In this article, I will outline the proper framework for building an effective risk management department within a lending organization.
The primary goals of building any credit granting operation are to perform:
Ongoing warehousing (and expansion) of your data variables.
Characteristic analysis to identify top variables, cancer segments and opportunity segments.
Implementation of top variables into effective criteria rules.
Implementation of custom scoring models to complement the criteria rules
Ongoing portfolio analysis to identify new cancer and opportunity segments.
Ongoing model validations to test need for replacement models.
Building a culture of risk management in any organization is best done with preconceived notions of the structure you intend to establish. In my experience, the best structure is one with top down acknowledgement of the following:
There is a big difference between your "model" and where you set your caps. You can build an incredibly powerful credit decision model, but if you set your caps too loose, excessive credit losses will follow.
The goal is clear: build the most powerful model possible given the data you have to work with, and then set the caps to produce the desired losses you are targeting. You only get so many applications through the door — do not waste them. The stronger your model, the more you will be able to approve while keeping defaults at desired levels.
Don't fall into the trap of thinking "volume" can somehow control losses. Too many lenders make the mistake of thinking if they continuously grow in volume, they will outrun their defaults. This way of thinking is cancerous! Volume cannot mask defaults for very long. Vintage static pool analysis will always reveal what's really going on.
Your "Model"
By your "model," I'm referring to the overall program that determines which applications to approve or decline. A very simple model could be: Approve all apps with FICO score > X and decline all the rest. A more comprehensive, and far better, model might be a hybrid of criteria rules and multiple internal scores all working together to determine which apps to approve and decline.
I like to think of our overall "model" as a simple pencil. And your model developer is the pencil sharpener. How sharp your pencil is depends partly on your model developer's technique, but a whole lot more on how much data your model developer has to work with. The best way to sharpen your pencil is to expand the number of variables in your possession.
There are various ways to determine the overall power of your model, and all of them focus on how well your model does at identifying/separating good performing loans from bad ones within a population. This "separation" can be measured, and it's how you know how sharp your pencil is. Where you set the caps is an entirely different concept than how sharp your pencil is. I would urge all C-Level executives to understand this important fact.
Net Profitability
How are we to know just how high our losses should be? On one hand, I suppose lower losses means more profits, right? So why not just set the caps super tight and the losses will be super low?
The answer is obvious — because you'll only be approving a tiny number of loans and you won't be able to grow your portfolio. But how many defaults are too many defaults?
This is the million dollar question. Getting it wrong is the leading cause of death for so many lending institutions. First, "losses" refers to the principal dollars lost when default (charge-off) occurs. For example, if you loan $10,000 and the customer pays back $7,000, then stops making payments (i.e., defaults), the principal lost is $3,000, or 30 percent.
In the graph below there are four different scenarios for modeling net profit using an industry standard formula.
Each lending organization should input their own "targets" in each of the rows. As you can see, while the framework is the same, it can be dialed-in in many different ways. As the portfolio's risk-profile increases (left to right), so does the volatility of achieving our targets. And perhaps the most volatile of all are the annualized losses, which speaks to the importance of having a robust risk management framework in place within your organization.
Controlling Losses
We've already acknowledged the importance of having a model in place that does a good job of separating goods from bads (i.e., a "sharp" pencil). I think it's wise at this stage to outline the three pillars of what impacts loan performance:
You Loan Originations "Model"
Your collections Department's Effectiveness
The Economy
Your Loan Originations "Model"
Here's an example of a weak originations model: Apps come through the door and go into a queue. A credit analyst pulls a credit report and makes a human assessment of the overall risk. Approvals and declines are issued manually and everyone hopes for the best.
A better approach: Apps come through the door and pass through a robust auto-decisioning program, built with custom scores and custom criteria rules proven to be good at separating goods from bads.
An estimate of "likelihood of default" based on the overall data profile is calculated. Approvals and declines are issued accordingly and instantly.
Such a system is built for speed, effectiveness and scale. Decisions are consistent, and performance is measured over time with continual improvements being made to the decision program based on identifying various cancer segments (high loss segments), and opportunity segments (low loss segments) in the resulting portfolio over time.
Cancer segments should get declined, opportunity segments incentivized with pricing discounts, and so on, to generate a higher proportion of those types of customers in your future portfolio.
Your Collections Department's Effectiveness
The effectiveness of one's collections department is paramount to controlling losses. As you go further into the subprime realm, it becomes more and more crucial to implement proactive strategies. In the deep subprime space, most customers simply won't make a payment unless someone calls them and asks them to pay.
An ill-equipped collections department is one mainly focused on customer service type calls (incoming calls) with very little reporting in place that measures staff-level performance. Data in the loan management software is either dirty (inconsistent) or lacking altogether.
A well-equipped collections department is one where good data and staff-level measurement is the main focus. Daily goals are assigned to each collector with commissions tied to reducing delinquency and collecting payments. As accounts move from DQ bucket to deeper buckets, roll rates are measured and monitored for adverse trends. Cyclical trends, as well as number of collection days each month, are well-known and factored in the monthly goals. Multiple payment channels exist, providing a variety of ways customers can make payments. And behavioral scores are used to identify accounts most likely to either cure or roll worse. Higher volumes of the toughest accounts are assigned to the best collectors based on results in prior months.
The Economy
You have no control of the economy, of course, but you should monitor where it is headed since defaults can swing pretty wide in response. The best you can do is be rock-solid at your loan originations "model" and collections department's effectiveness, then bake in a little buffer in your annualized net loss estimates.
Being good as a lender is not always common sense, and not something that simply requires great software. It really takes learning and know-how. There's an interesting catchphrase in the risk management field that's really quite telling: "Do you know, or do you think you know?"
If I were to ask you what your most predictive variables are, do you really know? Perhaps down payment? Term? Miles? FICO score?
Are you sure?
What if I said you should really have a list of your top 100 most powerful predictors readily available and sorted from strongest to weakest based on each variable's precise ability to separate good loans from bad loans?
I recently consulted a BHPH dealer who had a rising problem with CNLs, to the extent they were in default with their funding source. Over the past two years they had been implementing changes to characteristics like max loan term, max loan amount, and a few other items they thought would set things right.
I conducted a simple characteristic analysis and revealed that while they had tightened their averages in those areas, there were more important characteristics that carried even more power, with averages that were trending in a negative direction.
The net result was an overall further worsening of a portfolio that was already struggling.
They thought they knew what to tighten. It was common sense, right? But they ignored characteristics that had even more correlation to causing defaults.