Can I create a time series to forecast sales based on weather?

[#1]

I have sales data from a seasonal company and want to forecast the sales based on temperature. How can I do this in JMP?

[#2]

I'm not 100% sure of what you are trying to do here...perhaps you can clarify? The thread question is around creating a time series to forecast sales. Is that all you want to do? In that case using the time series analysis and modeling platforms is your best approach. Or ultimately are you trying to build a predictive model of sales as a function of temperature? Or perhaps it's both?

For the predictive modeling piece here are my intial thoughts.

The simplest place to start is by plotting the temperature and sales columns in the Distribution platform and ask these questions:

1. Where's the middle?

2. What's the spread?

3. What's the shape?

4. Are there any observations that look suspicious, nonsense, missing, or outliers, that might influence subsequent analysis?

The goal here is to validate that you have reasonable and appropriate values for any subsequent analysis.

Then I'd go to the Fit Y by X platform and just plot the data there to give you some idea of the relationship, if any, between the variables.

Now, since you want to build a predictive model rather than an explanatory model I'd think about a modeling validation strategy and tactics that you'd like to use. What you do next depends in part on if you are using JMP or JMP Pro. Too many different pathways here to suggest an appropriate approach...read up in the JMP documentation on validation to start.

Now, finally, it's time to actually create a model. Within the Fit Model platform (and others) you have a plethora of pathways. My recommendation is pick at least a few and see how well they perform with an eye towards answering the practical problem at hand.

Good luck. I'm sure others will offer additional thoughts.

[#3]

First, welcome to the community.

Adding to Pete's suggestions, you might look at a time series of the sales data (X, MR Charts).  As usual if the data isn't good, well you know garbage in-garbage out.

Here are a few of my random thoughts:

1. There may be a lag effect so if the temperature impacts sales in a future time period, you will have to account for this,

2. There may be lurking variables that are not included in your study (I doubt just temperature affects sales).  I suggest you develop a thorough list of hypotheses as to what may effect sales and think about how to get data to provide insight to your hypotheses.

3. Really I think the best you could do is to establish whether or not there is correlation between the Temp. and Sales, which does not imply causal relationships (which is what you want to establish for predictive modeling)

The platforms will include Fit Y by X (for one X), Fit Model for multiple variables, Multivariate Methods>Multivariate for correlation

[#4]

Here's a link to a (potentially?) related post. I think @peng_liu 's comments might be just what you're looking for.

Time Series Forecasting with Two Related Variables