Articles and Updates
Comparison of UA, REMI, and STAMP Simulations of Tax/Spending Increases
March 30, 2010
By Alberta H. Charney, Ph.D.
UA Estimates of Tax/Expenditure Impacts Compared to those of the Goldwater Institute (prepared by Beacon Hill) and REMI
The Goldwater Institute (GI) issued a press release indicating that a $1 billion increase in government revenue from a sales tax increase in Arizona would cut 14,415 private jobs from the economy, add 5,260 to the government sector for a net negative job impact of 9,155. The present study estimated a net positive job impact of 13,128 (20,510 due to government spending less 7,383 due to the tax increase).
The results of the two studies are somewhat difficult to compare. The UA study computes the negative impacts of the $918 million sales tax increase (net of the portion paid by tourists), not the $1 billion in taxes (all paid by Arizona residents) in the Beacon Hill (BH) study.The UA study then separately computes the positive economic impacts associated with the increased government expenditures, while the BH study has built in a link between tax increases and government expenditures. The UA study results also include the economic impact associated with the federal matching money that will be lost to the state as a result of expenditure cuts. The BH study does not remove the portion of sales taxes paid by tourists and does not attempt to measure the impact of the federal dollars that will be lost to Arizona's economy.
The following table compares the economic impacts referred to by the Goldwater Institute and compares them to the UA study and REMI simulations done by ASU. In Table 1, the results of the three studies have been normalized to $1 billion in a tax increase and consequently a $1 billion increase in government expenditures.
UA, Beacon Hill, and REMI Estimated Economic Impacts of a Tax Increase,
Normalized to $1 Billion and Excluding Portion Paid by Out of State Visitors
|Estimated Net Job Impact of $1 B Sales Tax Increase and Corresponding Increase in Government Spending|
|Beacon Hill STAMP||
*Estimate computed by simulations done by ASU in February 2009.1
The computation of net job impacts is necessary for comparing model results because of differences in the concepts of presentation. In the BH presentation of results, the impacts of both the tax and government spending increases are divided into private and public sector jobs. The UA and REMI results estimates the total (public and private) negative impact for the tax increase and the total (public and private) impact of government expenditures. For comparison, the impact figures in Table 1 have been normalized to a $1 billion increase in taxes associated with a $1 billion increase in government expenditures. Therefore, none of the net impacts shown in Table 4 include the federal matching dollars associated with certain components of government spending.
From Table 1, the BH net job impact estimates are negative 9,155 jobs (representing a loss of 14,415 jobs in the private sector and a gain of 5,260 jobs in the public sector), while both the UA study and the REMI results show positive net job impacts of a tax/government spending impact of over 8,000. The remainder of this study identifies some of the reasons why the BH model results are so different than both the UA IMPLAN estimates and REMI.
It must be noted that there is no Beacon Hill report available for Arizona. Rather, the Goldwater Institute's press release2 links to an unexplained half-page table, labeled "A $1 billion increase in transaction privilege use and severance tax revenue (Scenario 1): A 13% increase in current transaction privilege use and severance tax revenue." The values in Table 4 come from that table. A thorough search of the Beacon Hill website does not reveal a report for Arizona or a discussion of the model used for Arizona. The only report available was a description of the STAMP (State Tax Analysis Modeling Program) built for Pennsylvania3. Most of the following comments are based on that 40 page report. It must be noted that the PA STAMP model report is difficult to follow because many of the variables and notation used in the report are not carefully defined.
Before discussing Table 1, it is important to point out that the Goldwater Institute inexplicably simulated a 13% increase in both the transaction privilege (sales) and the severance tax, but the proposed sales tax increase will not increase the severance tax.
