Laws of Forecasting *

I. Forecasts are always wrong.

The only question is "how wrong is it?" The goal is to get it accurate enough that it doesn't hurt the business too much when it is within its normal variation. That can usually be accomplished either through better forecasting methods or changes to the business to make it less sensitive to a bad forecast. more ...

II. Correct forecasts are not proof that the forecast method is correct.

It could have been luck. Don't just look at the results, look at the methodology. more ...

III. All trends eventually end.

No matter how accurately the trend is forecasted, at some point in the future it will be wrong. Consider what might cause a trend to change (seasonality, new competition, saturated market, etc.) when evaluating a forecasted trend.

IV. Complicated forecast methodologies can be dangerous.

Simple forecast methods are easy to explain, understand, analyze and debug. Complicated methods tend to obscure key assumptions built into the forecast, which can lead to unexpected failures.

V. The underlying data in the forecast are nearly always wrong to some degree.

It is just a question of how far off it is. Therefore, the more data that is in the forecast, the more likely some important error will be missed.

VI. Data that has not been regularly used is almost useless for forecasting

Data quality is usually directly proportional to the amount it has been used. Without regular usage, errors remain undetected and inconsistencies develop. It's better to use solid data in a forecast even if additional assumptions have to be made in order to use it.

VII. Most forecasts are biased in some way -- usually accidentally.

It is very difficult to eliminate all bias in a forecast, since the forecaster always has to make certain assumptions about which factors to include, how strongly to weight them, and which to ignore. And sometimes the bias is intentional. more ...

VIII. Technology will not make up for a bad forecasting strategy.

Create an appropriate strategy first, then use the technology to make it better.

IX. Adding sophisticated technology to a bad model makes it worse.

If the model is bad, anything you add to it -- statistical methods, time-series methods, neural networks, etc. -- will make it worse. And it will be more difficult to figure out what is going wrong.

X. Large numbers are easier to forecast than small ones.

Everything gets easier as the numbers get bigger. A forecast of unit sales where there is an average of 1,000 units sold per month is a lot easier to get right than one where average sales are 2 per month. It is frequently better to forecast the bigger number and back into the component parts than to forecast the component parts and then sum them up to forecast the bigger number.

(Click on links above for more details about each law)

* OK, they're not scientifically tested laws like Newton's law of gravity or the second law of thermodynamics, but they do seem to be true most of the time in our experience.