So since Jasper is a student, we took advantage of that and visited one of the statistics professor at UT Austin, Dr. James G. Scott. He generously offered to meet with us during his office hour. To give a brief intro of him, he is an assistant professor in Department of Information, Risk, and Operations Management of McCombs school of business whose research focuses on the core issues of model choice, multiple testing, variable selection, and latent-feature extraction. Click here to see more of his bio.
He suggested that we start with a simple linear function with dummy variables. The dummy variables would be to comprehend different signals in the data – such as day of the week, month of the year, weather, sports game, festivals, and other local events that my affect the revenue performance of restaurants. Once we would build a simple table full of coefficients for each dummy variables, the calculation would be very easy and quick, unlike Holt Winters that requires heavy computational cycles.