Why the BH Government Spending Impacts are Low Compared to UA and REMI
The BH net job loss figure of -9,155 is the residual from a -14,415 private sector job loss and a positive 5,260 public sector job gain. The 5,260 employment figure for $1 billion of government expenditures is extraordinarily low, so that value will be discussed first. That's lower than the direct impact (excluding the multiplier effects) of $1 billion in expenditures on general merchandise, in which no goods sold are produced locally and only the retail margin is retained in the state. This is absurd. Governments produce or buy mostly services, which result in far more jobs per $1 billion than 5,260. No other studies produce government expenditure impacts that low ---- not REMI, not IMPLAN, or any other.
Careful reading of the PA STAMP report reveals why the positive impact of government spending is so low. In the STAMP model, government spending is supposed to be linked with revenues. The PA STAMP model description indicates that government revenues are summed and spent by governments and, in the case of transfer payments, by households. But there are several problems with their approach.
First, not all government revenue is necessarily spent in the STAMP model. Specifically, an equation for government savings is defined as the residual between government income and government spending. The half-page table that was linked to the Goldwater Institute's press release did not indicate the change in government savings as a result of the tax increase. It is very likely that government savings increases in the simulation so not all tax revenue are spent.
Second, the STAMP model explicitly prevents some government spending from responding to a change in tax revenues: "Some government spending is assumed to remain unchanged even if tax revenues vary; the rest of spending is endogenous, in that it responds to the availability of funds (p.25)." It is difficult to imagine which components of government spending remain unchanged when there are zero funds. Elsewhere in the PA STAMP report states "The purchases of goods and services by some government sectors are considered to be exogenous to [determined outside of] the model (p.31)." This whole concept is absurd and results in a) an increase in government savings and b) a very low job response to an increase in tax revenues because a portion of government spending continues on, no matter how revenues change. The whole point of the present study and this discussion is to compare economic impacts of raising taxes by $1 billion and increasing government spending by $1 billion. In the STAMP model, when taxes are increased, not all of the revenue increase is spent, and portions of government remain unchanged "even if the tax revenues vary." No wonder so few government jobs are affected by a tax increase of $1 billion in the STAMP model.
Third, the STAMP model specifies government spending in ways that will automatically result in low economic impacts of government spending. The major arguments of why government spending has larger economic impacts than household spending are because a) governments buy more goods and services locally (in-state) compared to households, b) governments spend mostly on services, and c) service sectors have high direct jobs/$million expenditures. In STAMP, the model structure prevents government spending to generate those comparatively larger impacts than household spending. For example, rather than healthcare expenditures in STAMP directly affecting health-related jobs (doctors, hospitals, nurses, long-term care facilities), this important government expenditure is treated as a transfer payment to low-income households (PA STAMP, p. 11). If portions of government spending are treated as household income instead of direct spending, then (by design) the job impact of government spending will be extremely low.
Finally, there is no explicit link in the STAMP model between state expenditures and the level of intergovernmental (federal matching) revenues. Thus the $442.5 million in federal matching funds associated with state government funding were not considered in the Goldwater Institute's press release. Not assessing the loss of federal matching funds is really not a failure of the STAMP model; rather it is the fault of the Goldwater Institute's use of the STAMP model when they neglected to incorporate those additional dollars.
In contrast, the UA study and the REMI simulations spend all the revenues from the tax increase. Further, the UA study carefully allocates all government spending listed in the FY2011 Conditional Enactments to appropriate NAICS and IMPLAN categories for impact analysis, based on detail data in the State of Arizona FY2010 Appropriations report.
Why BH Impacts of Tax Increases are Large Compared to UA Study
There are many similarities between the IMPLAN input-output model used by the UA and the STAMP model. Both are used for impact/policy analysis. The STAMP model is a Computable General Equilibrium (CGE) model that starts with the same type of data contained in an input-output model and many of the same underlying assumptions, then changes the nature of some of the relationships. The construction of the STAMP model (and a number of other CGE models) utilizes the IMPLAN database.
