It’s become a favorite holiday tradition around these parts: wait for the National Retail Federation annual holiday spending forecast each October, then write a few choice words about why their methodology is awful and their forecast track record is even worse.
I overlooked the Federation’s Oct. 3 forecast this year, as the 2016 presidential election obscured everything else. That was my mistake.
In case you missed it too, the Federation said it expects holiday salesin November and December (excluding autos, gas and restaurants) to “increase a solid 3.6 percent to $655.8 billion — significantly higher than the 10-year average of 2.5 percent and above the seven-year average of 3.4 percent since recovery began in 2009.” Online retail is forecast to increase between 7 percent and 10 percent.
Despite the usual over-optimism, this isn’t the most outrageous forecast we have seen from this group. The 2015 forecast was for a gain of a 3.7 percent (actual sales rose 3 percent), so it is relatively close to last year’s holiday shopping number. Regardless, the calendar being what it is, Black Friday is coming, and with it, all of the usual misinformation and statistical silliness. The news media and social networks play a role here, reporting and posting Federation promotional press releases as if they were actual “news.” Why these retail forecasts have amassed such a poor track record predicting holiday sales is the subject of today’s missive.
I will spare you our usual diatribe about why forecasting is so fraught with error, and why so many people are so bad at it. University of Pennsylvania professor Phillip Tetlock, author of “Superforecasting: The Art and Science of Prediction“(and a recent guest on Masters in Business), has spent decades analyzing decision-making. His explanation about why forecasts usually are so bad and what needs to done to make better predictions should be required reading on Wall Street. His basic outline for forecasting includes: work off of a baseline (i.e., how have initial public offerings or holidays sales done on average), recognize the role of chance, be brutally honest in admitting what you do and don’t know, remain humble, be granular, acknowledge your biases and overconfidence and most of all, describe possible outcomes in terms of probabilities.
Even though the track record of most forecasts has been pretty awful, Tetlock’s research shows it doesn’t have to be. Few seem to follow the professor’s edicts when making their regular predictions about things like retail sales or unemployment.
Yet there are signs that the media is wising up: the Washington Post, Wall Street Journal, The Atlantic, and Fivethirtyeight have all run columns on how and why the Black Friday forecast folly is so useless. Black Friday skepticism is going mainstream.
There is also some glimmer of hope that the Federation has stepped up its modeling game, and is using more real data to create its annual forecast. According to the group:
NRF’s holiday sales forecast is based on an economic model using several indicators including, consumer credit, disposable personal income and previous monthly retail sales releases.
If this turns out to be a robust data-driven model, it would be a worthwhile upgrade. The older model that the Federation used not only produced inaccurate forecasts, but also cranked out a few humdinger double-digit misses. I will revisit this topic in the coming weeks, and report back if the new version is any better than the old version.
Originally: In the Mood for Holiday Forecasting Follies