Overview of P2P Automation and Analytics Sites – Part 2

There is an ever increasing number of websites dedicated to aiding in analytics and investing in Lending Club and Prosper. In part two of this comprehensive two-part series (click here for Part 1) we will look at all of the popular tools that investors use today (there are ten in all) in order to backtest strategies, automate their investments and analyze data. Analytics and data has been a large part of p2p lending from the start and investors are always looking for an advantage to potentially maximize returns. Investment automation allows for a much better investment experience versus selecting notes on the platform by hand.

BlueVestment Logo

BlueVestment specializes in automation strictly for LendingClub. Through BlueVestment, users are able to automate investing in notes using credit models from P2P-Picks. P2P-Picks has created their own models based on historic data from Lending Club and Prosper. Investing using either the Profit Maxmizer or Loss Minimizer models is a hands off method of investing in Lending Club. As new data becomes available, the models are updated accordingly to achieve the results of the respective model.

Users can also opt to create their own basic filter criteria using the 22 attributes available to them.  In addition, users can create advanced filters using the node builder.  This allows for a customized investing experience with different operators as shown below.

BlueVestment Node Builder
Using Note builder to select criteria.
BlueVestment Advanced Filter
Example of an advanced loan filter.

BlueVestment also has a high level dashboard which includes investment amount, available loan distribution by grade and total loan availability by grade. BlueVestment is a completely free service.

PeerTrader Logo

PeerTrader currently offers automation and charting specifically for Prosper with Lending Club integration coming soon.  They are currently in open beta and will be free for anyone to use until they officially launch.  As of 2015, they are now an SEC Registered Investment Adviser which makes them one of two public retail tools that are an RIA.  PeerTrader has just completed their integration with P2P-Picks, which is publicly available when you create a PeerTrader account.

PeerTrader also offers streaming charts which are interesting to watch during the release times. Among the charts, you can see Total Investable Volume and Number of Active Listings as shown in the screenshot below.

PeerTrader Realtime Charting
Click to enlarge

The screenshot below shows an example of how to setup an auto-invest with PeerTrader.  They currently offer 19 filters to customize your auto invest.

PeerTrader AutoInvest
Setting up a PeerTrader filter.

 

P2PQuantLogo

PeerToPeerQuant’s main offering is a genetic algorithm credit model for Lending Club.  The model was created with the goal to determine which borrowers will pay the most to investors.  A genetic algorithm is built by inputting Lending Club’s historical data.  The computer program then uses this information to create investment strategies with anticipated returns.  Strategies can then be combined or other criteria can be added to create better results.  With an average portfolio age of 6 months, stated returns are 11.1% when calculated using XIRR. Notes using their model can be purchased by clicking “Invest” on the notes on their website. After the first free 5 investments every month, the cost is $0.15 per click.

P2PQuant Note Picks
Sample of PeerToPeer Quant’s buy notes page.

Besides being able to purchase notes, PeerToPeer Quant offers a Percentile Calculator. This uses the data from Lending Club’s “Understanding Your Returns” page and compares your results with others.  Inputting my data from my primary notes yields the below results.

P2PQuant Percentile Calculator Results
Results for ANAR: 12.13%, WAIR: 17.73%, Average Age: 10.8 months.

The above results show that compared to all investors, my account is performing pretty well. However, compared to those who have the same average age and weighted average interest rate, I am only in the 68th percentile.  The calculator is a free tool to use and doesn’t require a PeerToPeer Quant account.

Notient Logo

Notient is a relatively new player in the P2P lending automation and analytics space. They offer a dashboard which includes account activity, interest to date as well as portfolio breakdown. The daily charting is quite interesting, especially when gathered over a long period of time.  Below you can see my account activity since I opened my Notient account.

Notient Account Activity

To get started with automation, Notient offers several pre-created strategies. For instance, they have included strategies from Peter’s post on how he is investing in Lending Club and Prosper. Users are also able to create a custom strategy using fourteen attributes. From there, strategies can be assigned to auto-invest.

