To best measure the economic impact of the Michigan Business Development Program we have constructed what is known as a “spatial panel vector auto regression,” or PVAR, model. A full description of the technical aspects of this model and related output can be found in “Appendix A: Modeling the MBDP’s Economic Impacts.”
The model is constructed using data from the U.S. Census Bureau’s Quarterly Workforce Indicators from 2012 through 2016 and from data found in reports published by the MSF and MEDC. The QFI data capture the start of the MBDP as marked by its first approved deals and include the latest available quarterly data from the Census Bureau at the time of this writing. We use data from MSF and MEDC reports that span the life of the program, from these agencies’ MBDP “projects lists” and each MSF/MEDC annual report to the Legislature through Sept. 29, 2016. These documents were occasionally supplemented by information provided by an MEDC spokesperson.
We are particularly interested in the county employment effects of the MBDP program since its potential to create new jobs is how the program is marketed and defended. Our examination of county-level employment is useful because most MBDP awards are specific to one county, where new investments and job creation allegedly occurs.[*] Indeed, MEDC press releases, related news coverage and official state reports all seem to emphasize this angle of the program’s economic development results.
Evaluating the effect of incentives is a challenge. For instance, the timing of state investments and related subsidy payments influence the process of measuring their impacts, and these payments are paid out on an individual basis to each firm receiving state support. The unique geographic factors experienced by each firm further complicates the procedure.
Another concern is the possibility that firms seeking these incentives are not entirely random — that there’s something unique about them as a group. Perhaps, for instance, firms selected for subsidies are more likely to choose certain places to locate or expand their business than would otherwise be the norm. If these choices deviate from what should be expected without the influence of the MBDP, this risks biasing the estimated impact. Additionally, there may be something intrinsic to particular counties that make them more or less likely to contain firms that receive incentives. This, broadly speaking, is known as endogeneity bias, and it is a challenge for assessing incentive programs of all kinds.
To address this issue, we use a very flexible econometric model that allows the impact to materialize at different speeds. We favor this model since it allows us to test endogeneity, the technical problem arising from reverse causation, which may occur when incentives are provided to firms that choose their location on a nonrandom basis. We include a “spatial,” or geographic, component for most estimates because labor markets cross county borders.
The advantage of our approach — as opposed to the program administrators’ REMI-based model — is that we employ historical information about what has occurred. The state’s REMI forecasts rely on assuming information about both the future performance of the company receiving subsidies and the underlying economic conditions impacting the firm. In other words, our model relies on data based on observed reality, whereas the state relies on data based ultimately on predictions about the future.
[*] It is possible that some firms receiving MBDP support are credited for creating jobs outside the county in which their MBDP project is located. Further, some firms (by our count there are only eight) get support for projects located in more than one county. Some companies were allowed to count jobs created “statewide.” The impact of the potential employment growth from these types of projects may not be captured in our model if the new jobs were created in counties that did not host any other MBDP projects. We do not expect that these situations would significantly impact our findings.