Regression Prediction with Time Series Data

[Master1022]

TL;DR

I am conducting a linear regression using climate time series data (temperature, precipitation, etc.) in order to predict how a certain device will perform. Should I remove seasonality from the data even if we expect the device to perform differently at different times of the year?

Hi,

I am not sure what the correct forum is for this question.

Question: When do we need to remove seasonality from time series data to do a regression analysis?

Context:
I am planning to conduct a prediction analysis where I want to find out how a device performs. I hope to estimate a function that takes the form:

\begin{equation}<br /> \text{device output} = f(\text{temperature}, \text{precipitation}, \text{etc.}) <br /> \end{equation} <br />

and given the nature of the device, I expect that it will have less output in the summer and more in the winter. I am working with one year's worth of data to train the model and have some predicted data from the future. Therefore, I think the different temperature levels are very important factors in the regression. However, I have been told that I ought to 'remove seasonality from the data' (no further clarification was given upon asking again).

I have done some reading on the internet, but have been unable to find any information. Any help would be greatly appreciated.

 

[mfb]

With a single year of data you can't remove seasonality anyway. There would be nothing left!

If you are sure you understand every factor that influences the device output there is no reason to adjust anything. There is a good chance you are not sure, in that case it might be interesting removing average seasonal effects. To do that you need something to average over, however (and then look at deviations from the average).
If you produce a function f(temperature) but the actual device output depends on sunlight, for example, your model will still produce the right seasonal cycle but it misses the actual reason for it, so predictions within a given time of the year won't be good.