Notient - Find a Strategy

Notient is currently in beta and is not charging to use the platform.  Looking forward, they plan to offer automated note sales and support other platforms besides Lending Club and Prosper.

InterestRadar Logo

InterestRadar has been around in the Lending Club automation space for quite some time and offers many different features.  As a registered user, there are several public strategies that are available to choose from which aim for 10% returns. Users can edit these strategies if desired or create a completely custom filter strategy.  Interest Radar has two built in scoring models to predict default which can also be used when backtesting.  Both models are computed for each loan grade.

Besides investing, Interest Radar also offers an auto sell feature.  This can be setup based on formulas with specific variables as well as the FICO trend.  Since selling notes is a very manual process, auto sell could be useful for users who list many of their notes for sale or those who try to get rid of notes that they believe will be charged off.

By uploading your Lending Club portfolio. you have access to all sorts of insights. One of the options is “Risk”, which shows non-performing loans, loan concentration, score estimated risk and payment processing delay.  You can see below that I have several notes that are deemed riskier with Interest Radar’s IR01 score (Scores range from 330 – high risk of default to 409 – lower risk of default).

IR Score Estimated Risk

A subscription to Interest Radar is $9.99/month or $59.99/year, but there is a free 30 day trial offered.

P2P Automation and Analytics Sites – Wrap Up

It is great to be a P2P investor with so many tools available. I have only scratched the surface on the functionality that these tools offer and new features will only be added as tools mature. I encourage investors to explore the wide variety of tools available and keep an eye on new ones entering this space.

What tools do you prefer and use regularly? Let us know in the comments!

Subscribe
Notify of
10 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Jacob
Feb. 18, 2015 5:19 pm

Peter,

To me, the success of various analytics engines and services designed to outperform Lending Club or Prosper’s own risk assessment basically hinges on how effective Lending Club and Prosper are themselves on identifying risk arbitrage and updating their risk scores. Rennault’s comment with you in your interview pretty much hits the nail on the head:

“In general, we also have a lot more people working on risk management, and portfolio management than any single investor, and our people are not necessarily dumber than anybody else. I think the general point is it’s going to be hard to consistently outperform as a platform, because when there is an arbitrage opportunity, we see it as well…

I think over time, considering we have the data, we have the manpower, and we have the desire to continue to make the risk ranking as efficient as possible, I think it’s going to be hard for any single investor to consistently find these arbitrage opportunities that generally when they appear, they pretty quickly go away.”

My question to you Peter is this: do you think 3rd-party analytics engines and services will be able to offer arbitrage advantages above and beyond simply enrolling in Lending Club and Prosper’s own automated investing note grades without any sort of filter?

James Wood
James Wood
Feb. 18, 2015 8:23 pm

Right now, if you back test the data you can find significant differences in outcomes when testing on just a single factor such as income level, type of loan and state the loan is in. This has been the case for some time. While things may change, so far they have not.

The LC pitch for their own automated investing seems to be that it all works out in the wash. This pitch makes sense for their business model as they will do best if all, or at least most loans, are funded. However, for those of us paying attention, in the past and at least for now, the math is saying all loans at the same interest rate are NOT the same. The rate that certain loans sell out tells the same story. So do the returns on my three accounts – I could be very lucky but I doubt it is all due to luck.

James Wood

Jacob
Feb. 18, 2015 10:30 pm
Reply to  James Wood

I tend to agree with you James. That is why I still have filters set up. The question becomes how long this “arbitrage” can last. Mr. LaPlanche’s comments seem to indicate that they have data scientists hard at work and are constantly revising their lending models. So while back testing may seem to suggest profitable strategies, you can’t go back 3-5 years and pick a portfolio with those criteria. We have to go forward. If Mr. LaPlanche’s comments are to be believed, they have already adjusted their proprietary credit model (or are adjusting it quickly) so that the next 3-5 years won’t see the same holes that are revealed via back testing. I wonder if this is true.

