The ACA's Employer Penalty is Distorting Business

Taxes and regulations are known to affect the size distribution of businesses, due to the fact that smaller businesses are less subject to enforcement.  Large informal sectors are an obvious result in developing countries, but measurement challenges have hindered quantifying the size distortions’ impact on developed-country employment and productivity.  This paper uses new and unique data that is readily linked to a specific regulation: the 2010 Affordable Care Act’s (ACA) employer mandate.  The mandate’s size provision took effect in 2015 and is especially interesting, not only due to its notoriety, but because of its bright-line threshold and enforcement by monetary penalty.  This paper quantifies the size incentive of that penalty, develops a framework for combining evidence on size with evidence on voluntary compliance, and uses a new survey of businesses to quantify the number of businesses that changed from large to small as a consequence of the law.

The key size threshold in the ACA is 50 full-time equivalent employees (FTEs), which establishes the legal definition of a “large” business that is subject to the employer mandate.  Momentarily ignoring the distinction between FTEs and total employment, I display in Figure 1 a time series of the share of employment by small businesses, by a 50-total-employees criterion, among private businesses sized 25-99.  The data is sourced from the tables prepared by the Agency for Healthcare Research and Quality from the insurance/employer component of the Medical Expenditure Panel Survey. Both the 2015 and 2016 shares are well outside the range observed in the recent history 2008-14, and in the direction to be expected given that large employers were subject to a new regulation.

Garicano, Lelarge, and Van Reenen (2016) show how the distortionary effects of size-dependent regulations appear muted when the observer uses a different measure of size than regulators do.  This is the case in Figure 1, which looks at total employment as opposed to the full-time equivalents specified by the ACA and has total employment binned rather broadly (25-49 and 50-99).  Both Garicano, Lelarge, and Van Reenen (2016) and Gurio and Roys (2014) therefore obtain size measures that are especially close to regulator measures and find large size distortions in the French economy.  They do not link the distortions to specific regulations, but instead focus on France where there are many size-dependent regulations thought to be binding.  One of their estimation methods is to compare the actual firm size distribution to a Pareto distribution and measure the nonmonotonicity of the actual distribution in the neighborhood of the threshold.

The Mercatus-Mulligan data used in this paper has five measurement advantages.  First, it separately measures full- and part-time employment and therefore can produce good proxies for FTEs.  Second, the size distortion can be linked to a specific and relatively new regulation, which permits a before-after analysis as shown in Figure 1.  Third, voluntary compliance – that is, offering employer-sponsored health insurance (ESI) even when exempt from the mandate – can be measured.  This allows the measurement of size distortions to focus on businesses for which the employer mandate is binding.  Fourth, the survey was not conducted at the corporate level and therefore did not require any corporation’s approval to publish results.  Rather, individuals were confidentially surveyed, and these individuals happened to be managers at businesses.  If the sample aggregate happens to reveal politically-incorrect business practices, such a finding cannot impugn any particular business.  Fifth, the managers of the sample businesses were asked whether and how the law changed their hiring practices, with answers that can be compared to size and compliance.

Before-after comparisons between the Census Bureau business survey and the Mercatus-Mulligan survey show little change in the size distribution of businesses between 2012 and 2016, except among businesses in the total-employment range 40-74.  Among the latter businesses, the employment percentage of those with less than fifty employees has increased from 37 to 45, and this does not count the fact that a number of 49ers reduce employment below 50 full-time-equivalent employees (FTEs) without reducing their total employment below 50.  Annual time series from the MEPS-IC show an extraordinary jump in the employment percentage of those with less than fifty employees, beginning in 2015, which is the same year when the large-employer designation began its 50-FTE threshold.

            The size distortion is closely linked with whether a business offers employer-sponsored health insurance (ESI) to its employees.  Even by comparison with businesses employing fewer than 30 full-time workers, the propensity to offer ESI is low among employers with 30-49 full-time employees.  The size of this dip in the ESI propensity indicates the prevalence of 49er businesses: they do not offer ESI and thereby keep employment low enough to avoid the ACA’s large-employer designation.  The cross-section finding is my second and strongest piece of evidence that the ACA’s employer mandate is pushing a significant number of businesses below the 50-FTE threshold.

