Taking Estimated ROI to the Next Level

I had my first defaults start to come in the last 30 days.  What’s interesting about this is how the platforms calculate the ROI…  to describe it in one word:  “jumpy”.  Meaning, you can have a ROI that varies by more than 1% once a default happens.  If your ROI was to be plotted with respect to time it would like a saw blade. I believe with the tools (LendStats, Nickel Steamroller, et al) we might be able to get the plotted ROI to be more like a wave, and therefore much more accurate.

My main focus will be on Lending Club since they group everything that is 30-120 days late into one late group while Prosper has a status for 15 days late and every single month thereafter.

The LendStats method as you know will penalize late loans but it will only happen at discrete times (In Grace, 15-30, 30-120, and default) – admittedly Prosper is more granular.  I believe that a default loss curve could be constructed to give you a daily ROI value that would perform something like the following – “this note is 71 days late which means you are probably going to lose X % of remaining principal” (the 71 is just an example).

This information is impossible calculate with the Lending Club export as it. Only one new piece of information would be needed to open this new world of delinquency richness: days past delinquency.

Every late note in your portfolio has a level of “lateness”. A note that is 31 days late and 119 days late are both grouped into the same level loss but they both have different probable losses.   Prosper exports daily status payment history in great detail, which could allow for this to be constructed.  I’m going to explore this further and see what it will take to convert the platform to this varied approach. I believe it will provide much more accurate ROIs for the portfolio analyzer.


16 thoughts on “Taking Estimated ROI to the Next Level

    1. This would be for the portfolio analyzer on this site. It would prevent a situation where you analyze one day and have a 12% ROI and then run it the next day and then have an 11% ROI. In the situation I described it would be more like 11.2% on the day preceding 11%.  Maybe people won’t care, but I thought it would be nice to have the best possible ROI estimation.

      1. Ah, like some kind of “predictor” that you can plug your portfolio into, it analyzes each loan it and looks at, including its Payment History, and tells you what the chances it’s going to default are. (LendStats has some of that information, but I don’t think they have Payment History.)

        BTW, L.C.’s “Lateness” calculation can be way messed up. For example, here’s a loan that (according to Lending Club) hasn’t made a payment since 7/15/09.  Looking at the Payment History, how accurate would you say that is?


        (I use Accrued Interest when it’s available, but for whatever reason L.C. doesn’t make that information available on every page.)

        1. That note is on a payment plan. I think adding the days past delinquent or even better the payment history would add a lot of value to analysis. Prosper wins hands down here. They’re as transparent as glass with the data they provide.

          1. In fact (according to its status) it is no longer on a Payment Plan.

            Even then, I’m not sure why that should make any difference.

            (How about the “Payments to Date (5)” field — does that make any sense?)

  1. I think all this concern over calculating instantaneous and accurate ROI/NAR is antithetical to the this type of investment. It is not well-suited for these measures, which can only be accurately calculated after the resolution of the loan/portfolio.

    If you want continuous feedback on your picks, play the stock market or something with high liquidity. This ain’t it.

    1. I mostly disagree. I generally agree that any investor should not be watching their account daily, be it bonds or equities.

      However, more clarity into your returns is always a good thing. The point of these tools is to help you tune your selection going forward as well as mitigate potential issues with your portfolio. You should sign up for my webinar, I’m always interested to hear people’s opinion to improve and make the platform more useful. If you think it’s a waste I’d like to hear more. Thanks.

      1. Oh, I do think, however, that one could use the stochastic model to take a portfolio and calculate a 95% confidence interval for the returns using statistical models created with the past data. That might be a little more interesting (and realistic). People would be surprised how much variance there is even for loans with “good” characteristics.

        1. I actually like this idea a lot. I’m sort of doing that now with this here: http://www.nickelsteamroller.com/lendingclub_invest by publishing the estimated returns. I am sure there are more elaborate ways to do this, but I do not yet have the skills to do that.  I constantly work to provide better, more meaningful data. Everything is still very much a work in progress.

        2. I thought about this some more. All the returns in P2P lending are predicated in reinvestment of all repayment. I’m thinking future projections are almost pointless because models would have no way to figure out what notes you are going to buy. The best you could do is say “if you never buy another note your ROI would be X in Y months…” This would not be reflective of how people truly invest with P2P.  Thoughts?

          1. Partly the reason why I find ROI conversations boring. One has to make all sorts of assumptions and guesses to produce a number. It’s at best murky and at worst a distraction from the real issue.

            The real issue is how to best weigh risk versus reward and pick the notes that have the highest expected return. The portfolio only serves to reduce the variance in the expected return of a bunch of notes.

Leave a Reply