Ungated Post | 01 Mar 2017
Lights, Camera, but No Action?
A report by Michael Thom in the American Review of Public Administration investigates the impact of various types of tax incentive programs on economic activity in the Motion Picture Industry (MPI) on employment, Gross State Product (GSP), earnings, and industry concentration. The analysis implies that the alternative tax measures have different impacts on MPI activity:
- Both sales and lodging tax waivers have no effect on the metrics studied.
- Transferable tax credits have a permanent positive impact on MPI employment but no effect on earnings, GSP, or industry concentration.
- Refundable tax credits have a temporary positive impact on earnings but no effect on employment, GSP, or industry concentration.
Our review has identified issues which cast doubt on the conclusions of the paper. These include:
- Lack of precision of dependent variables: Measures of MPI activity use data from industry code 512. This covers activity in a broader set of industries than those directly affected by tax incentives, including music production distribution and film exhibition. The series offers superior coverage with all data points published by an official source during the relevant period. However, the broader measure can only act as a good proxy of trends in film production if there is a strong co-movement between the two datasets; unfortunately, we find virtually zero correlation between annual growth of industry 512 and the narrower 51211.
-
- Misinterpretation of coefficients: ** The tax credit measures are represented by two variables: the dummy, valued at zero or one depending on whether the tax credit is operating, and the duration, measuring the years that the tax credit has been operating. The impact of the tax credit should be understood by the combination of both variables. However, the only results presented are for the individual significance of the coefficients; a joint test could have shown statistical significance, even though the variables were not individually significant.
- Modelling policy variables simultaneously: For the indicators of MPI activity, the results presented test for the impact of the various incentive programs simultaneously. However, given the likelihood that some of these will have co-existed in different states, it would have been preferable to test each in turn. This feature of the regressions is likely to have affected the estimated coefficients for the policy variables.
- Growth rate and adjustment for state industry size: New York and California lead the US film industry. Running the regressions in terms of growth rates has two potential limitations. First, growth rates are likely less volatile in larger film-producing states, which can lead to estimation bias similar to the well-documented growth convergence modelling. Second, changes in larger states will be under-represented in the analysis. Therefore, an alternative would have been to test a model scaled to industry size and/or factored in initial state film production/employment level. Without this, it is not surprising that the sensitivity test where the model was run without New York and California had no discernible impact on the results.
- Inclusion of a variable measuring tax credit generosity in level terms: One of the control variables used in the equations measures the annual change in state tax credit spending. The specification of the variable as absolute change, instead of percentage, seems problematic. For example, due to discrepancy in state industry size, changes in “generosity” in California and New York are likely to have dwarfed changes in other states. Conversely, the greater absolute size of activity in California and New York should mean that MPI activity growth (measured in percentage change) is less volatile. This likely dampened the size of the estimated coefficient on generosity.
- No adjustment for possible endogeneity bias: the regressions contain control variables possibly leading to endogeneity bias. It is recognisable that increased film production in a given state would yield increased tax credit spending and employment. This will introduce simultaneity bias into the model, causing the estimated coefficients to be inconsistent. The paper notes that first differencing corrects this problem, but that only works for endogeneity bias caused by omitted variables, not simultaneity.
- Issues with diagnostic testing: we have also noted some small issues in the diagnostic tests used to assess the validity of the model. However, in our view, these unlikely had a substantive effect on the results.
Overall, our review questions the methodological approach and the central conclusions of the study. The use of a fairly broad indicator of sectoral activity is particularly problematic, given the lack of correlation with sub-industry trends.
Click here to read the report.
Oxford Economics’ team is expert at applying advanced economic tools that provide valuable insights into today’s most pressing business, financial, and policy issues.
To find out more about our capabilities, contact:
Americas
Diantha Redd
+1 (646) 503 3052
Email
Asia Pacific
Peter Suomi
+65 6850 0110
Email
EMEA
Aoife Pearson
+44 (0)203 910 8054
Email
Related Services
Post
KPMG M&A Outlook 2026: Between Uncertainty, Resilience, and Seizing Opportunities
Discover how Germany’s M&A landscape is evolving – with a focus on growth, AI and post-merger value creation.
Find Out More
Post
Silver, the next generation metal
This report highlights the critical role silver plays in data centres and artificial intelligence (AI), automotive and electric vehicles (EVs), and solar energy photovoltaics (PVs). With these sectors expected to expand significantly over the coming years, we expect future silver demand to be strong.
Find Out More
Post
Powering the UK Data Boom: The Nuclear Solution to the UK’s Data Centre Energy Crunch
The UK’s data centre sector is expanding rapidly as digitalisation, cloud computing, and artificial intelligence (AI) drive surging demand for high-performance computing infrastructure.
Find Out More
