2022 State of Streaming
With the credit squeeze, looming recession and high gas prices affecting U.S. consumer behavior, we are increasingly being greeted with ugly headlines foretelling the failures of several core American industries, with the U.S. auto industry at the top of this list of struggling sectors. Recent reports suggest that yearly sales declines have reached double digits1.
Even though the declining auto sales were widely expected by industry analysts, the speed at which the consumer demand collapsed still caught some -- including the big-3 auto makers -- by surprise. The short-term forecasts of auto sales tend to be based on the prior months’ sales trends and anticipated near-term state of economy. While this established method yields adequately accurate forecasts at normal times, it often misses the mark at times of drastic demand changes.
For mall-based retailers, foot traffic has often been a good indicator for sales. Similarly, for auto dealers, showroom traffic and incoming phone or email inquiries can be used as indicators for sales in coming months. However, these numbers are often difficult to compile and imprecise.
A more measurable and potentially more reliable leading indicator for vehicle sales may be the online visits to auto-related sites. According to a study by Capgemini published last month2, 88% of the car buyers used the Internet as a primary source of information during the purchasing process. J.D. Power reported that 75% of new vehicle buyers in 2008 are using the Internet during their shopping process, and the shoppers spent on average of 6.5 hours online researching automotive information3. A study by CNW Market Research found that the Internet information is most heavily relied upon by shoppers starting 4 months prior to a new vehicle acquisition4.
Based on these research results, I thought it would be interesting see if there is any correlation between online visits and the actual auto sales. The figure below shows the unique visitors to large auto-related sites. The visitors included are the ones that have made more than one visit to a site and viewed more than 10 pages on the site (as one would expect a serious car shopper to do).
The data shows a clear declining trend beginning with July 2007, which corresponds with the beginning of the downturn in actual auto sales. So, at least at the first glance, there seems to be a correlation between the two.
In order to ascertain if there is a real correlation between the two data series, I plotted the actual auto sales (in log) versus the 3-month trailing unique visitor totals (also in log) for the two months prior to sales. For example, May-July visitor data are plotted versus September auto sales. As it turns out, the correlation between the two data sets is very strong, with a correlation coefficient of 0.98 for the time period being analyzed.
The next step of the process was to determine whether or not these visitor data could actually provide an accurate forecast for U.S. auto sales. As shown in the figure, the fit of this simple model is quite good and it seems that a fairly accurate industry sales forecast can be obtained from the visitor data.
So it seems that the Internet traffic data may serve as a good leading indicator for actual auto sales, as long as the majority of the shoppers actually carry the purchase process to the end. One important caveat is that the correlation may fray in the event that the purchase process gets disrupted. A significant demand shock, such as consumers being spooked by the meltdown in the financial markets and holding off on major purchases or not being able to obtaining financing for a car loan, might cause this existing model to break down. That said, car manufacturers and others following the industry closely may want to take a look at what consumers Internet behavior indicates for auto sales forecasts.
1 Department Commerce Advance Monthly Sales for Retail and Food Services data.
2 Capgemini, Cars Online 08/09.
3 J.D.Power, 2008 New Autoshopper.com Study.4 CNW Marketing Research, Inc., Purchase process Wave IX.