Discuss (0)

Print

RSS

Home » Salescast » Here

Strengths and weaknesses of statistical forecasting

Statistical forecasting isn't magic. The fundamental assumption behind this methodology is that the most tangible source of information in order to predict the short term evolution of your business is nothing but your historical data of your business itself.

Statistical forecasting is one of many sources of information to build a reliable forecast. There are many sources of information available that could contribute to building a reliable forecast: surveys, newspapers, analyst reports, expert opinions, etc. All those sources are good, but what about automation and productivity?

We don't claim that a statistical forecast will do better than expert opinions when producing a very limited amount of forecasts refreshed on a monthly basis (think of forecasting GDP - Gross domestic product) - but what if you need hundreds, thousands or even more, forecasts to be produced on a regular basis? Very few companies can afford armies of experts to make forecasts.

More accurate than manual methods and expert opinion that does not involve a significant amount of time and skill. “30% increase in accuracy over the current status quo” is often what we hear from new clients. This is not because our clients are not capable – it is simply because our clients do not have the time to develop the information sources, and skills, and then apply these to hundreds and thousands of individual products. To be clear: The planning manager's insight or even gut feel might well be more accurate than any statistical forecast for a specific product – but the statistical forecast will be better over the whole product portfolio.

The tradeoff between accuracy and productivity is statistical forecasting's strength. It delivers the many forecasts your business needs while keeping costs under control and, in Lokad's case, reduces the required attention to absolute minimum.

It is no substitute for common sense and is best combined with it. This also means that statistical forecasts are dumb in the sense that no statistical method will ever be able to anticipate something that does not already exist in your data. If all the inventory of your closest competitor just turned to ashes due a major fire outbreak yesterday, then your sales will probably go up as clients turn to alternative suppliers - at least until your competitor recovers. No statistical approach - Lokad being no exception - will be able to forecast such a once-in-a-lifetime pattern. In practice it means that if important information for your business is not reflected in historical data, then there is no alternative to manual correction of statistical forecasts.

Short-term is where statistical forecasting shines the most. There are mathematical insights to this, but we will adopt here a non-technical angle. Let's divide the business changes into two broad categories: quantitative and qualitative. Quantitative changes represent typical day-to-day fluctuations in your business (ex: 5 units of product X are sold, then the next day, 7 units are sold). Qualitative changes reflect the ongoing transformation of the nature of your business and its market (ex: digital photography has gradually replaced film photography over the last decade). Looking at short periods of times, say a few months, then quantitative changes are typically vastly dominating qualitative changes. On the other hand, looking at long periods of time, say one decade, qualitative changes become much more important than quantitative evolution. Statistical forecasting is somewhat blind when it comes to qualitative changes. Statistics only applies to numbers, and, consequently, only to quantitative changes. Statistics takes its strength from the fact that quantitative changes are driving the business in short-term. Yet, the longer the horizon, the more important qualitative evolution is, gradually increasing the statistical inaccuracy that does not capture this type of changes.

Why statistical forecasts work despite economic downturns, business growth and external events. It's clear that global macroeconomics (say GDP) impacts your business. Yet, in order to refine you business forecasts, the current macroeconomic index is useless. It's the future macroeconomic index that is needed, and this one typically proves to be way harder to forecast than the sales of a typical business. In practice, we observe there is very little to be gained, as far short term sales forecasts are concerned, from country-wide indicators, because those are so hard to accurately forecast in the first place.

Good for inventory optimization, not strategy. In practice, it means that statistical forecasts are good for tactics, such as optimizing inventory replenishments, but not so good for strategy, such as reaching for a new market positioning.