Trendy forecasting
- Maurizio Cortesi, Ph.D. - Zegtraining
- Oct 9, 2015
- 2 min read
Data, big data, bigger data. Data mining, computer analysis, models, predictions. Decision making, humans vs computers: is there a winner, and who shall decide that? Big data is a big topic of our time, and businesses are orienting themselves towards a deeper integration of computers in the decision processes, automating a task where we presumably thought to be essential.
Still, can we have realible forecasts about highly volatile outcomes? What happens when historical data are less informational than the contextual ones? With these questions in mind a team of researchers at IE Business School asked experts in the music field to make predictions on where a new single would enter the Top 100 Chart.
Among the variables given to allow for predictions, some can be considered historical, some contextual. For instance, previous research has shown that having a big-name (an actor with a big history of success in previous movies) is not enough to be a sure hit in cinemas; advertising, marketing, and other more contextual aspects are seemingly more important. The research findings confirmed this same idea, where better predictions were made by those who weighted the contextual variables the most, leaving only little or no space to the historical data.
When your industry, or problem, or context, is highly uncertain and volatile, perhaps data is not enough. You can't read the future only in the past, because the present often makes a big difference. Forecasting is trendy, but a hit in the past doesn't always assure you a future hit. Your work is now, renewed every day, and here is your context, not the same as yesterday.
And then again, on many things you can make some good and reliable predictions: the time you'll take to reach the workplace, the problems you'll have to face during a typical day, how much you'll spend for your lunch. Less volatile, less uncertain, here yesterday is a good proxy for today, and even of tomorrow. Still, can you historically predict when this approach won't work?
Comments