Revisit: Simple Forecasting
Introduction
Scott Armstrong seeks evidence on simple vs complex forecasting methods
[This went out last week with a bad link. I apologize for that – ch]
Big data and sophisticated analytics are used to discover knowledge and to make forecasts. Do we have any evidence that this will be useful? Or is it better to rely on simple forecasting methods?
Our paper, "Simple Forecasting: Avoid Tears Before Bedtime" proposes that simplicity in forecasting requires (1) method, (2) representation of cumulative knowledge, (3) relationships in models, and (4) relationships among models, forecasts, and decisions are all sufficiently uncomplicated as to be easily understood by decision makers. We provide an operational definition in the paper.
Our review of studies comparing simple and complex methods found 93 comparisons in 28 papers. Complexity beyond the sophisticatedly simple did not improve accuracy in any of the studies. Rather, it increased forecast error by an average of 32 percent in the 21 studies with quantitative comparisons.
The effects are so consistent and substantial that we are concerned that we have might have overlooked disconfirming evidence. Please look at the references in the paper to see if we have overlooked any key studies, and send your suggestions to Scott Armstrong. Are there any conditions under which complex methods (as defined in our paper) are more accurate? Here is the .
Scott
–J. Scott Armstrong
Wharton School, JMHH 747
U. of Pennsylvania, Phila., PA 19104
Home Phone 610-622-6480
homepage: