Analysts, investors and news outlets are enamored with seasonally adjusted housing starts (US Census Seasonally Adjusted Annual Rate, SAAR). The number makes for a great sound bite each month as the US Census releases a new number by adjusting the monthly actuals by a seasonal factor to produce an annual figure.
The use of SAAR can mislead the reader. As an example, this headline makes it sound like 66,000 fewer homes were built in July than expected:US housing starts for July 1.534M vs 1.60M estimate (ForeXLive.com). This is just one of the pitfalls of the US Census SAAR. For operators in the industry, there is a bigger problem.
Using SAAR makes the annual housing cycle look like a smooth run.
But the seasonal change in demand is a critical factor for operators at lumber yards, mills, and on lumber trading floors to understand and anticipate price moves and supply constraints. The popular seasonally adjusted numbers intentionally ignore and even obfuscate the annual cliff in demand that occurs each year.
Lumber industry teams need to look at the actual - without adjustments - monthly housing starts data. The chart below uses the same underlying data from the same time period as the chart above, without seasonal adjustments.
It is very important that a lumber buyer anticipate the seasonal decline in demand to avoid overstocking inventory. Furthermore, depending on the production plans of mills and inventory levels downstream in the channel, this drop in demand can cause downward pressure on price.
About the Data
These two graphics are derived from the same US Census Housing Starts monthly data. The first chart, with the relatively flat line, shows monthly start values that have been adjusted by the US Census using seasonality factors to produce an annual number. The second chart uses the unadjusted monthly data. In order to express multiple years on the same axis, I have normalized each month to the peak month of the year. In other words, each monthly number is expressed as a percentage of the peak month within that year.
I only used the years between 1991-2002 and 2012-2020, with the basis of my selection being steadily rising markets, most similar to the current market, at normal levels of demand (1.0 to 1.7 million annual starts). Sure, I could have thrown out 2020 due to COVID, but 2020 fits these criteria.
My selection of these years explains why seasonally adjusted numbers in the first chart track higher throughout the year; these are rising markets. But if you want to use data to forecast housing starts, you have to be really careful about the data you pick. Including all years’ data picks up trends in anomalous markets (the big boom and big bust). Eliminating too much data leaves a small data set that makes extrapolation risky. That highlights the core problem of outsourcing your housing market insights to the headline grabbing seasonally adjusted numbers - the data may require a PhD in statistics to interpret accurately.
What are the implications for the Fall 2021 lumber market?
This year the lumber market experienced a historic high. As demand slackened, mill lead times shrank, buying slowed and prices dropped (read this blog article to understand why high prices fall so fast). Upcoming drops in prices were also visible in the data as described in this blog article on early data signaling an OSB price decline. The after effect was a wall of wood purchased on long lead times at high prices clogging lumberyard inventories. At today’s prices, inventory purchased before the price drops is overpriced and can lead to a margin drain on the business.
As buyers come back into the market, the key question for buyers is “how much inventory should I have headed into the fall season?” If you listen to headline seasonally adjusted housing starts you may misinterpret the reality of the fall market. Actual housing starts will tell the story.
Demand across the US typically drops dramatically after October. Sometimes in October. There are notable exceptions, like 2019, where housing production held high, draining channel inventories ahead of the COVID supply chain crunch. More reason to understand the real numbers, and not let seasonally adjusted numbers lead you off the cliff.
To illustrate the magnitude of the difference between the average monthly starts in winter relative to summer, take a look at the percentages on the graphic below:
In gently rising housing markets between 1.0 and 1.7 million annual starts (our market), December housing starts are about 25% lower than summer. That roughly translates into 25% less lumber sold across the US. When that 25% decrease in underlying demand is accompanied by high lumberyard inventory at high cost (today), then prices can be expected to remain low relative to the peak, unless the actual housing demand is anomalously high for the season.
While analysts, investors, and news outlets watch the US Census Seasonally Adjusted headline numbers in the coming months, I’ll be advising Yesler customers with insights into the actual underlying data to manage risk and steer clear of avoidable cliffs.
August 31st, 2021