3 things we can’t predict – and 1 we can
January 7, 2015
It’s the New Year, the time for predictions about what the future will bring.
Instead of looking into a crystal ball, I’d like to look into predictive analytics tools. A while ago I wrote about what predictive analytics is and why it can be useful (click this link to read about that).
The last months of 2014 brought me several contacts and proposals related to new commercial predictive analytics tools. For instance, IBM’s Watson Analytics is just out of beta. Watson and other tools of this generation are very user friendly, intuitive and don’t require any statistics or data science background to use. They are likely to land on most business leaders’ desks soon, and therefore the time is perfect to investigate what the predictive tools can NOT do.
1. Predictions are based on data collected on previous events and circumstances. Even if the mathematical model that performs the calculation is highly sophisticated, the results are only as good as the data going into the model. Terrorist attacks in Helsinki, for instance, are nearly impossible to predict, as there have been no precedents within this context. If data quality is poor and events are missing or impossible to quantify, the impact is the same as if the data didn’t exist.
2. Predicting a system that we have no influence on, like the weather, is different from predicting the behavior of a human system that reacts to predictions, like the stock market. In a system that reacts to predictions, any new information enforces the momentum that leads to a much wider impact than the original events or prediction. The recent Russian ruble crisis is a good example of this. The Russian central bank actions to stabilize the ruble had a practically opposite effect, when it was seen as a sign of confirming the seriousness of the situation by the rest of the market. On a more normal market day, the stock prices already reflect the analyst’s estimates of their performance in future. In a similar way, a favorable report on the performance of a school or university helps to direct the best talent to that institution, enforcing a positive loop of better performance by more talented students. Therefore, the behavior of a system that responds to prediction is much harder to predict as the multiplier effect of information must be taken into account.
3. The events that have the most impact on us are random and unanticipated, in the “Black Swan” territory. These are singular events that are highly improbable based on previous instances, but have a massive impact. Some natural disasters fall into this category, as well as disease outbreaks like Ebola, and unanticipated human actions like 9/11 or the Crimean invasion by Russia. It can be argued that with the right data all of the above could have been predicted, and some well-informed experts possibly did. However, for most of us they we complete shocks and surprises. These events will continue to impact our world in more fundamental ways than anything that we can predict either by human or machine learning.
Therefore, predictive analytics will unfortunately not predict the future. It will make sophisticated statistical analysis more accessible and applicable to a wider context, and that is actually not too bad.
The thing we can predict with good confidence is that the coming year will be an exciting one for anyone involved in analytics and digitalization. There are fantastic opportunities for both business users to benefit from new tools and technologies, and for the businesses selling these products and services. Wishing you a successful New Year 2015!