My point estimate is that the United States has 38,327 49er businesses that collectively employ 1.7 million people.  This translates to roughly 250,000 positions that are absent from 49er businesses because of the ACA, but the Mercatus-Mulligan sample by itself is not well suited for accurately assessing the average number of positions that the 38,327 49er businesses eliminated.  The sample also indicates that businesses continue to adjust their employment over time.  For example, many of them reported that, because of the ACA, they hire fewer workers or at least fewer full-time workers, but tried not to adjust the situations of their existing employees.  If the ACA and its employer mandate remains in place, perhaps the prevalence of 49er businesses will increase over time.

            By definition, the 49er businesses have less than 50 FTEs and do not offer ESI.  But it appears that a majority of them had been offering it in the prior year.  Employers with 30-49 FTEs are also disproportionately likely to report that they hire less or have shorter work schedules because of the ACA.  This is my third finding pointing toward an economically significant effect of the ACA on the size distribution of businesses.  To my knowledge, this is the first paper to find a business-size distortion that is readily visible in aggregate U.S. data.  It is also remarkable that the distortion can be linked to a specific regulation with a precisely known penalty for violations.

            Individual-based surveys of businesses are rarely used in economics, but that is bound to change as the survey industry is becoming more efficient (i.e., cheaper for the researcher).  It is worth noting the contrast between the Mercatus-Mulligan survey design and in-depth studies of a particular business (e.g., (Einav, Knoepfle, Levin, & Sundaresan, 2014; Handel & Kolstad, 2015)).  The former design has the advantage of representing a wide range of industries and geographic areas.  Moreover, this study is not sponsored by any business and therefore does not require a corporation’s approval for its release.  Corporate approval is a concern for studies of a particular business, especially when the topic involves public-relations-sensitive issues such as distorting business practices to lessen the cost of well-intended federal regulations.  Another dividend from using a professional survey research firm is that every respondent completed the survey.

            This paper does not put its estimates into an equilibrium framework. Future research needs to estimate the number of eliminated positions at 49er businesses that resulted in jobs created at businesses that compete with 49ers in product or labor markets.  To the extent that the employer mandate shifts employment from 49ers to other businesses, future research needs to assess the aggregate productivity loss from the shifts, recognizing that the ACA’s large-employer definition is just a vivid example of a more general pre-existing enforcement phenomenon.  Even without the ACA, businesses are taxed and regulated, and understand that adding to their payroll tends to increase the enforcement of those rules, albeit not discretely at 50 FTEs (Bigio & Zilberman, 2011; Bachas & Jensen, 2017).  One ingredient in such productivity calculations would be the number of positions shifted, which I found to be roughly 250,000.

From the equilibrium perspective, another interpretation of my cross-section finding – the nonmonotonic relationship between ESI and employer size around the threshold – is that businesses below the threshold did not adjust their size but merely dropped their coverage, in which case, I have mislabeled them as 49ers.  Indeed, I find that such businesses are disproportionately likely to have dropped their coverage in the past year.  However, this alternative explanation does not by itself explain why (i) so many businesses were added to the 25-49 (total employment) size category, (ii) so few were added 50-99, or (iii) coverage rates are not particularly low for businesses with less than 30 FTEs.

The implementation of the employer penalty in January 2015 coincides with a sudden slowdown in the post-recession recovery in aggregate work hours per capita, with 2016 national employment about 800,000 below the trend prior to the implementation of the employer penalty (Mulligan, 2016).  This paper’s estimates permit us to gauge the aggregate importance of the 49er phenomenon, not counting the marginal employment impact on non-ESI businesses that continue to employ 50 or more FTEs.  If 250,000 positions were the aggregate employment effect of 49ers (see the equilibrium caveat above), that would be about one third of the recovery slowdown.  Perhaps more important would be the social value of those positions, given that employment and income are substantially taxed by payroll, income, and sales taxes even without the ACA thereby creating a wedge between the positions’ social and private values.  If that wedge were $20,000 annually, that would be $5 billion of lost annual social value, plus the usual Harberger triangle, which is 38,327 businesses in the quantity dimension and up to $68,987 annually in the price dimension (about $1 billion annually).

