Crowdlending is one of the newer asset classes available to individual investors. Crowdlending enables individual investors to directly loan money to other individuals and small/mid-size companies. The basic idea is that one investor can give partial loans to a large number of borrowers and thus build a broadly diversified loan portfolio.

One of the pioneer platforms, and the biggest, is LendingClub. In December 2017, it had loans active with combined worth of over $8 billion. It has been existing since 2007, has over 1.5 million investors and over $28 billion borrowed historically. It is also a publicly listed company in New York stock exchange with market capitalization of almost $2 billion. So it is already a well-established company rather than just a startup company.

The basic idea is that Lending Club gives loans to individuals for a period of 36 or 60 months. The platform does largely automated, algorithm-based due diligence on each borrower and gives him borrower grade from A to G. A is the least risky grade so the borrower gets a lower interest rate for his loan. G is the most risky grade so borrower needs to pay high interest rate for the loan. Investors can then select on which loans they participate and with which amount. Minimum investment per loan is only $25. If there is a default, Lending Club tries to collect the money from the borrower but in most cases this can be done only partially and therefore the investor may face losses for individual loans. It is therefore good idea to diversify and invest in at least >200 individual loans.

What is a nice feature for a number oriented person such as myself is that Lending Club publishes their full historical loan data in a huge data file. This data file can then be read to a database for further analysis. So I analysed the historical loan data and came into the following return and risk numbers since the start of the data in July 2007. The numbers are after defaults and possible recovery of bad debt:

Grade | Return p.a. | Stdev p.a. | Max Drawdown |

A | 4.7% | 0.4% | 0.0% |

B | 6.0% | 0.9% | -0.4% |

C | 6.2% | 1.2% | -1.0% |

D | 5.8% | 1.9% | -4.8% |

E | 4.9% | 2.0% | -3.1% |

FG | 3.4% | 3.0% | -13.7% |

A-C | 5.6% | 0.8% | -0.5% |

How about correlation to large cap equities?

Grade | Correlation |

A | -0.04 |

B | 0.14 |

C | 0.13 |

D | 0.28 |

E | 0.12 |

FG | 0.09 |

A-C | 0.16 |

However there are three issues with these past returns that should not be forgotten:

- These returns include all loans ever made by Lending Club. As an individual investor it is not possible to have so diversified loan portfolios. Investing into less loans does not necessarily decrease the returns but it would increase the volatility of the investment.
- The maximum drawdowns are from 2008 when the platform had just been founded and there were not so many loans available as currently. Furthermore, there are some unclarities with the performance of the loans in 2008. This means that investing may be more risky than what is indicated by the past return numbers.
- The returns are not normally distributed.

Let’s now add crowdlending as new asset class to Adaquant Asset Allocation Suite. Open from the menu Tools / Asset Classes / Manage. Click the yellow plus button to create a new asset class. You see now the following window:

Write “Crowdlending” as name of the asset class and select “High-Yield Debt” as main asset class.

Next you need to set the price data for the new asset class. There are two possibilities.

First, by pressing “Edit Time Series”, you can manually set monthly prices to the new asset class. You can press “Copy” button and paste the data to Excel and modify it there. Once you are ready, select and copy the data in Excel (all three columns with the headers) and press “Paste” button in the tool. The problem with this approach is that you need to have all monthly data for the new asset class since December 1997 until the last month of data in the database. This is not the case with Lending Club data since it only exists since mid-2007.

Second alternative is to press “Set Time Series” button. You need to know the return and standard deviation of the new asset class, and you can optionally set the correlation to other asset classes if there is any. We have all the data we need in the above tables. Let’s say you plan to invest into low risk categories A-C. Your expected return is 5.6% and standard deviation 0.8% per annum. There is also 0.16 correlation to Large Cap Equities. So your settings are:

Press “Generate Series” button. You now have time series for the new asset class:

If you press the “Set Time Series” button again and repeat the generation of time series with the same parameters, the results would look different i.e. the monthly returns are not the same. What is however common is that the new time series would have the same long-term return, standard deviation and correlation to Large Cap Equities as the old one. And this is all that matters. The way the asset allocation simulations are done in Adaquant Asset Allocation Suite, by making set of sample returns from the monthly asset class returns, means that anyway the monthly returns will be mixed in the simulation phase.

The only problem with this “Set Time Series” approach is that currently it only supports normally distributed returns. But as written earlier, the returns in investing to Lending Club loans would not have been normally distributed:

If you compare the return distribution (blue bars) to the normal distribution (yellow area), you can see that the returns peak in the positive side (=high kurtosis) and have more negative returns than indicated by the normal distribution (=negative skewness). In other words this means that if you use normal distribution to model Lending Club returns, you would underestimate the risk and overestimate the frequency of high positive returns.

Unfortunately Adaquant Asset Allocation Suite “Set Time Series” functionality supports only normally distributed returns. What I do next is to create a better return model by utilizing Matlab (mathematical software) and then pasting the results to Adaquant Asset Allocation Suite.

I take the borrower grade A-C monthly returns between July 2007 – September 2017 as base. The mean return of these returns is 0.45%, standard deviation 0.22%, Pearson skewness -0.73 and kurtosis 3.69. With these distribution characteristics I create a random return series in Matlab for 234 months (number of months between December 1997 and June 2017). Then I adjust these random returns so that the correlation of the returns with S&P 500 index is 0.16. The distribution of these returns looks like this:

The return series can now be copied to Suite. Here is link to the Excel that contains the price series of the returns: crowdlending_returns.xls. You need to edit the asset class id to correspond to the one open in the Suite. Then copy all cells with values (including headers) and press Paste button in suite. The asset class window will then look like this:

Now that you have created the new asset class, you can use it in the simulations as well as set return, standard deviation and cost estimates for it.

Note that setting estimate for return and standard deviation does not change the shape of the return distribution. This means that for example, if you invest into very high return but high risk crowdlending investment, then you can set the estimate e.g. to 9% p.a. and standard deviation to 15% p.a., but you don’t need to create another asset class for this.