Variable Selection

Variables represent point-in-time data points that are used to train your model. 

Variable Selection is how you choose what data will be driving trading decisions, which makes it a required step in creating a model. The absolute minimum is 1 variable, but it is highly recommended to choose many more than that. We recommend 20 - 100 variables as a good range to use. Some machine learning algorithms are better at handling a larger number of variables.



Presets are a way to quickly populate your selection with a collection of variables that we have already thoroughly tested. We generally recommend using the Technicals preset for daily and weekly investment horizon models and the Boosted Ratios for Monthly or longer investment horizons. 


The variable browser allows you to pick individual variables from each of our provided data sets. They can be found by expanding the hierarchy trees in the left section of the screen or by using the search bar.


All variables require at least one transformation to be chosen before they will be added to your model. In some cases you may just want the "Actual" value of the point in time data, but in many cases you will care more about the "% change" of a value. When dealing with fundamental or estimates, a relative time transformation is required since the data represents a certain number of quarters previous or ahead respectively.


When creating a model, one of the important decisions you have to make is what variables to include.

All Variables screen feature dataset coverage data to facilitate this decision. Coverage showcases data fill rates at the beginning of the backtest and at the current date, enabling better understanding of the quality of the dataset for a given stock universe.