Stanford University findings on the ACA and the labor market

Stanford Economics Professor Mark Duggan was quoted as concluding that

"While the Affordable Care Act had a significant effect on health insurance coverage, it did not have a substantial effect on the U.S. labor market as many had expected" and

"the Affordable Care Act has not had the negative effect on jobs the law’s critics claimed it would."

He was referring to a working paper distributed by the National Bureau Economic of Research, which states that

“labor market outcomes in the aggregate were not significantly affected.”

Theirs is a working paper and I'm sure that the authors are eager to add data and analysis, so I understand the above conclusions to have been modestly offered. With that said, it is worth recognizing that the above conclusions are not what the working paper shows.

1. Table 4 (the paper's first table on labor market outcomes) shows that the ACA reduced nationwide labor force participation by 349,190 in 2016, plus however much the ACA reduced labor force participation in a geographic area that was fully insured before the ACA, which I call the HFIA (Hypothetically Fully Insured Area).  This effect is economically significant and, when combined with items (2) and (3) below, is easily in line with "the negative effect on jobs that the law's critics claimed it would be."

   [Admittedly, the 349,190 is probably not statistically significant by the usual criteria, but the quotes above are not claiming that either side could be correct. Rather they claim to decisively reject "critics" who made claims right in line with the Duggan-Goda-Jackson point estimate. See below for the derivation of the 349,190]

2. The paper assumes, without much explanation, that the HFIA part of the ACA's impact is zero.  But other work has shown that near-elderly insured people were given a tremendous incentive to retire early.  In other words, basic economics tells us that the HFIA part is likely positive (i.e., in the same direction as the 349,190) and we should not assume it to be close to zero until we have further measurement.

3. The empirical methods in the paper, which emphasize differences among geographic areas such as Medicaid expansion states versus other states, are not designed to detect effects of the employer penalty.  The employer penalty is the same amount throughout the nation.  The penalty creates large labor-market distortions; those distortions that have been measured in other studies have proven to be similar across geographic areas.  Moreover, the employer penalty did not apply until the 2016 coverage year, whereas 8/9 of the working paper's data is before that date. This is an especially serious problem for the low-income population, where the employer penalty in effect has them working 50-60 days per year for the government, on top of the implicit and explicit employment/income taxes they would pay even without the ACA (this fact is nowhere mentioned in the paper). For this reason, the authors' claim than that "lower income individuals were actually incentivized to work more" is especially incredible.

To derive the 349,190, look at the first "Out of the labor force" column of that table.  The first row says that the M variable reduces out of the force by 0.0847 on average for each working-age person in the U.S.  The second row says that the E variable increases out of the labor force by 0.0962.  The mean of the M and E variables are, respectively, 0.073 and 0.086 (p. 11 of the paper). So, relative to the HFIA, their regression says that the U.S. has increased out of the labor force by 0.0021 per working-age person:

0.0021 = 0.073*(-0.0847) + 0.086*(.0962)

To get a number of people, multiply by the number of working age people (difference between these two), and you get 349,190.

Read here to feel the Joy of Voting

The economic analysis of voting primarily takes voting as instrumental: like a bank account, a vote is supposed to be nothing more than a means to an end. A few of us have argued against this: e.g., Geoffrey Brennan, Bryan Caplan, and recently Becker and Mulligan, but that is a small minority.

Another way to appreciate the same point is to see what actual Catalan voters had to say last weekend:

The polling station workers thought that if they had computers with older technology they may be able to connect to a wifi system [the Spanish government was blocking polling stations' internet access] ... we all started clapping – it had worked! They were connected. One man inside excitedly ran to inform the others... “I’m going to be the first to vote!” he yelled excitedly, to laughter. The two elderly women and a handful of others inside took up their ballot papers and voted.

Then the gates opened and the first round of people walked through. Everyone was cheering and applauding jubilantly ...the faces of those who came through were still calm and resolute but some became tearful after they voted. It was a really moving moment, and it’s hard to accurately put it in words. The best way I can describe it to say there was an overwhelming sense of dignity about both the moment and the people.

You can read the full account here.

That voting is to many people not merely a means to an end is better understood by Catalan separatists than the Spanish government.

