: 1,738
|
Post by benaj on Oct 31, 2019 13:26:50 GMT
www.zopa.com/invest/proof ( as of April 1st 2019) Some interesting stats from Zopa. Median returnCore 3.9% Plus 4.6% Less than 1% made a loss90% investors returnCore: over 2.7% Plus: over 3.4%
|
|
|
|
Post by propman on Oct 31, 2019 15:35:50 GMT
This suggests that the variation in returns for Plus are similar to Core which I am surprised about. Does anyone know whether the matching algorithm aims to equalise actual returns or expected returns? I have always assumed the latter, but the former would reduce the variability from defaults and would mean anyone with below average performance would get better loans than the lucky ones! (When it was first introduced they did say that the averaging was over a relatively short period and so would not have made sence to include defaults, but I assume it has been amended since).
|
|
zlb
Member of DD Central
Posts: 1,422
Likes: 333
|
Post by zlb on Oct 31, 2019 17:31:13 GMT
My plus does worse than core.
|
|
aju
Member of DD Central
Posts: 3,500
Likes: 924
|
Post by aju on Oct 31, 2019 17:59:27 GMT
Me thinks this is suffering from that great old "Lies, damned lies, and statistics" quote. It remind me of the days I cobble together yet one more spreadsheet and then 3 months later wonder what on earth I was thinking. I wonder what Zopa was thinking in this page ... there seems a lot of possible assumptions in the article I feel. Not sure NAR is that representative either but hey I'm up on the deal at present  |
|
trevor
Member of DD Central
Posts: 557
Likes: 381
|
Post by trevor on Oct 31, 2019 22:01:12 GMT
My core has returned a rubbish 2.4% over the last 6 months. Rubbish performance. I've reduced my holding recently in favour of LW. If Z don't improve in the next 6 months I will reduce to zero.
|
|
|
|
Post by propman on Nov 1, 2019 10:18:42 GMT
I'm sure that there will be some investors who put in little and were unlucky, hence the cut off at 10%. Also, I would assume that this will ignore any sale fees and will be over the whole term that this asset class has been held. As discussed this is likely to dip after 4 months and again when reinvestment is curtailed. I would guess that >50% have probably reinvested all returns thus flattering the median.
|
|
Greenwood2
Member of DD Central
Posts: 4,384
Likes: 2,784
|
Post by Greenwood2 on Nov 1, 2019 10:57:53 GMT
Mmm stats, you have to love them.
I also don't think there are any lucky investors who have had huge returns, these stats do not imply that at all.
- They use median not mean, so we can safely assume the mean is lower.
- The median figures appear lower than all projected returns email I checked from the last few months.
- 90% return at least x% => so the unlucky 10% are getting less, probably quite a bit less.
- The median could even be the maximum anyone received, see table below (of course that's an extreme example).
| User | Core return (%)
| Plus return (%)
| 1
| 0
| 0
| | 2 | 2.7
| 3.4
| 3
| 2.7
| 3.4
| | 4 | 2.7
| 3.4
| | 5 | 3.9
| 4.6
| | 6 | 3.9
| 4.6
| | 7 | 3.9
| 4.6
| | 8 | 3.9
| 4.6
| | 9 | 3.9
| 4.6
| | 10 | 3.9
| 4.6
| | Median | 3.9
| 4.6
| 90% return at least:
| 2.7
| 3.4
|
In this example the mean returns are 3.15% and 3.78%. It would be easy to construct even lower mean returns with more samples (so some could be negative!).
But you would also have to add more at the median or above to keep the median correct.
|
|
aju
Member of DD Central
Posts: 3,500
Likes: 924
|
Post by aju on Nov 1, 2019 12:34:30 GMT
I love hard sums!
The figures above are they real figures or are they just number pulled out of the hat to demonstrate the reasoning.
The most damning thing from my perspective is that Zopa does not show an accurate loanbook - if they show it at all - and even if they did surely we cannot even hope to corroborate or challenge their stats, can we?
Edit: I checked the loanbook and it seems I was wrong they have actually updated it since I was complaining a few months ago (I don't remember them getting back to me and saying they had fixed it after I wrote them twice once in feb and once in july). Still not sure if it would help though.
|
|
aju
Member of DD Central
Posts: 3,500
Likes: 924
|
Post by aju on Nov 2, 2019 0:08:37 GMT
aju The figures are very made up!
@greenwood2 It would be easy to add mode points to keep the median the same but skew mean returns lower.
My main point is the blog is just marketing fluff and unless Zopa shows the true distribution of returns it is totally meaningless.
I thought as much ...
|
|
|
|
Post by propman on Nov 4, 2019 13:51:43 GMT
These are not meaningless. Yes they could provide more information, but it does give an indication of the likely returns. Unless there was some manipulation via the algorithm, although theoretically you could have strange distributions giving these results, this is extremely unlikely. Although it would be good to get a methodology (ie what fees are included, the period the info is taken from and any excluded data).
