What if there were no new construction starts beyond today?
What if the last new construction starts recorded for May (released by Dodge June 21) were the last to be posted and once those projects reached completion there would be no more work?
Of course this is a totally unlikely scenario, but deleting all future predicted starts allows to perform an important test. All the construction starts recorded as of today make up the backlog, and eventually that backlog will run out. So, if the new starts spigot was turned off today, how much spending would remain for 2019, 2020 and beyond? (For use later, new construction starts recorded through May generally equal an average of 40% of all starts expected each year).
The questions then are: How dependent is the spending forecast on construction backlog? How dependent is the construction spending forecast on new construction starts? What magnitude of miscalculation in the new starts forecast would be imparted to the spending forecast?
Single-family residential projects can take as little as 6 to 9 months to reach completion, multi-family perhaps twice as long. For the average nonresidential building, completion would be reached in about 24 months, but some large industrial projects will take three years or more. For some of the airport, highway and rail expansion mega-projects, the cash flow schedule of spending will take four to eight years to reach completion.
An average of ten years of monthly cash flows produces an average spending schedule for the various construction market sectors. Recognize that starts are posted every month, so January starts have twelve months of spending in the 1st year while projects that start in December have only one month of spending in the 1st year.
Residential project starts net about 65% of money spent in the 1st year, the year started, 30% spent in the following year and 5% spent in the third year, or 65-30-5. Although each type of nonresidential work has a more specific cash flow schedule, the average for nonresidential buildings is 20% spent in the year started, 50% in the second year and 30% in the third year, or 20-50-30. Very long duration infrastructure projects have a spending distribution on average that looks like 15-30-30-15-10.
Residential projects have the shortest schedule to completion. Work flow needs continual replenishment from new starts to support spending. The amount of work in backlog today would support only two thirds of anticipated 2019 spending and less than 10% of 2020 spending.
All Nonresidential buildings type currently have enough work in backlog to support 90%-93% of the total forecast spending in 2019. Current backlog would support only 50% of the total spending forecast for 2020. There’s only enough to support 10%-20% in 2021.
Power and Highway backlog as of today would support 95% of the total forecast spending in 2019 and 70%+ in 2020. Because these are long duration projects, there is enough in backlog today to support 40% of spending in 2021.
That’s a lot of good facts, but how can we use that information to perform an important test?
Let’s use the average nonresidential building for an example. For this example, let’s try to determine the validity of our 2019 forecast based on what we have in backlog today. New starts through May is about 40% of total starts expected in the year. Backlog through May supports 92% of spending in the current year. Spending in any given month has cash flow from an average of the previous 24 months of project starts, so the average of large numbers reduces potential error from backlog. The validity of our annual spending forecast is dependent on whether or not we correctly predicted the remaining 60% of starts for the year, and those starts support 8% of the spending forecast.
Therefore if we incorrectly forecast the remaining 60% of starts by 25%, then we incorrectly forecast total annual spending by 25% x 8% = 2%.
For the 2020 forecast, the math gets just a little more complicated. Remember we stated earlier that the typical spending schedule for nonresidential buildings is 20-50-30. So 20% of 2020 spending comes from new starts in 2020. Only 80% of 2020 spending comes from work in backlog at the start of the year. Based on what we have in backlog today, new starts through May 2019 supports 50% of 2020 spending. We are dependent on the expected new starts in 2019 to get us up to 80% of the expected spending in 2020.
We are expecting 60% more in starts in 2019 and that will support the currently missing 30% of 2020 spending. If we incorrectly forecast the remaining 60% of starts by 25%, then we incorrectly forecast total annual spending for 2020 by 25% x 30% = 7.5%.
Also for 2020, since 20% of all spending within the year comes from new starts within the year, if we incorrectly forecast 2020 new starts by 25%, then we incorrectly forecast total annual spending for 2020 by 25% x 20% = 5%.
I’ve posed this scenario by asking what would happen if we incorrectly forecast the remaining starts by an error of 25%. That would be a huge error, not very likely to occur. I’ve been tracking Dodge Data & Analytics construction starts for more than 10 years and have seen enough data to expect that by mid-year the unanticipated error in forecast starts for the end of the year might be more on the order of 5% to 10%, not 25%. And in fact, historically, revisions to year end starts data is usually UP, not down.
So, by deleting all remaining forecast starts data, we see the spending forecast based on cash flow of new starts would require a very large error in the starts forecast to translate into a large error in the spending forecast. If we apply a more reasonable and yet still conservative error of 10% in all projections of future starts, the forecast for 2019 spending would be off by less than 1% and the forecast for 2020 off by a total of 5%.