The Upside-down Economics of Regulated and Otherwise Rigid Prices

Although not always highly visible outside of Communist countries, price regulations apply to a large fraction of economic transactions, even in the United States.  There are, of course, controls on apartment rents and taxi fares in major cities, and minimum wages for low-skill workers.  A number of states regulate interest rates on loans with usury laws and the federal government regulates interest and insurance rates with redlining prohibitions and antidiscrimination rules. Outside the state of Nevada, the price of sex is legislated to be zero.  Basic telephone and cable TV rates are regulated.  Price controls are the norm in the health sector, which by itself is already a sixth of the U.S. economy.  Much modern research on business cycles features “sticky” prices, and the technology sector includes several markets with natural constraints on monetary prices: these are not exactly regulated prices but potentially share many of their economic characteristics.

One view is that price regulations are, in the neighborhood of the unregulated price, more redistributive than they are socially costly even though they reduce the quantity traded. For example, a price ceiling is supposed to create a benefit for buyers that almost offsets the loss it imposes on sellers.  A number of studies have qualified these incidence and efficiency presumptions, noting that in addition to reducing supply, price ceilings may harm consumers by allocating the good to low-value customers (Barzel 1997, Glaeser and Luttmer 2003) or wasting consumer resources through queueing and search costs (Bulow and Klemperer 2012, Deacon and Sonstelie 1985).  But product-quality changes, which have been widely documented and explicitly considered in a few of the previous models of price regulation, are another concern.  Using a more general model of the technologically possible quality-quantity tradeoffs, our paper shows how a price ceiling imposed on a competitive market may increase the quantity traded, benefit producers at the expense of consumers, and have worse than first-order effects on efficiency – solely because the regulation affects non-price product attributes.  We also concisely characterize the features of tastes and technology that lead to such outcomes.

Practically all goods and services have non-price dimensions (hereafter, “quality”), with sellers often spending considerable amounts as they attempt to make their product more attractive to buyers. Non-price product attributes provide markets considerable scope for complying with a price ceiling without necessarily trading less quantity.  Take apartments, for which it is sometimes said that the purchase price of land and structure equals the expected present value of the rental income to be received from tenants.  In fact, about half of the revenues obtained from tenants is spent on short-run variable inputs rather than financing the structure’s purchase or construction.  Figure 1 shows the claims on national tenant-occupied housing output for 2006, as reported by Mayerhauser and Reinsdorf (2007).  Almost half of housing output went to intermediate goods and services (e.g., banking, realtor and advertising activities) and depreciation (a proxy for normal repairs and maintenance).  Another five percent went to labor (largely management), and about three percent went to compensate landlords for holding vacant units.  Landlords could adjust any of these items in order to reduce the ratio of costs to revenue.

Price ceilings have, in many real-world instances, increased quantity by reducing quality.  One is the case of doctor appointments, where ceilings on the price per visit have sometimes resulted in patients’ visiting the doctor more frequently for the same health condition.  As Frech (2001p. 338) puts it, Japanese patients “are often told to come back for return visits.  And, even injections of drugs were often split in half to make two visits necessary.”  Cuba, among other places, has ceilings the retail prices of eggs and other grocery items.  Even though the grocery-price ceilings are far below what the retail prices would be, the grocery quantities sold are not.  Instead, groceries there are sold in large containers and without refrigeration and other retail amenities.[1] Another example is the case of rent control of pre-war premises in Hong Kong, which appears to have increased the number of leases and perhaps even the number of square feet under lease as tenants engaged in partial subletting and landlords rented to “rooftop squatters.”  Elsewhere, rent control appears to increase the fraction of apartments that are under lease.

The quality-quantity tradeoff may be the source of these results. Holding expenditure constant, a price ceiling prohibits low quantities.  Take retail grocery sales.  Absent regulations, suppliers spend resources to preserve, cull, and promptly deliver their produce inventories so that the consumer receives fresh items.  With a price ceiling set on, say, a per-ounce basis, suppliers cut down on their quality-enhancing expenditures and thereby reduce the fraction of the groceries obtained by the consumer that is edible.  Consumers with a price-inelastic demand for edible groceries purchase more total groceries because the survival rate of purchased groceries is reduced by the price ceiling.  A variety of goods from apartments to light bulbs to doctor appointments have this feature that the unregulated market serves customers with less, but more expensive, quantity because that quantity is efficiently managed to provide the maximum value for the customer’s dollar.  Our model does not assume that controlled goods necessarily have such ease of substitution between quality and quantity, but these examples begin to show why the textbook predictions may not be reliable.

