Collections Capacity Optimization - Part 2

The following article is the second in a series from credit cycle consultants PIC Solutions that looks at optimizing the capacity of collection operations in a bank or financial institution environment. PIC has written many articles on collector effectiveness at the creditor level.

In this month’s article we highlight some of the forecasting technique issues that organizations need to consider.

Why forecasting?

Forecasting is fundamental to optimization as there are a number of variables, such as volume of accounts, which need to be predicted in order for workforce planning to occur.

Linear Forecasting

Linear forecasting, described by the equation y = mx + b (where m is the slope and b is the intercept), is both the easiest and most common form of forecasting and can be performed on most commercially available software applications, such as MS Excel.

However, linear forecasting only determines the value for long term forecasting. When time periods such as monthly, or even daily forecasting is being done, it finds itself wanting. Linear forecasting effectively smoothes over any variability associated with time, including variability that may be described by underlying patterns.

Example: From experience we know that collections activity shows distinct seasonal variation, peaking for vehicle and asset finance over December, January and February. Hinging on which months, or series of months, were used in planning capacity, this seasonal peak would either be under planned for resulting in an increase in the bad rate for that time period, or the peak would be properly planned for resulting in potential excess capacity for the remainder of the year.

Time Series Decomposition Forecasting

Because of the limitations with linear forecasting, we have found time series decomposition forecasting a viable alternative that is able to add much to our clients’ understanding of their collections businesses. Time series decomposition forecasting is described by the equation Value = (Mean) x (Trend) x (Seasonality) x (Cycle) x (Random). In broad terms, the calculation identifies an underlying trend within a data series after which underlying seasonality is identified as well. As is the case with collections activity where distinct, reoccurring seasonality is apparent, the overall r-squared (predictability, in other words) is able to be improved upon when compared with linear forecasting.

Despite being only accessible through more advanced analytical tools such as SAS and NCSS, from our experience, time series decomposition forecasting has proven to be most accessible and readily comprehensible.

Some Forecasting Challenges

From a technical perspective, collections forecasting holds some challenges. Unlike inbound call centers where the demand call patterns are largely set outside of the control of the business – but nonetheless still able to be forecasted through Erlang methods – the drivers of collections forecasting range from internal policies, to macro-economic variables, to collections strategies.

To treat forecasts uncritically without considering alternative modifiers can be, if not perilous, short-sighted. To illustrate by way of example, although a forecasting model might suggest future volumes within early stage collections to be x, because the modeler has the hindsight of knowing that borrower interest rates increased in the recent past, he or she may modify the forecast up.

PIC Solutions provides consumer credit solutions to a wide range of blue-chip organizations. We are experts in the fields of credit, risk and software and have an established track record of success powered by solutions.