Clicky

X

Timing your loan portfolio sale using predictive analytics

LoIn our last article, we introduced a few statistical tools commonly used to predict the timing of future events. In this article, we extend the discussion of time-to-event to predicting the timing of account transition for maximum profit. To fuel the discussion, we reach beyond the common toolset and offer a novel approach. The approach we will introduce offers a synergy between different areas of statistics to find factors that lie beyond your data that drive your business.

The prediction of time-to-event has many useful and profitable implications for SMART BANKING.

Value creation:

  • Predicting the life-stage of a customer based on transactional history
  • Identifying of the ‘stages’ of the evolution of a customer relationship in order to maximize lifetime value (CLV)
  • Cross-sell and up-sell customized products and services based on life-stage and CLV
  • Understanding and predicting the estimated business value of a new customer

Risk management:

  • Predicting, if and when an account is going into default
  • The ability to simulate various ‘what if’ scenarios of loan origination vs. attrition
  • Insight into how different products are affected by macro-economic events across product offerings
  • Knowing when to buy and when to sell a loan portfolio
  • Development of action plans by account type to effectively extend time-to-event.

A little about data, statistics and real life

The performance of most statistical algorithms come with varying levels of success.  The reason is, most algorithms rely only on actual data; not possessing the ability to read between the lines.  Often, this is where the solution is.  Finding the solution is having the ability to see beneath the surface of the data.

Data in its simplest form is the recording of the events that characterize our daily lives.  There is much hype in analytic circles about data.  Data, data, data… we overhear the conversations every day. Massive data collection companies make fortunes capitalizing on the collection and sale of data.  The truth about data is that all data is good but most data is irrelevant.

We like to think of data like a pizza. Even a bad pizza still tastes good.

With all of this conversation about data, let us not get trapped into the perception that more data is better. Let us not subscribe to the notion of throwing data at a problem until the problem goes away.  It is not the size of the model we build or the amount of data that we use, it is the quality of the results that matter most.

Another myth I would personally like to extinguish is that a statistical model can show causation, this is a major misconception, but more on that in our next article.

Back to the problem at hand

The science of statistics is evolving.  New methods are constantly being developed and mathematically proven to find more information from data.  Subscribing to the pursuit of perfection in analytics we introduce a novel approach to the prediction of time-to-event as it relates to the management of the loan portfolio.  The focus of our discussion will be the timing of the acquisition and sale of loan accounts to reduce risk.

To accomplish our goal, we marry two branches of statistics commonly referred to as Multivariate Statistics and Survival Analysis.  The approach involves modeling what we have referred to before as the environment of a problem.  In our approach, we will use calculated data called canonical variates to read between the lines in order to find a better solution.

Canonical variates are calculated quantities from the original data that describe a relationship.

In order to achieve maximum profitability we must know which accounts are more valuable than another and for how long. Value on the surface is obvious.  We therefore concentrate our efforts on answering the question of how long an account will be more profitable than another.   Since we are modeling a relationship we want data that describes the relationship that we are modeling. Canonical variates follow this intuition. We calculate the variates that describe the relationship between types of loan accounts in the portfolio and then use the variates to predict when to buy and sell each type, in tandem.

The synergy of multivariate statistics and survival analysis allows us to look at a portfolio holistically and model the timing of events against each other.

Knowing what to keep and for how long

Loans are sold and purchased every day.  Loans are originated or acquired and then become an asset that may be sold at a later time.

We consider a loan portfolio to be a dynamic entity that evolves over time. The need for continuous monitoring exists to keep loan accounts that are important and divest transitional loan accounts appropriately. Predicting the time to a potential adverse event therefore allows more flexibility to effectively managing a diverse portfolio. In short, canonical variates allow us to look beyond the data and read between the lines.

We predict how long to hold on, and when to let go for maximizing profitability or satisfying strategic goals.

We provide the analytics that give you the power to know.

With the ability to mathematically prioritize which individual accounts should be sold at what time and what attributes of new accounts should be pursued; you maximize not only immediate profitability but also predict the duration of the difference in profitability that can be potentially achieved.  The process offers a quantifiable solution to loan origination or acquisition and the timing of sale.

In future articles we will be begin exploring various methods of determining attributes that are key drivers within your data and how these attributes ‘stack up’ into segments of customer behavior.

Stay tuned as we continue to explore the many facets of SMART BANKING.

Tags:

Leave a reply

Your email address will not be published. Required fields are marked *