Even if a price ceiling does not increase the quantity sold, changes in non-price product attributes mean that the impacts of a ceiling on quantity and the surplus of buyers and sellers have little to do with the supply and demand for the controlled good by comparison to not having/producing the good at all.  On the demand side, it is not the same when price falls by regulation as when it changes due to a reduction in the marginal costs of producing the services delivered by the controlled good.  On the supply side, it is not the same when price falls by regulation as when it falls due to a reduction in the buyers’ marginal willingness to pay for the services delivered by the controlled good.  Even when the curves are properly adjusted to reflect changes in non-price attributes, the usual supply and demand diagram is not necessarily suitable for welfare analysis.

To the extent that supply slopes up, producers tend to benefit, relative to the unregulated allocation, from the increase in quantity and lose from the reduction in quality.  Indeed, we find a simple supply-elasticity condition that indicates whether a price ceiling net redistributes from consumers to producers, or vice versa.  For some of the same reasons, the possibility for producer gains is still present even when the equilibrium quantity impact of a price ceiling is not positive.

For conciseness, the scope of price regulations considered here is limited in three ways.  First, the rest of this paper refers to ceilings, but not floors.  Our framework applies to price floors too, but ignoring them removes numerous provisos, inversions, etc., from the discussion.  Also, the contrast between our results and previous ones are less subtle with ceilings than floors.  Second, we do not consider price ceiling regulations that also specify the amount supplied. Third, this paper features regulation-induced changes in non-price attributes that, holding price and expenditure constant, primarily affect the services consumers receive from the controlled good, rather than affecting the resources that the consumer has available for consuming other goods. The featured case encompasses the examples cited above: the price regulation is misspecified in the sense that it normalizes expenditure with a quantity (say, ounces of produce received from a retailer) that is different from what consumers ultimately value from the controlled good (edible ounces of produce).  In the latter model, not treated in this paper, the price regulation is misspecified in that some of the expenditure on the controlled good occurs downstream of the price regulation, so that compliance is achieved by moving production downstream.

In formulating a competitive hedonic model with a variable quantity but a lack of heterogeneity among producers and consumers, our goal is ease of exposition.  As with the textbook analysis of price ceilings and other public policies, we view the competitive case as a helpful starting point that focuses on tastes and technology, which by themselves have interesting and subtle features.  Instead, the standard hedonic-model framework with unit demand must be extended to allow for variable quantity in order to highlight quality-quantity tradeoffs that occur in the marketplace.  Having a homogenous group of consumers (producers) allows us to show that quality adjustments are distinct in principle from the question of who consumes (produces) in the regulated market, respectively.  Market power and heterogeneity can be added later, and we presume that doing so will only enrich the already surprising range of market outcomes that come from quality adjustments by themselves.

---

[1] Take eggs.  Expressed as a ratio to the price received by producers in the U.S. market (there is no reliable data for the Cuban producer prices because the Cuban government has vertically integrated the production chain), U.S. retail egg prices are about 3 whereas the Cuban retail price is controlled at about 4/3.  Controlled Cuban eggs are sold in trays of 30 without lids or refrigeration (Mulligan 2016) while Cuban egg consumption per capita is at the world average and above that in comparable countries such as the Dominican Republic. 

This paper was updated at nber.org in January 2017. The June 2016 version is here.

The Affordable Care Act and Marginal Tax Rates

I updated my ACA and Romneycare papers. The updates add the so-called "family glitch," Medicaid expansion, and end-of-year reconciliation of advance premium credits.

nber.org is hosting excel files with those updates and extensions (use the version with "update" in the file name).

The Marriage Tax and the Labor Tax

I received this question by email

If households respond to -- what I understand is -- a steep marriage penalty embedded in the subsidy formulae by postponing marriage or even by divorcing, would not this further broaden the effect of the Act in the labor market. In the limit, if every household sought subsidies under such an "income splitting" basis, would not more households qualify, and would not more households be influenced over a larger range of the income curve than even your study assumed? And thus would not the impact on the labor market be even larger [than calculated here]?

