Today it's no secret that A/L managers, auditors, and examiners are enamored by forecasting models. They use them to stress-test interest rate risk, liquidity risk, and capital. More and more modeling results are driving management decisions - and increasingly regulatory approaches.
Here's a paper published by IBM a few years ago entitled, "Seven Symptoms of Forecasting Illness". I think it's a good read (along with a similar article, "Management's Great Addition") given our increased reliance on forecasting models. The paper really hit home for me when they began talking about "False remedies to avoid":
Forecasting illness is hardly new, and several cures have tried and failed to cure this malady. Don’t waste time or money with the following “false remedies.”
Too often, forecasting illness is diagnosed as a statistical defect, the cure for which is a better statistical method. The notion is that managers can derive better information if they look at the same historical performance through a different lens. There are three problems with this solution. First, even the most advanced statistical algorithms do not deliver much value. The consensus among academics who study business forecasting is that relatively basic extrapolation techniques, such as moving averages, generally perform as well as, if not better than, more complex techniques. Second, individual companies simply cannot glean future trends from historical performance in the same way that someone forecasting macroeconomic trends can. Businesses change too often and too drastically. Additionally, the reliability of business data, which is prone to manipulation (for example, nudging sales forward or shoving costs backward), is sometimes questionable. Finally, most, if not all, businesses are subject to events and conditions that have never occurred before; working these unknowns into forecasts ranges from excruciatingly difficult to impossible.
Seven Symptoms of Forecasting Illness, p12 - IBM Business Analytics 2010
The first two sentences - "Too often, forecasting illness is diagnosed as a statistical defect, the cure for which is a better statistical method. The notion is that managers can derive better information if they look at the same historical performance through a different lens." - remind me an awful lot of what's going on with core deposit analysis these days.