At Kwelia, we’re big fans of looking for answers in data. While we often explore rental market trends, we’ve also explored other topics, such as the correlation between gas prices and rental prices, and performed Twitter sentiment analysis for a VP debate. Pulling in outside data sources such as gas prices can help to highlight interesting trends while also revealing connections between data. With that in mind, if you’re looking to cast your data in a different context, which outside data should you investigate and where can you find it? While the answer will certainly vary depending on what you are interested in, when it comes to data, few, if any, companies have more of it than Google. In fact, the billions of searches per day that Google uses to tailor advertisements to users can also act in the aggregate as a powerful tool to observe trends.
Google search data has been used for a diverse set of purposes from influenza epidemiology and estimating retail sales to finding movements in unemployment, inflation, automobile demand, and vacation destinations. Others have also found a correlation between Google searches for home real estate agencies and the Case-Shiller housing price index (incidentally, Robert Shiller, one of the creators of the index, was recently a co-recipient of the 2013 Nobel Memorial Prize in Economics).
Does search interest translate into rental demand?
Given how well search volume is able to detect changes in a plethora of “real-world” data, how well does search volume track changes in rental prices? Put differently, do rent prices vary with more searches? With fewer?
Using Google Trends, we obtained time-series data on searches for apartments in the San Francisco Bay Area, New York City, Philadelphia, and Atlanta. The search data contains relative search volume rather than absolute search volume. That means, for example, that if we lookup the prevalence of the search term “smartphone” in San Francisco in September, rather than receiving results saying that “smartphone” was 5% of all SF searches in one week vs. 3% of SF searches in another week, the search interest is reported as numbers between 1 to 100. This is done by indexing the maximum frequency for the “smartphone” searches over the course of September at 100, and presenting all other data relative to that maximum. So, for this example, if the fractions of searches for the term “smartphone” out of all searches in SF over four weeks were 3%, 5%, 1%, and 4%, then their respective relative search indices would be 60, 100, 20, and 80 (since 3/5= 60% , 5/5= 100%, 1/5=20%, and so on).
Nevertheless, these relative measures are sufficient for comparing how shifts in search interest compare to rental price changes. Remember, our interest is in finding correlations, which do not have dimensions. Against the Google Trends data, we pulled our own data on median rental rates (in dollars per square foot) for the relevant metropolitan areas over the same time periods.
We graphed the data for the Bay Area first:
Let’s get a bit formal: a little statistics
There appears to be a fairly strong similarity in how the search interest and the median rental rates move. But are we imagining things or is there a significant relationship here? To answer that question, we got nerdy so please excuse us if your eyes start to bleed! We calculated the correlation coefficient for the two data sets and performed a t-test to measure the significance of the correlation. From the t-distribution we found a two-tailed P value of less than .0001, which by conventional standards is extremely statistically significant. After repeating this for the New York, Philadelphia, and Atlanta data, we found that median rental rates in the other three cities also had statistically significant correlations with search interest.
So what does this all mean? Well, the correlation between searches and rental prices implies that search interest acts as a proxy for rental demand. In other words, keeping all things equal, if search volume increases, then prices increase (and vice versa).
Furthermore, the graphs clearly indicate that there is a time-lag between shifts in search interest and changes in rental rates. Thinking about this practically, this makes perfect sense, as savvy prospective buyers typically research apartments before purchasing.
Seasonality is also interesting. We can see that apartment searching research ramps up in late Winter before rental season begins and continues into Autumn, with consistently relatively low search interest in December. This, too, is consistent with our data which generally shows higher prices following periods of increased search interest.
Finally, here are a couple more pretty charts just for good measure:
NYC: we see a slight divergence in the search interest and price around late October 2012 that is partially due to time lags, but may also be due to the influence of Hurricane Sandy.