March 9, 2014

Ultimate Forecasting Disclaimer

Hello!  If you're reading this, that means you're interested in picking up some tricks about forecasting.  I'm going to write down all of the ones that I know, and maybe they'll be helpful.   

I promise to do the best that I can.   But, there's a big but...

Brett's Ultimate Disclaimer These are just tricks and techniques that worked for me.  I can't say for certain that they will apply to your situation(s).  And I don't think I'll ever be able to say more than that.  Indeed, when it comes to forecasting, anybody who says more than that is selling something.

Here's why:

1. Forecasting is intensely case-specific.  This is true in many ways -- here are just a few:
  • What, exactly, are you trying to predict?  (In database parlance, what is your prediction granularity?)
  • What levels of accuracy are acceptable?
  • What information is at your disposal to help with your prediction?
  • How often do you need to update your predictions?
  • Do you need to explain how your model works to a broad audience, or can you run amok with obscure math?
  • What tools do you have at your disposal?  Excel?  Relational databases?  Map reduce software?

There are profound ramifications to changing any one of these dynamics. Consider a  master chef, who is asked to make one truly excellent meal, under certain conditions.
  • Does it matter if he's working with fish or fowl?  Hell yes.  
  • Does it matter whether he has fresh or canned tomatoes?  Of course it does.
  • Does it matter whether he's preparing a meal for teenagers or foodies?  Naturally.
  • Does it matter how long he has to prepare the dish?  Yep.
Even this analogy under-sells just how different forecasts are, from one circumstance to another.

2. There are many, many paths to the top of the mountain.
Conversely, don't assume that there is a single best way to make your forecast. Some people insist upon MatLab or Map Reduce tools.  Others work in R, or Excel, or Python, or Oracle.  

Naturally, everybody would love to believe that theirs are the best tools to solve the problem.  But the proof is in the pudding -- if it works, the tools were suitable.  It's as simple as that.

Warning: Salespeople Abound
Why am I telling you this up front?  Because if there's one piece of advice, it's that you need to stay on guard from easy answers.

Consider your situation for a moment:
  • You probably want to forecast something in a business capacity.  That means that maybe you or your company has a budget for tools and resources.  
  • Meanwhile, forecasting is confusing as hell.  It's a heady mixture of math, algorithmic strategies, sophisticated tools, and more math.   It sure would be nice if there were an easy way to do it.
Where-ever there is money and a desire for a quick fix, there will be people peddling answers.  So it should come as little surprise that forecasting is CHOCK FULL of people who will INSIST that:
  1. They've solved problems exactly like yours before.
  2. They know the best (indeed, the only) way to do it right.
These people are either (a) selling you something, (b) blowhard assholes, or (c) both.   So -- be on guard!  This is a complicated topic, and while it's tempting to just believe somebody who tells you that he or she has the answers, you must stay on guard.  Even from me.

No comments:

Post a Comment