In the end, this is somewhat of a philosophical discussion. Picking individual notes by algorithms or data mining might be similar to the difference between buying a broad market, low cost mutual fund and trying to pick individual stocks based on analyst insights. For stocks, the studies I’ve read seem to indicate that active management yields no better results than, say, a Vanguard low cost fund over the long haul. In the early days of P2P, maybe arbitrage did exist. Maybe it still does. The question is, how much of out performance is statistically significant and how much is dumb luck. There may very well be market beating strategies, similar to how having stocks that have high dividends tend to outperform no dividend stocks. Time will tell of course.

RawRaw
RawRaw
Feb. 20, 2015 8:48 am
Reply to  Jacob

I wouldn’t expect it to last very long, unless you are using predictive data that LendingClub is not. Perhaps that is private data sets or local economic conditions or something else.

JB
JB
Feb. 24, 2015 8:07 am

I think Jacob has brought up an interesting viewpoint, well thought out and coherent. Something I hadn’t thought of before. Well done!

James Wood
James Wood
Feb. 24, 2015 9:03 am

I do agree that the target is a moving one. Some of the movement may be from LC and others changing how they rate loans in order to even out returns. While they may be making such changes and things could change overnight, I heard the same “pitch” over a year and a half ago and am not seeing major movement on their side, at least not so far. I do think it is in their best interest to talk up their automatic investing tools and use it to buy loans that aren’t getting purchased immediately but I’m not yet convinced that they will buy the most in demand loans right when they hit the market.

Another game changer is us. The more time in, the more we all refine our methods, the m ore our filters pick out the same criteria and the more competition there is between us all for the “best” loans the more access to top loans will change. Same thing as big money and professionals moves into this space.

The more automated investors get, the faster the best loans will sell out. I’m still hand picking loans with filters right after they are released. It is clear the the right tail of the bell curve, and many of you here, have automated that process outside of LC tools and are out competing me on speed. However, I can still find enough qualified loans to continue cherry picking what has historically been great performers.

James

PS I view 3 to 5 years as a near eternity with any new Tech related product or service. Part of how you frame this thread depends on your view of time frames. I have a long term view of peer to peer lending but a short term view of the current algorithms and especially the competition. Right now, most changes seem to be from folks like yall and big money moving in.

Jacob
Feb. 24, 2015 9:49 am

“The more automated investors get, the faster the best loans will sell out”.

The presumption being, of course, that we (collectively) know which loans are the best loans beforehand and will go after those loans preferentially. I’m not so sure we do know which loans are the best, or that we know better than LC or Prosper. I like to think I do, but I could just be fooling myself.

Before running my own back testing on Nickelsteamroller, I thought I had a pretty good intuitive idea about what might constitute a good loan. I was amazed at how far off my intuition was. Several criteria that I thought would be important didn’t seem to make a lick of difference. For now, my filters only concentrate on income amount (the higher, the better), credit inquiries (zero), and specific loan categories (I filter out just a few categories). These seem to be winning strategies, but the devil is in the details.

Even if one plugs in these filters and confirms that over the past several years a certain filter set resulted in a higher than average return (due to a lower default rate), it takes complex statistical analysis to be able to determine the probability that such filter criteria are actually statistically significant. And as always, past performance is not necessarily indicative of future results. After all, one individual’s winning formula could simply be luck. For a good discussion on how we humans are prone to make patterns where none exist, I recommend the book “Fooled By Randomness”.

James Wood
James Wood
Feb. 24, 2015 11:56 am

I admit that some of my initial intuition was also off. Most wasn’t but a few things I thought to be so, especially in the category of reason for the loan, simply were not what I thought them to be. For example, I thought taking on debt to get married was about the dumbest idea out there but it isn’t one of the worst categories.

And I missed that the state of Nevada would be such a poor performer In retrospect it makes perfect sense and seems obvious. Hindsight is like that.

But that is why we develop models, back test them, revise them, back test them again, apply them, monitor the results, and pay attention to what is going on around us, etc. What ever you are doing, being aware of randomness and small sample size (using a lot of filters that yield great back test results but pick out only a few loans can be dangerous) is a good thing. I get that.

