The 2008-09 financial crisis, often referred to as “The Great Recession” or “The Great Depression” is almost impossible to forget. It took its toll on individuals and institutions around the globe. It was a result of many factors present leading to the crisis.
However, did you ever wonder if the financial firms, banks had predictive models? Yes, they did, but the predictions failed.
Did you know why? The assumptions made while making those predictive models were invalid.
The models didn’t consider the possibility that housing prices might stop rising and that they even might fall. And, when they started falling it turned out that the models were poor predictors of mortgage repayment. The belief that housing prices would always rise was a hidden assumption in the models.
The assumptions in predictive analytics are so powerful that having invalid or obsolete assumptions can bring down even whole economics.
The quality of assumptions made while building predictive models need to be clearly examined. Business managers should always ask what the key assumptions are and when do we call them invalid. It’s vital that the business and the analysts continually monitor if the key factors involved in assumptions have changed over time.
“The HBR guide to data analytics basics for managers” suggests a few questions that businesses can ask their analysts:
- Can you tell me something about the source of the data you used in your analysis?
- Are you sure the sample data is representative of the population?
- Are there any outliers in your data distribution? How did they affect the results?
- What assumptions are behind your analysis? and,
- Are there any conditions that would make your assumptions invalid?
The big assumption in predictive analytics is that the future will continue to be like the past. But over a period they may become invalid.
The most common reason for making the assumptions invalid is TIME. If the model is created in the past, years ago, it may not predict correctly.
Another reason the assumption may no longer be valid is if a key variable is not included in the model and that variable has considerably changed over time.
So, the conclusion is the quality of assumptions made is a highly critical part of a predictive model along with the right data and the right statistical model, and both the business and analysts cannot afford to not monitor them.