Predictive analytics is defined in terms of extracting information about future conditional distributions of data, and the importance of a leverageable information advantage is discussed including the measurement of value with respect to benchmarks and the importance of properly modelling causality.
Differences between financial market contexts and retail contexts in terms of the nature of the analysis and the nature of the data is examined. It includes non-normality, heteroskedasticity, the large degree of correlation found within the data and the existence of power-laws and genuine outliers.
Graham ends by reviewing the actual differences in what we can and should do with data and which are the consequences of commodity storage and commodity computation in the big data era.
Presentation held at the RavenPack 4th Annual Research Symposium, New York, June 16th 2016.Request Event Materials