This plot overlays two data sources, starts cash flow and spending. For each of the three major market sectors, Residential, Nonresidential Buildings and Non-building Infrastructure, a light colored line and a dark line are plotted. The light line is the expected direction spending will move estimated from the monthly cash flows from Dodge Data construction starts data. The dark line is the movement in actual spending. All the nonresidential buildings and infrastructure data is the sum of all the markets within the sectors. This plot is not generated for each market.
The actual spending should follow a pattern that is similar to the direction of movement predicted by the estimated spending. If the patterns are similar, it is an indication that the forecasting tools are generally accurate. The thing to watch for is that the direction of movement was predicted accurately. For instance, the Non-building Infrastructure lines show pretty well from Jan 2017 through June 2019 that the starts data estimated the direction spending would move. To a lesser degree, Nonresidential Buildings shows correlation. Residential spending long term agrees with the estimate from starts cash flows, but spending is much more smoothed and actual spending inflection points seem to lag the estimated.
This adjusted plot below shows Residential Estimated from Starts moved out 6 months vs actual spending. The correlation is much better. This may be an indication actual residential work has a longer duration than I use in my model to cash flow the starts data. A test for this adjustment could become clear over the next 6 months. Whereas the original plot predicted the residential bottom in spending for 2019 already occurred and the next 6 months would post an increase in spending, the adjusted cash flow that shows better correlation indicates the bottom is yet to occur before spending starts to increase later in the year.
Reliability of Predicted Construction Forecast Data is always foremost in the thoughts of an analyst. Cash flow models provide for approximately 75% to 80% actual jobs data in the predicted spending for the forecast 12 months out. For next month’s forecast we have 96%-97% actual jobs data in the forecast, only missing jobs that start within the next 30 days. Reliability trails off each month. So, this is useful as a way to check the forecasting model. Essentially, this provides a check on the method I use for forecasting spending. It lends credence to the validity of the future forecast.
The forecast monthly changes in cash flow generated by the starts data are used to predict future spending in all Construction Analytics spending forecasts.