|
|
aju
Member of DD Central
Posts: 3,500
Likes: 924
|
Post by aju on Nov 4, 2019 14:00:21 GMT
Lets hope their predictions/projections are not being performed/created by their software team who have been promising performance improvements for the best part of nearly 2 years and the whole web views thing seems to just get slower and slower everytime I log in. I use my own bookmarks of the relevant pages if I want to get something done quickly!. |
|
|
|
Post by fuzzyiceberg on Nov 7, 2019 11:09:25 GMT
This suggests that the variation in returns for Plus are similar to Core which I am surprised about. Does anyone know whether the matching algorithm aims to equalise actual returns or expected returns? I have always assumed the latter, but the former would reduce the variability from defaults and would mean anyone with below average performance would get better loans than the lucky ones! (When it was first introduced they did say that the averaging was over a relatively short period and so would not have made sence to include defaults, but I assume it has been amended since). Taking Zopa's figures at face value (as I think we should, no reason to think they would deliberately lie) then we can see that Zopa has failed to meet their target returns. Using Core (the arguments apply equally to plus) the median return is 0.2% lower than target. In Core 10% of investors missed target by more than 1.2%, achieving less than a 2.7% return. It seems to me that sort of distribution would be unlikely to occur if the algorithm was actively taking into account actual returns over the mid to long term. I suspect it simply allocates loans which average the target return over the short term. Of course if the 'target return' were less than the average of the expected return for the product there would be scope to minimise losers by allocating them more high return loans at the expense of the winners (ie reducing the deviation of the overall return distribution) but this comes at the cost of a seemingly less competitive product.
The real concern is that we don't really know how any of this works and Zopa seems to have carte blanche to change their credit vetting, markets, targets, algorithms etc at will and without any by your leave form investors.
I am minded of the well known advice that you should not invest in something you don't understand...
|
|
|
|
Post by propman on Nov 7, 2019 12:04:19 GMT
This suggests that the variation in returns for Plus are similar to Core which I am surprised about. Does anyone know whether the matching algorithm aims to equalise actual returns or expected returns? I have always assumed the latter, but the former would reduce the variability from defaults and would mean anyone with below average performance would get better loans than the lucky ones! (When it was first introduced they did say that the averaging was over a relatively short period and so would not have made sence to include defaults, but I assume it has been amended since). Taking Zopa's figures at face value (as I think we should, no reason to think they would deliberately lie) then we can see that Zopa has failed to meet their target returns. Using Core (the arguments apply equally to plus) the median return is 0.2% lower than target. In Core 10% of investors missed target by more than 1.2%, achieving less than a 2.7% return. It seems to me that sort of distribution would be unlikely to occur if the algorithm was actively taking into account actual returns over the mid to long term. I suspect it simply allocates loans which average the target return over the short term. Of course if the 'target return' were less than the average of the expected return for the product there would be scope to minimise losers by allocating them more high return loans at the expense of the winners (ie reducing the deviation of the overall return distribution) but this comes at the cost of a seemingly less competitive product.
The real concern is that we don't really know how any of this works and Zopa seems to have carte blanche to change their credit vetting, markets, targets, algorithms etc at will and without any by your leave form investors.
I am minded of the well known advice that you should not invest in something you don't understand... Many thanks for your thoughts. It is possible that the outliers are people who have stopped investing some time ago and therefore who cannot be effected by the current allocation algorithm, although I agree that the results do make this possibility less likely. I can't see that their results would be lowered by this approach. It would narrow the variation for the majority who are at least relending. Indeed it might be possible to uplift the median a little by helping those near it at the expense of those above.
|
|
zlb
Member of DD Central
Posts: 1,422
Likes: 333
|
Post by zlb on Nov 7, 2019 12:28:53 GMT
OK, any statisticians?
I still think poor returns are caused in part by the 1% diversification - where an investor doesn't realise until it is too late, that if they deposit 1x£10K that they are lending £100 per borrower - £95 defaults would then occur - how long would it need to be invested, for such high defaults to be equivalent to 10x£1k invested in to £10 loans?
Zopa have acknowledged that this is a problem - they have a long term promise to make diversification far more granular in some way.
I agree they could compensate the worst off with the wealthiest here - but isn't that what their bank is going to do anyway?
|
|
aju
Member of DD Central
Posts: 3,500
Likes: 924
|
Post by aju on Nov 7, 2019 14:33:49 GMT
OK, any statisticians? I still think poor returns are caused in part by the 1% diversification - where an investor doesn't realise until it is too late, that if they deposit 1x£10K that they are lending £100 per borrower - £95 defaults would then occur - how long would it need to be invested, for such high defaults to be equivalent to 10x£1k invested in to £10 loans? Zopa have acknowledged that this is a problem - they have a long term promise to make diversification far more granular in some way. I agree they could compensate the worst off with the wealthiest here - but isn't that what their bank is going to do anyway? I'm not sure you are completely right about diversification, we discovered a couple of years back that we could ensure £10 loans as a maximum just by limiting lending to £1999 per block. Trouble is lately we are getting many more defaults that seem to add up to nearly the same as the monthly interest. We have sold off in last 6 months and are waiting for the relend to catch up but its nail biting stuff some months as I have mentioned elsewhere. At some point though I believe statistically we should more to the middle of the average. Perhaps! Zopa as you say have been suggesting for quite a while, at least 2 years, they are looking into diversification of <1% but their main concerns has always been lending speeds being unacceptable for customers - I think they are also concerned the amount of additional data it will generate with their systems already baulking in the overload. That's another thing they keep promising a shiny new all singing, more efficient platform again for the last couple of years that we are still waiting for. I think the comment a few posts back - I think it was this thread, regarding their engine balancing lending with the target interest rather than the expected interest rates after defaults says it all. But the comment below is suggesting its more measured - I'm not sure if this is referencing the original methods though as there are lots of changes but none of the faq's are dated as far as I can tell..
|
|