CGE models are more complex than I-O models, requiring additional data and additional assumptions. For a simulation of an increase in the transaction privilege (sales) tax, the assumptions regarding prices are critical. In an input-output model, there are no price effects, other than what the model user assumes. For the increase in the sales tax, the UA impact study assumed price elasticities of one and zero cross price elasticities, thereby reducing consumption of all taxable categories by the amount of the tax, but having no effect on the purchase of other items. These identical assumptions are incorporated into the CGE model via the consumption function outlined in the PA STAMP model (p. 15).
What is unclear, however, is how Arizona's transaction privilege tax was interfaced with the STAMP model. In particular, price appears in every relationship in the CGE model, including the demand for industrial inputs and consumer demand, by category. However, Arizona's transaction privilege tax does not apply to most industrial inputs. All inputs that are directly incorporated into the manufacture of a product are not taxable in Arizona. In addition, there are exemptions for equipment used in the production process. Further, most services (telephone and utilities are exceptions) are not taxed for either consumers or businesses.
Since the STAMP model structure is the same for all states and since there is no description of which prices the sales tax applies to in the PA STAMP model, the obvious question arises: When sales taxes are increased in the STAMP model, does it impact the purchase of all inputs and the purchase of all consumption categories? Just how state-specific is the stamp model structure for Arizona? There is absolutely no way of telling from the PA STAMP write-up. If the transaction privilege tax is specified to impact all industrial inputs and all consumption categories in STAMP, then the 13 percent increase simulation of the tax would result in an extremely overstated negative impact of the tax.
Whether or not the STAMP CGE model results in reasonable estimates also depends on the realism and reasonableness of the added layer of assumptions required for the CGE model. A major difference between IMPLAN and the BH model is how imports/exports to the state are specified. In IMPLAN, imports/exports are determined by regional purchase coefficients (RPCs) that represent the portion of local demand that is supplied locally for each sector. RPCs do not change with price in an input-output model and therefore do not change with an increase in the sales tax rate. In STAMP, the same RPCs are combined with STAMP's import "elasticities" to create domestic share elasticities for each industry. Import elasticities of demand measure how domestic prices affect industries'mix of purchases between in-state produced inputs and out-of-state produced products. When domestic input prices increase, the price elasticity of demand for imports determines how much businesses shift from buying in-state inputs to inputs purchased somewhere else in the world. Although there may be price effects on domestic vs. imported input purchases, the parameters needed to determine the size of those effects are simply unknown, so they are set arbitrarily and by assumption in the STAMP model.
The PA STAMP description said that the import elasticities were taken from literature and referenced an article and a book4. This author did not review the referenced book, however the import elasticities with respect to domestic prices estimated and presented in the article by Reinert, et. al. ranged from .02 to 3.49, with a mean of 0.91. One-third of the estimated import elasticities were not reported in the article because they were estimated to be negative, a finding which is inconsistent with economic theory. If the negative estimates are set to zero (the lowest values consistent with theory), the mean import elasticities would be approximately 0.61. The STAMP model assumes import price elasticities much larger than this -- 1.50 -- for producers'purchases of intermediates, indicating that producers have a very strong response to in-state price changes, reducing their purchases of in-state produced inputs and choosing instead to import inputs. Thus the builders of the STAMP model have chosen to incorporate into its structure very large import responses to changes in prices.
Further, and even more objectionable, the STAMP model applies those same 1.50 import elasticities of demand to most of the economic sectors in the model, despite the fact that the referenced elasticities were estimated for manufactured goods and mining commodities, i.e., items that are typically transported. But STAMP applies import elasticities to all sectors and sets most of them at the very high level of 1.50. Some of these make no sense at all. For example, they apply a 1.50 import elasticity of demand to construction, implying that if domestic construction costs increase, industries will "import" more construction, which is nonsensical. Industries will reduce the amount of construction they undertake because of the increase in price (both the UA study and STAMP apply a price elasticity of one). But STAMP assumes that not only will they purchase less construction overall, they will import more and buy less of it in-state. It just doesn't make any sense. According to STAMP, industries will change their in-state vs. import shares of utilities, banking services, real estate services, insurance services, and communications if domestic prices increase, all with the same 1.50 import elasticities of demand that were estimated for manufactured goods and mining commodities. These import elasticities are extremely unrealistic for most of these sectors. For example, if in-state utility prices increase, most businesses can't just choose to import more of their electricity and natural gas from out of state. They are stuck buying utilities from local utility companies. A handful of sectors -- wholesale, retailing, health, other services, and something labeled ENTRHO (variable undefined) -- are arbitrarily assigned lower import elasticities of demand of 0.50 in STAMP. Most should have been set close to zero.
