When forecasting is powerless

In previous articles, I described proven statistical forecasting methods.

And now let's think. All we can do is predict using short time periods. After all, modern companies live very little: how many of you work where there is statistics for at least five years? Our goal is to build a reliable forecast for the year, for example. But modern business is grimacing and does not want to be predictable.


All scientific approaches, methods for analyzing time series (even the most complex ones), successfully fitting a complex process, turn out to be absolutely useless in the future. All because the director of logistics did not agree with the customs, and the ridiculous number of bicycles from China stood at the border while the season was over. I would like to look at the face of Finir, comparing the plan (built according to your super-method) with the fact :)
')
Does this mean that scientific forecasting methods are an absolute crap that you shouldn’t spend time on. By no means! Even despite the skepticism of the crowd that predicting, based on historical data, is the same as driving a car, looking only in the rear-view mirror. Of course, do not be so distorted. Methods and algorithms are evolving - and in many cases the very simple and simple methods that I have described are very effective.

But it is stupid, really stupid to take on tasks whose solution by standard means is a priori impossible.

Exchange rates, futures

Here is a chart of exchange rates. Probably many will be right in asserting that the trend will continue. But when it changes, even the super investment analysis professionals do not know.


And Murphy with technical analysis is not at all and. Trend changes are predicted by completely different entities - not according to such graphs. Just do not mess with currencies. Nostradamus among us is not enough.

Product Forecasting

A very common mistake. Well, yes, our company sells a thousand items, three hundred of them are very important to us. We make a forecast for them! This misconception and no one needed detail. Always try to aggregate - this will keep the prediction error to a minimum. And secondly, the groups always behave more steadily.

Take a look at this picture at the top of the post. I doubt very much that someone would correctly predict in a coherent way. But in the group they behave very clearly.
And then, who wants to make dozens of times more unnecessary predictions;)

Mad behavior

A business process does not have to be in a state of resuscitation; it’s just that there are always peaks and falls in life. But if they are chronic, ready to bet, you will not always be able to honestly look the boss in the eye.
For such cases there is always outlier cutting. Any professional statistical program in two accounts will identify rascals and replace with a moving average.


And you know, definitely, the power of the analyst is in the balance of the approach, and not in the knowledge of a couple of methods. Therefore, I described this kitchen from two opposite sides.

PS I put together all the articles on forecasting. In my opinion it was funny.
Methods of forecasting, and how it can be useful in a crazy business (as well as one tattoo on a bare shoulder;) Download in pdf format (~ 1.3Mb)