Last Summer, we updated several thousand contractor profiles with a history of building permits they received over the past ten years or so. Since then, we’ve expanded our system for analyzing permits and are at a point where we have started using them to make broader inferences about the remodeling & construction market.
To verify the legitimacy of our approach (explained in more detail below), we compared the relative value of our index over time versus the US Census C-30.
BuildZoom’s Residential Index versus the C-30 (total expenditures in residential construction)
There was a very strong correlation (R=.8) between the trend in our residential index versus the C-30 over time. Plotting the two series on a line graph suggested our residential index lead by about a month. Adjusting the time series accordingly resulted in an increased correlation (R=.82).
BuildZoom’s Commercial Index versus the C-30 (total expenditures in commercial construction)
Plotting our commercial index versus the C-30 suggested a 12-month lead in our index. Adjusting it by 12 months, we saw a medium correlation (R=.53). Interestingly, the correlation from July 2009 through July 2013 was significantly higher (R=.76) than over the original time series. We are in the process of investigating the lower correlation from January 2009 – July 2009.
What it all means
There is compelling evidence that our approach provides a lead indicator for activity in the remodeling/construction industry. Due to the depth of our data set and nature of our methodology, we will be able to answer critical questions moving forward, with a level of accuracy and velocity that is unprecedented.
Why would we want to do this?
To preface, I should note that BuildZoom is fundamentally an online marketplace. If we weren’t already gathering and analyzing this data to support our core product platform, we wouldn’t have invested the effort required to answer market questions to this level of accuracy.
That being said, we believe it is important to have a comprehensive understanding of the market in order to best serve our users and grow our business. Specifically, here are some of the things we want to know:
- How is demand evolving?
- Which sectors of the market are growing?
- Which regional markets have large supply/demand gaps?
- Where is the market headed?
Are we reinventing the wheel?
There are two main indices that have gained broad acceptance in the market:
- The US Census’ C-30 uses a blended methodology that relies heavily on survey data from multiple sources to provide monthly estimates of the total dollar value of construction work done in the US.
- The Harvard Joint Center for Housing Studies Leading Indicator of Remodeling Activity (LIRA) blends survey data from several sources (including the US Census) to estimate national expenditures on home improvements for the current quarter and subsequent three quarters.
Both are useful for understanding the macro trends in remodeling and/or construction (the LIRA focuses primarily on remodeling). Our perspective is that neither supports the level of analysis needed to answer specific questions and inform tactical decisions.
Our methodology
- Data collection – We gather data from a blended collection of sources. On the supply side, we gather licensure data from over 90 regulatory entities in addition to input from over 35,000 contractors registered on BuildZoom. On the demand side, we collect and analyze permit records from over 110 regulatory entities and study the activity of over 500k monthly site visitors.
- Normalization – Attributes of licenses and building permit vary from source-to-source so once the data is collected, significant work needed to be done to normalize it.
- Reconciliation – Building permits and licensing are regulated by different entities. Work needed to be done to develop a matching algorithm capable of systematically reconciling licenses and permits with distinct contracting businesses.
- Classification – To effectively analyze license and permit trends across different regions, a canonical classification system was needed. While classification is a common practice with licensing, many permit authorities simply provide an unstructured description of the work being performed (e.g. Remodel existing shower in Bath adjacent to Master Bedroom). To address this issue, we developed and applied a multinomial classification system (using naive Bayes) to automatically determine (a) whether a building permit is residential or commercial in nature; and (b) the type of work performed (e.g. bathroom remodel, new home construction, etc.).
Leave a Reply