If you want a real test, all you have to do is apply the models developed by back testing, tea leaves, or whatever to future purchases, real or virtual. Purchases where you don’t already know the answer is where the rubber hits the road.

The statistics can be as complex. Or not. You can ask yourself if you are doing better than average, or not. You can ask yourself how much better than average, or not, you are doing. And you can replicate this many times with different models in different portfolios.

At this point I’l mention two things. One is that I’m a scientist and am more comfortable with statistics than most people and two, I’ve not actually run a formal statistical tests to see if my results are significant. I have run loads or regressions to back test ideas.

I’d like to run more stats but I don’t have the data. Besides, I don’t think I need to as I can read the graphs. Graphs don’t generate a p value but they do tell you what is going on.

Graphing is another excellent way to look at the data. LC allows us to do that for each account. Comparing vs other accounts that have 100 or more loans and vs others that are within the tightest interest rate band similar to mine, I’m above average.

Of course that is just one data point and there was a 50% chance of being above average so that doesn’t carry any weight. But lets look at how much it is above average, it is over 3% above average. Well, you say, it had to be somewhere and one data point, again, is useless statistically. Fair enough.. However, I have three different accounts and they are all 3% or better than the average. That does tell me something. Flipping a quarter 3 times in a row and getting all heads isn’t that rare, it will happen 12.5% of the time. But that isn’t a good comparison as 3% or better isn’t just a 50/50 comparison of better or worse than average. It is three results that are a lot better than average. Eye balling the data points, 3% or more is VERY conservatively a standard deviation above average. That three times in a row is going to be significant.

Still not convinced? Lets break it down further. My three accounts have 20 different portfolios in them. These are combinations of different strategies and/or were purchased at a different time period. Most of these are my hand done picks and and a few are LC automated picks using my criteria. So now we have a sample size of 20 but some don’t have many loans (for example, one has just 3 loans and is the I forgot what model I was using when I bought this loan folder) in them so if we combine the smallest 5 together that is a sample size of 16. I don’t have the tools to run a test at this level but the data look like if I did, I’d suspect I’d get 16 above average results. They won’t all be 3% above average but it does seem like what most of us are doing is working, repeatedly.

Yes, most of the above is qualitative, especially since I don’t have a way to quantitatively compare my results to similar results and generate an average and standard deviation vs similar loans. Nor do I have the tools to quantitatively look at my portfolios vs other portfolios that are similar. Despite this, the graph that compares my accounts to other similar accounts tells a story that what I’m doing, at least so far, matters quite a bit. When I drill down to the portfolio level , and look at different models, some of which are quite different and range from my first steps just copying Peter’s basic model to developing my own, that they all seem to be out preforming the average,

Yes, it could all be luck, but every time I create a new portfolio and test a new model vs reality and continue to outperform, the argument that the results are just due to luck becomes weaker and weaker. As a scientist I know that the possibility that it is all due to luck will never fully go away, but from where I sit it seems to be less and less likely every time I repeat the process and produce a similar result.

If you want to buy loans from people in Nevada and other bottom 20% states, that have used over 90% of their credit limit, have been working in a low paying job for less than a year and want an unsecured loan for money for a car, you can do so. I’m willing to bet that moving forward at lest in the short term, the model I just proposed will continue to under-preform.

I recommend the book “measuring behavior” for a good applied introduction to stats and the various biases we all can fool ourselves with. While this book is on animal behavior, the explanations of stats are very well done and understandable.

James Wood
James Wood
Feb. 24, 2015 12:05 pm

“I’m not so sure we do know which loans are the best, or that we know better than LC or Prosper. I like to think I do, but I could just be fooling myself”

I like to think we do as well. And I also like to think that when large amounts of data are available we can crowd source better results faster than when they are head behind closed doors by brokers that don’t always have our best interests at heart, are not generally paid for performance and may not be as bright as they think they are. Of course, I may have just revealed a lot of my own biases.

The proof is in the pudding. Can you or can’t you fairly reliably produce models that out preform the average moving forward. I think I can. And I think you can too.