Learn about fixed effects regressions.
- [Instructor] Regression analysis is a great tool for making forecasts and predictions. But, it's not perfect. In particular, there's a number of problems that often come up with regression analysis. And one of the most pernicious problems is what we call omitted variables, or omitted variables bias. In particular, what sometimes happens is that we lack data on one factor that might impact the variable we're trying to predict. Recall our friend Ed, who is looking at predicting the value of a commercial property building.
One thing that might impact that commercial property building is a zoning change. If the building Ed is looking at has historically been zoned as a commercial property, but the town's zoning committee has recently re-zoned the property for a landfill, that's probably going to have an impact on valuation, right? Just makes sense. We lack data on zoning changes. Now, in some cases, we can try and deal with the omitted variables problems by looking at gathering additional data.
For example, data on zoning changes and how that might impact predicted values historically. But, in other cases, we simply can't realistically find the right kind of data to deal with this. One opportunity that we have then to deal with omitted variables is a powerful technique that we call fixed effects. Fixed effects, or a fixed effects regression, lets us deal with omitted variables using a special modeling technique.
What is a fixed effects regression? Well, fixed effects is a statistical technique that essentially creates a placeholder variable for a unit of interest and lets us avoid problems with omitted variables. Here's an example. Let's pretend we're trying to predict a particular salesperson's sales next month. Well, the problem is, that no matter how much data we capture, we're still probably going to be missing that human element. Maybe this person feels really motivated next month, and so they're able to sell better.
Well, we're going to be missing that particular data point. We have an omitted variable there. One way to deal with this is to create a fixed effects variable. In essence, to create a single dummy variable or binary variable that will capture each individual person. So if we have 10 different salespeople, we will have a variable for salesperson one, for salesperson two, for salesperson three, et cetera. And it'll capture the intangibles related to that particular salesperson.
Intangibles such as likability. In essence then, fixed effects lets us capture individual elements that are unique and specific to a particular individual, be it a person or a company, for example. Fixed effects is a very powerful tool that we can use in our predictions in regressions going forward.
Professor Michael McDonald demonstrates how to harness the wealth of information available on the Internet to forecast statistics such as industry growth, GDP, and unemployment rates, as well as factors that directly affect your business, like property prices and future interest rate hikes. All you need is Microsoft Excel. Michael uses the built-in formulas, functions, and calculations to perform regression analysis, calculate confidence intervals, and stress test your results. He also covers time series exponential smoothing, fixed effects regression, and difference estimators. You'll walk away from the course able to immediately begin creating forecasts for your own business needs.
- Understanding big data and economic forecasting
- Predicting values with regressions
- Analyzing economic trends and economic cycles
- Using fixed-effects regressions and binary regressions for forecasting
- Assessing the accuracy of an economic forecast
- Using scenario analysis