Predictions with machine learning: Credit Corp's Consumer Lending11 November 2018
Four years ago, Credit Corp Group (ASX:CCP) was a difficult company to hold. Although the core Aus/NZ debt purchasing business was doing well, there were significant headwinds. Increased competition was causing prices to rise and returns to decline. Also being the largest player in a mature market, one could also foresee that growth would likely become supply constrained.
The Consumer Lending business was something that was started organically just two years prior. While growing its loan book rapidly, it was having a negative contribution to net profits. At the time I suspected it was more of a case of management optimising cashflow over profits, rather than an unprofitable business that was struggling.
I wrote on 2014-08-171:
The best part is that it contributed -$2.5m NPAT to the FY14 results due to the upfront provisioning (20-25% of value) of the loans upon establishment. As a result of this provisioning model, we're unlikely to see the real contributions to the reported NPAT until 2 years after the loan book stops growing (average term is around 1-3 years).
Assuming that Credit Corp can make a 16.5% annual NPAT return on the net loan book. If we conservatively estimate the net loanbook will hit $100m (gross ~$125m) in FY16 and flatline from there; Provided everything else stays the same, by FY18 CCP would have added a further $16.5m to reach an annual NPAT of $51.3m. A 47.4% increase from FY14's NPAT or 10.2% annually over the next 4 years.
My thoughts at the time provided the resolve to hold tight in a period when the growth story appeared over. Through the passage of time, this has been proven correct. The segment posted a NPAT of $16.1m for FY18 and has become a significant contributor to the bottom line.
A couple of years ago, I became interested in predicting the profit of the Consumer Lending division, with just the "net lending" guidance provided by management. A method that only involved historic data, and didn't require projections or a deep understanding of ROA, ROE, expected losses, etc.
Recently, I've turned my manual methods of creating the profit equation into a computer program. One that uses machine learning libraries to complete linear regression model. Code and data are available at the link below2.
Note: Machine learning models can be wildly inaccurate at predicting the future, especially with the limited data points provided. It's not going to be able to take into account product mix changes, tweaks to gross margins, variations in expected/actual losses, etc. Please take all the calculations with a large grain of salt
FY19 NPAT Prediction
With an assumed average gross loan book for the period of $191.5 and net lending of $50m (management guidance: $45-50m), I'm predicting an NPAT of $18.5m, which is at the top end of management guidance's of $17-19m.
FY19 NPAT Run-rate
The faster the business grows its loan book, the more negatively the NPAT will be impacted. This is due to the upfront provisioning of expected losses (~20%) that occur. By reducing the net lending to "maintenance level", we get an idea of what the actual NPAT run-rate is for the business.
I'm estimating the "maintenance level" for net lending to be at $31.2m. Plugging this back into the model gives a FY19 NPAT run-rate of $21.9m. This is 18% above the reported NPAT for the segment and is proportional to 7% of the group's NPAT.
Full calculations and assumptions are also detailed in the Jupyter Notebook2.
With management teams that prioritise cashflow over reported profits, often the true financial health of the company is understated. I personally view this as a margin of safety and a bit of an extra bonus. However there are times where the gap can become quite substantial which becomes an opportunity.
Similar "maintenance level" calculations can be applied to the larger PDL segment of the Credit Corp. This might be a good topic for another day.
CCP last traded at $19.12.
Disclosure: At the time of publishing I own shares in ASX:CCP.