Graphic 3 illustrates the cumulative impulse response function of total employment to disbursed MBDP funds as estimated in equation 1 above.

Graphic 3: IRF of Total Employment to One Standard Deviation Change in Disbursed MBDP Incentives

Graphic 3: IRF of Total Employment to One Standard Deviation Change in Disbursed MBDP Incentives - click to enlarge

 This IRF is calculated from the PVAR in equation one, and we report only the effect of a one standard deviation change in inflation-adjusted disbursed MBDP funds on total employment in each county. The confidence intervals were estimated using bootstrap sampling (100 iterations). As is apparent in this graphic, the confidence interval for the MBDP impact on jobs is both above and below zero, suggesting that the impact of the program is indistinguishable from zero.

We also considered an alternative specification using a log/log transformation prior to first differencing (Arnade and Gehlar, 2005; Malone and Lusk, 2016 ). This estimate provides a similar result, with responses not statistically or economically different from zero.

Graphic 4: IRF with Log/Log Specification

Graphic 4: IRF with Log/Log Specification - click to enlarge

The total employment effects represent the bulk of the policy analysis. However, most of the MBDP dollars are allocated to manufacturing firms. Thus, estimating the impact of these investments on manufacturing might identify better sectoral effect of these investments. In the following IRF, we illustrate both the quarterly and cumulative effects.

Graphic 5: Cumulative IRF of Manufacturing Employment to One Standard Deviation Change in Disbursed MBDP Incentives

Graphic 5: Cumulative IRF of Manufacturing Employment to One Standard Deviation Change in Disbursed MBDP Incentives - click to enlarge

We find no statistically meaningful effect of MBDP disbursement on manufacturing employment. However, we are interested in evaluating alternative specifications using a log/log transformation prior to first differencing (Arnade and Gehlar, 2005; Malone and Lusk, 2016). This estimate provides a similar result, with responses not statistically or economically different from zero. Note, that we do not include the confidence intervals here, since they are so large relative to the point estimates they obscure any variation.

Graphic 6: IRF in Log/Log Specification of Manufacturing Employment

Graphic 6: IRF in Log/Log Specification of Manufacturing Employment - click to enlarge

Together, these impulse response functions provide evidence of the effect of the MBDP loan and grant programs on job creation in Michigan counties.

The MBDP program reports through fiscal 2016 roughly $156.7 million was disbursed to create 17,913 jobs, or roughly $8,747 per job. Our estimates of the impact on manufacturing show no statistically significant impact. However, the effects on total employment are negative. This implies that the MBDP program is having no effect on total or manufacturing employment.

As a robustness test we relax the Bayesian approach to ordering impacts, allowing total employment changes to influence MBDP incentives. Roughly the same result ensued for total employment, with near zero point estimates on the effect of a one standard deviation change of total employment on the incentive allocation to a county. The same results held with manufacturing employment. Also, altering the lag length to four or eight quarters likewise had no meaningful effect on the point estimates or the confidence intervals.

Together these results suggest the MBDP program of grants and loans (disbursed values only) play no role in job creation in Michigan counties where they were deployed. The only robust caveat to this conclusion is that the materialization of the impacts is longer than the five-year period under observation. That however seems unlikely given the reporting results of MBDP.

One other estimation is necessitated by these observations and one of the reviewed papers above. Gabe and Kraybill (2002) report that business optimism played a part in incentive payments in Ohio during the early 1990s. For the MBDP we evaluate whether or not the expected job created reported by businesses plays a role in the decision to fund an individual business. We find that it does.

In a simple ordinary least square model, estimating the percent of overestimate of job creation from the initial application, on the logarithm of the approved dollars is very statistically and economically meaningful.[*] Each 10-percent increase in “overestimated jobs” results in a 7-percent increase in approved MBDP funds. This result mimics those of Gabe and Kraybill (2002) and suggest that businesses that provide more optimistic estimates of job creation enjoy a higher level of approved funding. See Graphic 7 below.

Graphic 7: Percent Job Overestimate and Approved MBDP Dollars

Graphic 7: Percent Job Overestimate and Approved MBDP Dollars - click to enlarge

[*] F = 12.44 (p=0.001), B = 0.07, with Huber White standard errors of 0.011 (p-value=0.001).