Forecasting, in general, is based on extrapolation from the historical data. While it is certainly the best way to get things started, relying only on the historical data for future predictions spells a recipe for disaster!
I think the two most important things in making good forecasts are:
1-Considering a plethora of factors that could affect your forecast - consumer preferences, economy, technology, competition to name a few.
Nintendo recently updated its forecasts from a 55 billion yen profit to a 25 billion yen annual loss. This was mainly attributed to the decrease in demand of the Wii U.
Nintendo had initially predicted that it will sell about 9 million Wii U and about 38 million Wii U games. It then changed its forecast to 2.8 million Wii U’s and about 19 million Wii U games.
The problem here that Nintendo faced was that it did not consider or properly predict the impact of two main factors affecting the video game industry:
(i)-consumer preferences (ii)-technology
More and more casual gamers now prefer playing games on their mobile devices instead of having a separate device like Wii. Moreover, if a person can download most of the latest games for free from the Android app-store, why would he/she be willing to spend about $300 on hardware and spend another $15-$20 on buying games to go with that hardware.
In addition, the hardcore gamers prefer the more powerful Sony Playstation 4 or the Microsoft Xbox one. Because of this situation, Nintendo is kind of stuck in the middle mainly because it did not give the required weightage to the aforementioned factors. And now, here they are, reconsidering their entire business model after a 25 billion yen loss forecast.
2-Re-forecasting as soon as new data or information is made available.
General Motors has been doing great work in this front. They announced a third new business plan in a span of 4 months in the summer of 2009. As prediction of car market gets tougher and the target keeps on moving, it is imperative that they revise their forecasts and work on achieving a target better than the forecasted one.
As David Cole, the chairman of the Center for Automotive Research said “It is a lot better to exceed a forecast than not live up to it."
They had to take these steps of forecasting again because of the influence of the economic crisis and the ever-changing fuel prices
I remember vividly, the struggle of figuring out the business values and the factors that would affect the predictions of the probability that a customer would repeat purchase a product that he/she bought on offer previously. I developed about a 100 odd features from the gigabytes worth of original data having about 15 features. Understanding the economic and business factors like consumer choices, pricing, elasticity of demand over and above the historical basket-level data was the most important factor in coming up with these features and developing a model based on a subset of the most important features.
Hence, more than just the data, forecasting is about how you integrate the domain knowledge with the data, the speed and frequency of forecast updates and finding the useful data from a large pool which can help achieve the focused goal of your forecast.
Sources/References:
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.