and his team listen to the community's feedback and offer, within a window of three years, two fantastic, insightful and impactful forecasting competitions, the M4 and M5. I am tired with all the "sceptics" that try to dismiss the competitions' results.
1/8

First, the ongoing discussion on error measures. It seems that everyone in the forecasting community has devised their own measures, and if someone is using another one then s/he is doomed. The real question is: does the chosen error measure fit the purpose of the exercise?
2/8
Second, the disbelief on the results especially from people that did not participate in the competitions. I have said that before, but skin in the game is important in forecasting (if you want to call your self a forecaster).
3/8
The fact that a method won a competition does not mean that it will win all future competitions. But, there are important lessons to be learnt. In the case of M3 and Theta: the value of information decomposition. In the case of M4 & M5: the value of cross-learning.
4/8
Such insights should be taken into account when designing future methods and algorithms. Since 2000, I have seen many new implementations of the Theta method that have improved on the original proposition. The same will surely be the case with the top methods of M4 & M5.
5/8
Third, the inability to understand the inherent limitations of designing a forecasting competition. No forecasting competition is going to be perfect. The same way that no RCT experiment will fully reflect reality. But, this does not mean that each competition does not add...
6/8
...to our understanding of forecasting. If the father of forecasting, @spyrosmakrid himself, has changed his view towards "complex" methods and saw the value of ML methods, this should be a paradigm for everyone else in the community.
7/8
I am not affiliated with the Makridakis Open Forecasting Centre, but its motto says it all: "Learn. Forecast. Compete. Disseminate. Excel." @spyrosmakrid
I invite others to do the same.
8/8
@threadreaderapp
#unroll

More from Society

This is a piece I've been thinking about for a long time. One of the most dominant policy ideas in Washington is that policy should, always and everywhere, move parents into paid labor. But what if that's wrong?

My reporting here convinced me that there's no large effect in either direction on labor force participation from child allowances. Canada has a bigger one than either Romney or Biden are considering, and more labor force participation among women.

But what if that wasn't true?

Forcing parents into low-wage, often exploitative, jobs by threatening them and their children with poverty may be counted as a success by some policymakers, but it’s a sign of a society that doesn’t value the most essential forms of labor.

The problem is in the very language we use. If I left my job as a New York Times columnist to care for my 2-year-old son, I’d be described as leaving the labor force. But as much as I adore him, there is no doubt I’d be working harder. I wouldn't have stopped working!

I tried to render conservative objections here fairly. I appreciate that @swinshi talked with me, and I'm sorry I couldn't include everything he said. I'll say I believe I used his strongest arguments, not more speculative ones, in the piece.

You May Also Like