So you think you are smarter than the average forecaster, and can identify a trend in time series data?
You now have a chance to put your trend detection skills (aka trendar) to the test, and help the cause of forecasting research in the process. Nikos Kourentzes, Associate Professor at Lancaster University, is conducting an online experiment on how humans perceive trends compared to statistical forecasting models.
It's fun, and takes just a few minutes. And you can see how your trendar rates against several statistical tests and over 1400 humans who have participated. (Take the experiment)
Four Legs Bad, Two Legs Only Slightly Better
Results so far (1460 participants as of this morning) show humans getting it right slightly over 50% of the time, with the best statistical methods correct on over 90% of the test series.
So our species is only slightly better than chance, or what could be achieved by a sheep. But don't let that get your goat. No four-legged ruminant will ever best us at detecting fashion trends like Moon Boots, Mullets, Fanny Packs, Shoulder Pads, and Puffy Shirts.
Leave Time Series Trend Detection to the Experts
When it comes to finding trends in your time series data, leave it to the experts. When you rely on SASĀ® Forecast Server, automatic time series diagnosis and modeling will give you more time to pick out that next pair of Crocs.
1 Comment
Thanks for publicizing the experiment! I will post some preliminary results at my blog, once I have adequate sample. I am sure you will find them interesting - so far they are somewhat surprising for me! Nikos