Asymmetric risk is crucial to marketing. Option traders may bet on a crop price movement in a direction that they believe is unlikely to happen but in the knowledge that if the unlikely event should occur the price movement would be so extreme that the return would more than cover similar bets where the unlikely did not happen. A bit like applying insecticide to cereals to control BYDV.

Similarly, we analyse trials to show the likelihood of an event not occurring by chance and often plump for a threshold of say 95% confidence that the response was meaningful. However, in the real world it is the return on investment that counts and where the cost of the input is zero, and the distribution of the response is symmetrical, a much lower threshold would be appropriate - in fact well below 50% in this case.

Big data should allow the production of actual distribution curves to give insight into the response of an action. In my experience it is rare to find a distribution in trials or markets that is symmetrical but rare to find an analyst that does not assume that it is. As a result opportunities are lost.

Can we improve our data communication to take into account cost and risk?