June 10, 2014

How to Think About Forecasting

When I graduated college in 1995, I got a job working in Data Warehousing -- setting up reporting systems using a variety of relational and multi-dimensional databases.  It required a balance of financial reporting (i.e., what to report) and data design skills (i.e., how to report it).

Yet, if you asked my grandmother, you'd hear a slightly different story.  As far as she was concerned, "Brett works in computers."   Now, people often said such things in the 1980's, and it's probably acceptable septuagenarian-speak today.    But for most of us, it no longer cuts it.

Working "in Computers"
Why is that?  How is it possible that a phrase that was totally legit in the 80's can now sound so ridiculous? 

The answer:  As a society, we have learned things, and as we learned, language got more precise.   The phrase "in computers" was fine, back when few people did.   But today, it's just a catch-all for a broad array of specialties -- networking, security, graphics, databases, software development, etc. etc. 

Here's the kicker:  As our language evolves, even some of these terms will become too generic.  I already know term that's bound for the trash heap:  Forecasting.

Forecasting isn't really a thing.
These days, I tell people that I do forecasting, and they nod approvingly.  Quite often, they exclaim, "You should talk with So-and-So."  Sometimes I do talk with So-and-So, and find that we have very little, if anything, in common.

Why this?   The answer is simple:  Forecasting isn't one thing.  People picture it as a specialty, like something specific, like database administration or UI development.   People think that it surely has some common base of skills that all forecasters share, the equivalent of a DBA's schema or accountant's general ledger -- but it doesn't.

If you choose any two forecasts, they're likely to be wildly, crazily different.  They can use radically different techniques, different math (or as my Oxford-trained friend puts it: different "maths"),  different error tolerances,  different outcome distributions, etc. etc.

How to Think About Forecasting
I've dreamed up many analogies and descriptions for forecasting, to try to convey to people just how broad forecasting really is (and consequently how devoid of meaning the term is).

Here's the best one I've come up with:  Forecasting is just guessing what's going to happen next.   That's it.

If you think of it in those terms, a lot of what I've written becomes obvious.  You might presume that forecasting is a formal skill -- but what about "guessing what's going to happen next"?  Is that a formal skill?  Not quite so much.

Also, if someone at a party says that they do "Forecasting", you might not your head.  But what if they said, "I guess what's going to happen next."  ...There, you'd probably give them a blank-eyed look, because it's such a meaningless thing to say.

But that's all forecasting really is.

No comments:

Post a Comment