Data Insight


 

What does the Future Hold for Net Absorption in Houston’s Industrial Market?  Consequences of Reduced Job Growth Arising from the Oil Downturn

 

July 2016 I Vol. 2, ISSUE 7 I Download PDF

 

Executive Summary

We are now seeing the sustained pullback in the oil industry trickle throughout Houston’s economy, including substantial declines in Houston’s base employment. According to Dr. Robert Gilmer of the Institute of Regional Forecasting of the University of Houston, the depth of the oil downturn will likely lead to net job losses in Houston of -10,800 in 2016 and -30,000 jobs in 2017, then rebounding to 62,900 jobs in 2018 and 97,100 jobs in 2019, with the assumption that the U.S. economy remains strong. Demand for industrial space in Houston will likely continue to push forward in some sectors (e.g., warehouse, distribution), though remain weaker in others (e.g., manufacturing) with the ongoing fallout of oil downturn and declining job growth in Houston. Here, we assess how Houston’s industrial market will  likely perform in coming years using our statistical forecasting model for the influences of Houston’s economy (job growth) on demand (net absorption) for industrial space. 

Based on various scenarios for job growth in coming years, we forecast decreasing net absorption in 2016 and 2017 with reduced job growth, but absorption then rebounds in 2018 and 2019 with increasing job growth (Figure 1). Specifically, in 2016 and 2017, net absorption is forecasted to dip to 5.583 and 5.025 million sq. ft., respectively. Net absorption will likely rebound in 2018 and 2019 upwards of 7.728 and 8.723 million sq. ft., respectively, as strong job growth returns to Houston. However, annual job growth (Q4/Q4) only explains 27% of variation in annual net absorption of Houston’s industrial space, likely because demand for various industrial products (e.g., warehouse/distribution, manufacturing, flex) differ greatly in their response to changes in employment. Moreover, our forecasts of net absorption are only as good as the job projections on which they are based. To this end, the job forecasts hinge on continued strength in the national economy and also the level at which active rig counts return following their bottom.

Figure 1: Office Net Absorption Fluctuates with Job Growth

Motivation

The oil pullback has been longer and deeper than initially anticipated when it began in 2014. The United States has shown to be a much slower swing producer compared to OPEC.  As a result, the oil downturn has lasted longer than even the Saudis expected. This is not a repeat of the 1980s, in which Houston was simultaneously experiencing a banking crises, an overbuilt commercial real estate industry, and an overall slowdown in the national economy. Nevertheless, the current pullback in the oil industry is as deep as the 1980s and the 2008-2009 Great Recession, as measured by drops in WTI prices and the 75% decline in active rig counts. The effects of the oil downturn are spreading throughout Houston’s overall economy, including job growth. 

At NAI Partners, we have developed a statistical model for forecasting how annual net absorption in Houston’s industrial market varies with changes in job growth. In 2015 we forecasted annual net absorption of industrial space to be about 6,000,000 sq. ft., with an 80% prediction interval of 394,000 - 101,479,000 sq. ft. of net absorption, based on projections for job growth of about 14,500 in Houston. Houston ended up adding about 15,800 jobs and its net absorption of industrial space was about 10,627,000 sq. ft.  Our forecast was within the 80% prediction interval—a solid outcome considering that annual net absorption has ranged from 2 - 14 million sq. ft. over the past 16 years. 

Here, we use the most recent employment forecast for Houston from the Institute for Regional Forecasting at the University of Houston to make quantitative predictions for how demand for industrial space will change in coming years. Demand is measured by net absorption—the change in occupied space in units of square feet of rentable building area from one time period to another.  Positive net absorption occurs when there is an increase in occupied space, while negative net absorption occurs when there is a decrease in occupied space.   

Forecasts of Job Growth Based on Recovery of Rig Counts

One of the most prominent economists that forecasts Houston’s job growth is the aforementioned Dr. Gilmer, formerly of the Dallas Federal Reserve Bank and current Director of the Institute of Regional Forecasting at the University of Houston. Figure 2 shows Dr. Gilmer’s forecasts for job growth under three scenarios of recovery from the oil downturn, each of which is based on a different level of active U.S. rig counts following its likely bottom of Q1 2016. Dr. Gilmer’s forecasts are based on three different scenarios for the return of rig counts, while assuming a strong, stable U.S. economy.  

