In Northern Michigan, winter is coming to a close. The first snowmelt occurred last week for the lower and the next will start tomorrow. Northern Michigan received plenty of snow this year and last month was terrific for snowmobiling.

Last month, I adopted significant changes to my forecasting methodology which insofar appear to be extremely successful. During the first year of this project, I adopted a methodology that was very sensitive to recent search levels but was also provided very inconsistent forecasts from month to month. Last year I adopted a methodology which increased the consistency of the forecasts, but didn’t capture recent changes in the overall trend of search interest. This new method allows for both consistent forecasts and sensitivity to recent trends while still accounting for changes in weather patterns from year to year. This approach led to a very successful forecast of the February search interest levels. For the Upper Peninsula, at least, this wouldn’t have been possible without considering both current trends and weather.

This month, I have adopted changes to how I calculate the actual data series both for the aggregate peninsulas and individual places. Previously, I repulled this data in its entirety from month to month. This is practically simpler than just updating the data because Google trends internally normalized its data. However, the data provided for Google trends also changes from data pull to data pull. Since I now have 12 monthly data pulls for each of the data series (although not necessarily 12 values for the most recent values), I have averaged all of my available data pulls to achieve a more consistent series.

The main page for this project can be found here:
Northern Michigan Places Google Search Interest – dataandoutdoors