jofra49 wrote:
Can linear regression be used to determine how rates change over time?
I am trying to do a project that determines how employment rates change each year for 5 years, for each industry, for certain geographical regions, and how to predict said rates for the next 5 years.
How can this be done with Python?
You could do it in MS Excel if you wanted to.
For most regression techniques they assume all variables have the same level of "integration."
If you violate this assumption you get bad predictions.
For example, lets say you ran a regression of unemployment rates on GDP. Well, unemployment rates are between 0 and 1 and GDP trends upwards over a long horizon. So lets say over a short time you found a correlation. Lets say it was negative, higher GDP -> lower unemployment rate. You would probably find this depending on where you took the data.
But if you started plugging in numbers for the future you'd get this result where as GDP goes higher, predicted unemployment rates go to zero - which doesn't make sense. In fact, with linear regression you'd get negative unemployment rate predictions.
Instead what you do is run a regression on another variable that is bounded or "stationary." So instead of GDP, you use GDP growth rates. While they can be negative infinity to positive infinity, for practical uses in the data set they tend to vary in a bounded range.
To make up for short comings of linear regression, you'd use logistical regression instead, which would bound the output between 0 and 1. Linear regression won't give you biased or incorrect estimates and in fact might be preferable to logistic regression for practical usage. A problem with logistic regression is that as estimates get close to 0 and 1, they tend to get unstable. By that I mean, a small change in the inputs can lead to a large swing in outputs.
05 Apr 2023, 7:01 am