    Total U.S. rig counts have dropped from 1,930 in August 2014 to 404 in May 2016, a 79% decrease. In recent weeks, rigs have slightly increased to 417.  The three scenarios for rig counts concern the level at which active rig counts return, including a high return of 1,650 active rigs, a medium level of 1,500 rigs, and a lower level of 1,300 active rigs. Notably, the number of active rigs is strongly tied to the performance of the oil industry. Dr. Gilmer places a 30%, 50%, and 20% chance on each of these three scenarios, respectively.  

    Figure 2 shows historic job growth through 2015, and job forecasts for each of these three scenarios in the level of return in active rig counts, along with a weighted average of the three scenarios of rig counts. In all three scenarios, jobs decline in 2016 between -7,400 and -13,500.  In 2017, positive job growth of 10,300 occurs under the scenario of high rig count return, but the medium and low rig count scenarios show job losses of -34,200 to -57,600.  In 2018, job growth returns, ranging from 20,100 to 97,200 new jobs and in 2019 job growth ranges from 69,400 to 107,300 new jobs. Recall, these scenarios are based on the assumption of a strong, stable U.S. economy, which itself may well falter in 2017, 2018, or 2019.  

Figure 2: 20116-2019 Job Growth Forecasts for Different Rig Count Scenarios

Job Growth Predicts Net Absorption in the Industrial Market

Job growth is a modest economic predictor of net absorption of all industrial real estate (Figure 3). Demand for industrial space as measured by net absorption does increase with job growth (Figure 3).  The explanatory variable of job growth (Q4/Q4, year over year change) on the x-axis is scaled in thousands of jobs per year. The response variable of total net absorption on the y-axis is scaled in millions of square feet of all industrial space combined. The solid red circles are the empirical data points for 1999 - 2015, for which the one extreme point of 2009 corresponds with the the Great Recession.  

The solid red line in Figure 3 is the linear regression model of the statistical relationship between job growth and net absorption, of the form y = mx + b. Specifically, y = 0.0291x + 5.898, where y is net absorption, x is job growth, m is the slope of the line, and b is the y-intercept. While we have plotted the relationships in terms of their original raw data, due to lack of normality net absorption was log transformed for the statistical test yielding a coefficient of determination (r2) of how well the data fit this log-linear statistical model of r2 = 0.27 (0.17 for untransformed analysis).  That is, 27% of variation in net absorption is explained by job growth. This is a modest percentage given the many factors simultaneously occurring in economics and industrial real estate which could obscure any such relationship. At the same time, this leaves 73% of variation in net absorption explained by other factors. In particular, this analysis was performed for all industrial real estate combined. When flex, manufacturing and warehouse/distribution space are analyzed separately, it is only net absorption of warehouse/distribution space that shows a significant relationship with job growth, indicating that demand for flex and manufacturing space are not very well predicted by job growth. 

The slope of the line, m = 0.0291, describes how y changes as x increases—that is, an increase by 1 unit of the x variable increases the y variable by how much. Accounting for the y-axis scaled in millions and the x-axis in thousands, the slope of 0.0291 means that on average 29.1 sq. ft. of net absorption occur for every one new job. The dashed blue lines are the 80% prediction intervals (upper and lower bounds) for net absorption. Put another way, there is an 80% probability that absorption will be in this range for a given level of job growth. The y-intercept, b = 5.898, describes how much absorption occurs when job growth is zero. Even with low to near zero job growth, Houston still tends to experience net absorption of about 5.898 million sq. ft. of industrial space. This aspect of net absorption becomes more important for estimates of job growth that approach or go below values of zero in recessionary periods. 

2016 Forecast for Job Growth and Net Absorption

    We can evaluate the statistical model of Figure 3 using 2016 numbers to date for job growth and net absorption of industrial space.  Note, the statistical model is only for data from 1999 - 2015. Houston’s job growth through May 2016 is a loss of about -11,500 jobs, consistent with the predicted loss of -7,400 to -13,500 jobs for 2016 as a whole.  With a weighted average forecast of -10,800 jobs lost, the prediction is for 5.583 million sq. ft. of positive net absorption in 2016, with an 80% prediction interval of 0.060 to 11.110 million sq. ft.  As of the end of Q2 2016, there has been about 1.5 million sq. ft. of net absorption, on track to meet the forecasted value.