There are numerous other elasticities in STAMP that are set by the model builders' "professional judgement." Few of them are documented. For STAMP to become an acceptable model alternative to REMI or IMPLAN, the model should be fully described, simulated for sensitivity of each of the assumed elasticities in the model, and be subjected to the referee process of publishing such a model.
Summary of Discussion of BH Model
The STAMP model structure allows for model builders to affect model outcomes depending on how taxes are specified , on how the parameters (elasticities) are set, and how the resulting revenues are spent..
The underlying model structure effectively minimizes the effects of government spending.
- Governments save a portion of revenue increases, thereby not allowing that revenue to have an impact.
- Part of the government is specified to be exogenous (determined outside the model), so that even if revenues fall to zero it remains unchanged.
- Important government spending categories are treated as transfer payments, thereby not creating direct (public or private) jobs and having the same (comparatively low) impact of household spending.
We don't know how Arizona's transaction privilege (sales) tax is incorporated into the STAMP model, i.e., we don't know which prices are affected when the sales tax rate changes. In Arizona, most inputs are not taxed (particularly for export industries) and most services are not taxed.
- If all industry input prices are increased as a result of the tax, the estimated impact will be too large.
- If all industry export prices are increased as a result of the tax, the estimated impact will be too large.
- If all consumer prices are increased as a result of the tax, the estimated impact will be too large.
The elasticities that are assumed in the STAMP model essentially pre-determine that tax increases will create large impacts.
- The import elasticities for producer manufactured and mined inputs are much larger than those estimated in Reinert, et.al.
- The import elasticities that are estimated for manufacturing and mining commodities are applied to most other economic sectors.
Both the model structure and the parameter assumptions suggest that the STAMP model was both designed and specified in a way that biases the results toward the findings of very low impacts of government expenditures and very high impacts of tax increases.
What Will It Cost If Arizona Voters Reject the One-Cent Sales Tax Hike On May 18, 2010? Alberta H. Charney, Ph.D. 3/30/2010
Sales Tax Expenditure vs. Expenditure cuts: an Economic Impact Study. Alberta H. Charney, Ph.D. 3/19/2010
Find more research and data on Arizona Fiscal Issues .
1. Computation based on findings of Hoffman and Rex, "The Economic Effects of Government Spending Reductions Relative to Other Options: A Report from the Office of the University Economist." Feb 2009, W.P. Carey School of Business, Arizona State University. Computation assumptions: a) REMI results are approximately linear, so they can be scaled and b) REMI results are approximately symmetrical so that the results of a tax cut and tax increase are equal but opposite.
3. PA STAMP (State Tax Analysis Modeling Program), A Complete Tax Model for Pennsylvania State, The Beacon Hill Institute at Suffolk University, 2009. http://www.beaconhill.org/BHIStudies/PA-STAMP2009/PA-STAMP2009.pdf
4. The article was K.A. Reinert and D.W. Roland-Holst. "Armington Elasticities for United States Manufacturing Sectors." Journal of Policy Modeling. 14, no. 5 (1992): 631-639. The book was: D.W. Roland-Holst K.A. Reinert, and C.R. Shiells. "A General Equilibrium Analysis of North American Economic Integration." Modeling Trade Policy: Applied General Equilibrium Assessments of North American Free Trade. Cambridge University Press (1994): 47-82.
For additional information, please contact us.