Thursday, November 10, 2011

Does Qualitative Analysis still matter in Forecasting?

In the field of predictive analytics, we learned that the inventory lifecycles of products can be driven further and further down by sophisticated modeling tools and methods such as ARIMA and Box-Jenkins. These tools exploit the data available to a firm (whether it be in the form of sales data or receipts from suppliers) in order to pinpoint the exact number of materials that would be needed to produce the optimal number of product.

In class and in the readings, human input was dismissed because of its inability to provide a sound foundation for the actual predictive process. The precision of our modeling techniques, it seemed, would be all we needed in order to make great forecasts.

However, in this brief article about predictive analytics, we can see that maybe letting computers run the show isn't such a good idea. In the writer's brief experiment with 100 random variables in his regression equation (presumptively part of an ARIMA exercise, if we were to use it for actual forecasting), he explains that because there are so many factors that may even be of fractional importance to the model, there may be inherent bias that a machine cannot detect.

For example, in the author's experiment, he talks about the correlation between a student's average SAT score and the amount of money that is spend per student in a state every year and its surprising trend. It appears as though spending more money on education DECREASES a student's SAT score, contrary to intuition. He goes on to further explain that there is bias inherent in the dataset that the model does not explain. The same problems exist in terms of predicting supply requirements: we need to know what the key factors that influence our expected supply level are, and we need to examine whether or not they have a true impact independent of other interactions from other variables. After all, you wouldn't say that shipping companies with red trucks ship more efficiently than those with blue trucks based on paint color alone would you?

We may be able to replace humans hard at work doing calculations by hand, but we may never be able to replace the inherent knowledge necessary to decide what constitutes a "good" forecasting model. What do you think the future is for qualitative analysis in forecasting? What do you think the challenges are with integrating qualitative and quantitative analysis?

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.