My answer:

It depends, even though I agree that the ACA includes a steep marriage tax as well as its steep taxes on labor income. My labor income MTR estimates are based on, among other things, estimates (from the CBO and others) of the number of people participating in the ACA exchanges. Your email provides a good reason to suspect that exchange participation will ultimately exceed the estimates. There are other reasons too.

Table 8 in my paper shows what would happen if one increased the exchange participation estimates -- see esp. the "exchange take-up" and "percentage of ESI moving to exchanges" rows.

But if the exchange participation estimates are accurate (maybe other factors offset the one you mentioned), then my headline estimates (Table 1) are fine even though the ACA reduces the incidence of marriage.

Ungated version of my marginal tax rate paper

http://caseybmulligan.com/MulliganMTRACA.pdf

excel spreadsheets with the marginal tax rate series are here (use the version with "update" in the file name).

The ARRA: Some Unpleasant Welfare Arithmetic

Food stamps, unemployment insurance, and other subsidies to persons who are unemployed and otherwise with low incomes, have recently been made more generous and available in more situations.  Did extra transfers help prevent a deeper recession, or did it amplify and prolong it?  Economists cannot fully answer these questions without examining the incentives of persons receiving the transfers.  The purpose of this paper is to quantify the number of people who recently had essentially no short-term financial reward from working, and how that number might have been different if safety net program rules had been made more generous, or if they had remained what they were in 2007.

American economists often discuss the unemployment insurance (hereafter, UI) system and its moral hazards as if the penalty for accepting a new job were about 50 percent of compensation,[1] which would suggest that the financial reward to working would be positive and significant in all but a few rare circumstances.  At the same time it is commonly noted that the average weekly unemployment benefit of about $300 barely exceeds the compensation from a full-time minimum wage job, and for this reason alone UI is almost always inferior to a real paycheck.  These claims are incorrect because they ignore payroll taxes, income taxes, and other safety net programs.  The tax arithmetic suggests that many UI participants would, even under 2007 rules and even ignoring all safety net programs aside from UI and the personal income tax, keep about 30 percent – and maybe as little as ten percent – of the compensation generated by accepting a new above-minimum-wage job because taxes typically took as much of the reward from working as foregone unemployment benefits did.  These thin margins essentially disappeared under the American Recovery and Reinvestment Act of 2009 (hereafter, ARRA).

Even when helping the poor is a primary policy motivation and the wage elasticity of labor supply is low, optimal tax theory frowns on labor income tax rates that equal or exceed one hundred percent (as long as work is not socially harmful) because at a one hundred percent rate there is no longer a tradeoff between efficiency and government revenue.  From a positive point of view, economists expect that employment rates will be low, if not zero, in groups of people who are aware that they receive no financial reward from working.  These are a couple of more reasons to quantify the prevalence of marginal tax rates that are near or exceed one hundred percent.[2]

The paper begins with a brief overview of the major safety net programs affecting the financial reward to working.  The first quantitative results are 2009 marginal tax rates and their components for some of the more common tax situations encountered by American workers and their families.  The rates are calculated for three scenarios: actual benefit and tax rules, benefit and tax rules as they would have been if they had not been changed since 2007, and benefit and tax rules as they might have been in a bigger stimulus.  The following section considers the rich and complicated variety of possible tax situations in order to arrive at estimates of the number of household heads and spouses with little or no financial reward to accepting a new job.  A “demand shocks and job search gambles” section shows how job acceptance rewards are nonlinear in the amount of a job offer, and the final section concludes.

Conclusions

            Before the recession began, going from unemployment back to work did not pay that well for someone eligible for unemployment benefits, but almost always paid a little something, with at least twenty percent of compensation from a job going toward enhancing the new employee’s disposable income above what it was during the spell.  Despite its inclusion of a “making work pay” tax credit and its expansion of the “earned income tax credit,” the ARRA increased marginal tax or “job acceptance penalty” rates for the vast majority of the unemployed and essentially erased the short-term financial benefits from working for two to three million non-elderly and unemployed household heads and spouses.  About five million had their job acceptance penalty rates increased above 80 percent by the ARRA.

