Can the What If Tool be used for economic time-series data?

[Hal Segal]

Can the WIT be used like this?

-- We have a data set of monthly data of the Canadian economy - approximately 30 years of monthly data with about 30 variables. It contains things like gross national product, size of labor force, housing starts, etc. 

-- We want to use ML to determine the relationship between the variables and predict future values. (Any suggestions on which algorithms would work best?)

-- THE BIG QUESTION: will the WIT allow us to change certain independent variables in the data and then see the results on the dependent variables in the data? 

[jwe...@google.com]

WIT can be used once you've trained a model that tries to predict one dependent variable (either a classification or a regression model) from some a set of input variables. So, if you can turn your problem into something where you take some of those ~30 variables as input and try to predict one of the other variables, then you could train a model on that data and then use WIT to analyze that model, and adjust the input variables to see the change in output. All of our demos on the WIT website are problems of this form.

But, if the problem you are trying to solve is really a time-series prediction (how will variable Y change over time given these variables X over the last many months, for example) then WIT isn't a great fit, because WIT isn't built for analyzing time-series prediction models.

[Hal Segal]

Ah, what a shame. I was hoping it would work for time-series data. Are you aware of any other techniques for what-if analysis with time-series data?