Case Study: Previse, the BMW M3, and the BMW X3

Posted on: June 23rd, 2022 | Posted by: Landan G

I was quite surprised to find out that BMW is one of the most popular and commonly searched for car brands on Google Search (link here). However, it does make sense for BMW to be so popular, they make beautiful high end cars while being high performance, luxurious, and exiting. While I was researching my next vehicle BMW did appear on my searches a fair amount of times, which made me wonder, how many people search for BMW vehicles throughout the year? Would more people search for a sedan/coupe like the 2 series during the summer months compared to a SUV like the X2? Or would SUVs be more popular in the winter when drivers typically desire a higher ride, AWD, and a larger vehicle.

With that being said, I wanted to look at the popularity of two BMW models throughout the year, namely the BMW M3 and the BMW X3, a 4-door sedan and a 4-door SUV. The data I'm analyzing is gathered from publicly available sources and is dated back a few years, so we can truly understand popularity throughout the year without being affected by individual outliers.

One thing I also wanted to include was the ability to forecast and predict future popularity for the BMW M3, so we could pinpoint a specific week or month a year or two in the future and understand how popular that model will be. So why do this extra step? What could we use this forecasting for? Such applications could include inventory forecasting, service scheduling, and inventory planning.

Without further delay, let's dive into the data and results.

BMW M3 and BMW X3 Google Search Interest

The data used for this case study is available publicly available and the notebook used to create these visualizations is available upon request, use the contact page on our website to get the notebook. The data contains search interest for the two BMW models from Calgary and Edmonton Alberta in the past 5 years. We're going to start off by plotting the data for the BMW M3 and BMW X3 on a single line plot to get an initial glimpse of our data, but as you'll see below we are unable to conclude and findings from the chart.

As you can see, the data isn't quite clear in the graph, so to help make things a bit more clear we can plot the two BMW models on separate graphs to try and get a better look.

While the dual plots helps a bit, one thing we can add to the plot is a moving average. In this case, I'll apply a 14 week moving average to smooth out the graph and try to eliminate some of the sudden swings.

So you may be looking at the plot and noticing some very obvious swings in the graph, and funny enough, it does look like the two graphs move together, meaning that there may be some positive correlation between search interest for the BMW M3 and the BMW X3 (I'll touch on this later). Ignoring the idea of the two makes moving together for now, I wanted to dive down and try to find out why the interest for these luxury vehicles had such drastic peaks and valleys for interest.
One theory I had was that the sudden swings in popularity for these models has something to do with when BMW announce, tease, or release the vehicles. I didn't want to waste too much time on this hypothesis I had so I gathered three press releases from BMW themselves and plotted those dates on the same chart as above. The results are below:

As you may be able to see, the interest for the models were already on the upswing when the announcements for the models were made. You can see a little upswing of interest when those announcements were made but overall nothing drastic. So, with my hypothesis proven false, I wanted to move on. The next step to take in looking at the interest of these models is to look at whether or not there is a positive correlation in search interest between the two models, a.k.a. if one model becomes popular then will the other model also become popular? To look at this I simply created a scatter plot and calculated the correlation value between the two models. You can view the results below:

Looking at the results, there is a correlation score of ~50.6 on a scale from -100 to +100. Given that score, we can reasonably say that there is a medium positive correlation of popularity between search interest of the two models.

Taking it one step further: temperature

Another idea I had was regarding the temperature throughout the year and how that affects search interest for these two models. In thoery, it would make sense for SUVs like the X3 to be more popular in the winter months than the M3, whereas the M3 should be more popular in the spring/summer months. I feel like I should take a moment to acknowledge that temperature != seasonality != other conditions like snow, rain, wind, or other factors that could affect the differences in search interest for the models. I chose temperature as an easily obtainable metric that wouldn't be hard to get publicly and put t use, however if this was a fully fledged commercial product then I would utilize more data than just temperature. Regardless, I downloaded publicly accessible temperature data for Calgary and Edmonton (the two cities that we got the BMW interest data from) and averaged the temperatures out. Again, this isn't a perfect strategy but for this case study I find it beyond sufficient. One last thing to point out quickly is that I had to remove some data from the vehicle interest to match it with the temperature data, however the results are still valid.

Let's use the same plot as above except now add the temperature as well, so we can get a rough idea of how the temperature moves with the search interest.

From first glimpse of the chart, you can probably see that the temperature of these two major Alberta cities does in fact move roughly alongside the interest of the BMW models. That being said, we can use the same correlation plot as we used above to look at exactly how temperature interacts with the search interest.

What was really surprising to me when I initially looked at the graph was the very high correlation between the M3 and average temperature compared to the X3. One could make a fairly accurate conclusion then that more people tend to search for coupe/sedan style vehicles when it is warmer out, especially in Alberta where we can get some pretty harsh winters. That isn't to say there's a negative correlation between the average temperature and the X3, but it was especially noticeable with the M3.

Like I mentioned above, this section wasn't perfect. We could factor in many other variables that could potentially impact the popularity of a specific vehicle, but this case study was designed to show what can be done, and not an exhaustive end-to-end demo.

Predicting the future of the BMW M3

Let's take this a step further shall we? Let's try to perform some basic predictions for the search interest of the BMW M3. Now, as we know from above factors such as the temperature can affect the values, but for the basic predictions we're solely gonna focus on the historical data. You can see that we were able to detect clear patterns in the data using advanced artificial intelligence showing that we're able to forecast the future based on historical data (while ignoring external factors) and easily utilize AI to power your business!

Looking into the future, we can predict that the interest for October 22nd, 2022 will be roughly 36.793. We'll have to wait a few months to truly know how accurate this model was!


Hopefully this case study was a small glimpse into what we can do with Bizatta Previse. Like I've mentioned above, there is a lot more we can do with our data to truly create the most powerful forecasting engine. If you need any forecasting or analytic services, feel free to reach out!