Hello together,

I want to build a predictive model for sales data, and herefore I want to analyze sales data and its correlations with other time series using Symbolic Regression (Eureqa: nutonian.com).

In traditional multivariate time series analysis (e.g. VAR models) there is the risk of spurious regression. In order to prevent spurious regression, stationarity (and also cointegration) of the various time series are tested before setting up the model.

Within the user guide of Eureqa nothing concerning stationarity or cointegration is mentioned (http://formulize.nutonian.com/docume...g-time-series/). Also in the description by Koza (Genetic programming; 1992) nothing regarding stationarity & cointegration was mentioned when applying Symbolic Regression to time series data.

I am still very skeptical whether it is possible to just throw any time series into the tool and receive valid results – however I do not have a statistics degree. Does anyone know more about this topic?

Secondly, I have stumbled upon the paper “Statistical Modeling: The Two Cultures” by Leo Breimann (2001) in my research, where he differentiates between data models and algorithmic models. This confused me even more – does this mean that when I use algorithmic modelling (e.g. Symbolic Regression), I do not have to check for spurious regression etc.?

Thanks for your help and kind regards,
Michael