This is my sixth installment of my project looking at Google search interest for ten types of outdoor recreation in Michigan: atving, boating, camping, fishing, hiking, kayaking, rving, hunting, skiing, and snowmobiling.

Previous installments were exploratory analysis of the data. This is my first attempt/option for fitting models to and forecasting the data series. I plan to create one or two additional options.

This model employs some of the methods I learned in my micromasters courses for denoising data and fitting trend and seasonality. Subsequently, I also employ a Vector Autoregressive (VAR) model that also considers exogenous weather variables.

In my Northern Michigan search interest project, I encountered (and overcame) many issues. One goal of this project was to use daily data to overcome these issues in a more straight forward and methodologically sound manner.

Problems this option overcomes include measurement error in Google trends data and accounting for trend and seasonality. Continued limitations are that this model doesn’t account for how the impact of weather variables changes for different times of year (or weekday/holidays). This method does not account for the impact of holidays. Also, I wasn’t successful in fitting the camping series with this model.