Figure 3: Net Absorption Increases with Job Growth

Industrial Absorption for Different Scenarios of Recovery from the Oil Downturn 

Based on recently released job forecasts for Houston, we make quantitative predictions of how net absorption in the industrial market will change with job growth in Houston. We further forecast net absorption in 2016, 2017, 2018, and 2019 based on job growth under the three scenarios of return in active rig counts. Figure 4 shows historic net absorption of industrial space from 1999 - 2015 in solid black line with open circles, with an annual mean of 7.228 million sq. ft.  The forecasted values for net absorption from 2016 - 2019 are plotted in various colors with dashed lines. The different scenarios of oil recovery suggest similar levels of net absorption in 2016, ranging from 5.5 - 5.6 million sq. ft., but then diverging in 2017. 

Under a recovery with a return to a higher number of active rigs (red line, Figure 4), net absorption is predicted to be 5.683 million sq. ft. in 2016, followed by 6.197 million sq. ft. in 2017, 8.727 million sq. ft. in 2018, and 8.634 million sq. ft. in 2019.  This is the most optimistic scenario for demand for industrial space given Houston’s economy and the oil downturn. Under a medium rig count recovery (blue line, Figure 4), net absorption is predicted to be 5.575 million sq. ft. in 2016, followed by 4.902 million sq. ft. in 2017, 7.813 million sq. ft. in 2018, and 9.021 million sq. ft. in 2019.  This is the most optimistic scenario for demand for industrial space in coming years given the different likelihoods of oil recovery in Houston.  The third scenario is a low rig count recovery (green line, Figure 4),  which predicts net absorption to be 5.505 million sq. ft. in 2016, followed by 4.222 million sq. ft. in 2017, 6.483 million sq. ft. in 2018, and 7.918 million sq. ft. in 2019.

Figure 4: Historic (1999-2015) and Forecasted (2016-2019) Net Absorption

Caveats and Uncertainty in Absorption Forecasts

We have assumed 80% prediction intervals. This is a probability of 0.80, which means that, while we are 80% certain, 2 out of 10 cases may fall outside this prediction interval given the noise associated with the data. If we were evaluating NBA free throw percentage, we would like our chances with the shooter who carries an 80% success rate, but he’s still going to miss 20% of the time. In predictive analytics, it is important to note whether the new values of the predictor variable (job growth) is within the range of the original data on which the projections are based. Extrapolation far outside the original data range can lead to unreliable predictions. In our case, job growth of original data ranges from -110,000 to +115,000. Most forecasted job numbers are well within this data range, which increases the likelihood of a reliable prediction.

Methodology

Commercial real estate data on industrial space were obtained from CoStar in June 2016. Data for all industrial buildings were combined for industrial space, and then separated by flex, manufacturing, and warehouse/distribution. Job and employment data were obtained from the Federal Reserve Bank of Dallas, based on Q4/Q4 year over year changes in job growth.  The statistical analyses and data visualizations were performed using the R software and programming language: 

R Core Team (2014). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/.

We used linear regression to examine the predictive effects of annual changes in employment (Q4/Q4  year over year change) on annual total net absorption (direct plus sublease) from 1999 - 2015, along with log transformed absorption to improve normality. Assumptions of linear regression that could render a biased statistical model were tested. None of the assumptions were violated, including statistical outliers in absorption, overly influential points in job growth, statistical outliers in employment, unequal variance, heteroscadascity, and serially correlated residuals (nonwhite noise error).  

J. Nathaniel Holland, Ph.D., Chief Research and Data Scientist

Dr. J. Nathaniel Holland is a research scientist with 20 years of experience in using the scientific method to extract information from complex multi-dimensional data. He joined NAI Partners in 2014 as Chief Research and Data Scientist. At NAI Partners, Nat leverages his sharp intellectual curiosity with his skills in statistical modeling to guide data-driven business decisions in commercial real estate. Like many data scientists in the private sector, Nat joined NAI Partners following a career in academia. Prior to taking up data analytics at NAI Partners, he held professorial and research positions at Rice University, University of Houston, and the University of Arizona between the years of 2001 and 2014. Nat is the author of more than 50 scientific publications, and he has been an invited expert speaker for more than 60 presentations. Trained as a quantitative ecologist, he holds a Ph.D. from the University of Miami, a M.S. from the University of Georgia, and a B.S. from Ferrum College.