Layoffs have also long been subsidized by unemployment insurance and other safety net programs, but again typically public treasuries would pay for less than 90 percent of the compensation lost from a layoff, while employer and employee had to absorb the rest.  When the ARRA was in full force, over three million workers could be laid off with a subsidy of 90 percent or more, and another five million with a subsidy rate of 80 to 89 percent.  A bigger stimulus would have put as many as 30 million workers in that situation.

To the degree that unemployment responds to the financial incentives for working, the ARRA and other programs assisting the unemployed interact with demand shocks in determining the number unemployed: an adverse demand shock increases unemployment more under the ARRA than it would if the same demand shock were experienced under 2007 tax and subsidy rules.

None of these results hinge on the increase of the duration of unemployment benefits from 26 to 99 weeks, which was achieved by legislation separate from the ARRA (United States Department of Labor 2011).  I count each unemployed person only when they are laid off; the results here reflect the level of benefits delivered by tax and subsidy programs to unemployed persons beginning to receive UI.  UI and other program eligibility rule changes are not considered in this paper but are important for quantifying changes in marginal tax rates between 2007 and 2009, and comparing such changes across demographic groups.

My findings of large, even confiscatory, job acceptance penalty rates are not the result of “cliffs” in transfer program formulas in which many dollars of benefits are lost for earning a particular marginal dollar (Yelowitz 1995) because I look at the consequence of more “discrete” decisions of accepting a job, or initiating a layoff, that change calendar year income by thousands of dollars.  Instead, my large rates reflect the combination of tax and subsidy rules, especially unemployment insurance.  Not surprisingly, my rate estimates exceed those of previous studies of transfer program marginal tax rates that omit unemployment insurance (Holt and Romich 2007) and exceed those of previous studies of unemployment insurance that ignored taxes (Chetty 2008).  But taxes, unemployment insurance, and other transfer programs have recently contributed significantly to the living standards of the poor and unemployed (Sherman 2011), so we cannot have a full understanding of the magnitude of marginal tax rates without considering the safety net broadly.

I have likely somewhat under-estimated the number of people with marginal tax rates in excess of one hundred percent because I have omitted a number of other possible sources of implicit taxes.  They include other means-tested cash assistance programs such as Disability Insurance, TANF and Supplemental Security Income; means-tested housing subsidies; means-tested tuition assistance; and means-tested energy assistance programs.  They also include court-enforced wage garnishment associated with the collection of delinquent consumer, tax, and child support debts.

            At the same time that incentives to retain and accept jobs were erased for millions, millions were laid off from their jobs and remained unemployed for an extended duration.  I estimate that 2.3 million additional non-elderly household heads and spouses were laid off in 2009 than would have been laid off if the 2000-2007 average number of layoffs had persisted through 2009.  The number of unemployed household heads and spouses were about 5 million greater than normal.  In other words, the extraordinary numbers of persons laid off and unemployed are of roughly the same magnitude as the numbers of persons having their incentives essentially erased by the ARRA.  The fact that more persons would have had incentives erased if the ARRA had been more generous to the unemployed suggests that it is possible that a bigger stimulus would have resulted in more unemployment than the actual stimulus did.

            It is beyond the scope of this paper to quantify the impacts that the large penalties for work from the ARRA (or other legislation) had on the labor market for people laid off during the recent recession.  Nor do I attempt to determine whether increasing marginal tax rates beyond 100 percent matters more or less than increasing them beyond, say, 70 percent.  But even before obtaining such estimates we should not expect that a labor market would function normally while the private benefit to working was zero or negative.  For this reason, the arithmetic presented in this paper is indeed unpleasant, and disturbingly similar to discredited welfare program rules of the distant past.  As James Tobin put it in 1965,

“[A 100 percent tax rate] does just that, causing needless waste and demoralization.  This application of the means test is bad economics as well as bad sociology.  It is almost as if our present programs of public assistance had been consciously contrived to perpetuate the conditions they are supposed to alleviate.” (Tobin 1965, 890)

---

[1] Chetty (2008) estimates the U.S. UI replacement rate as 50 percent for the purposes of demonstrating that it might be slightly less than optimal.  See also Fujita (2010).

[2] Behavior in the neighborhood of 100 percent tax rates would be especially interesting if it were true that (a) when tax rates are lower and more typical of their historical values, the amount of unemployment were insensitive to the amount of the UI benefits and (b) unemployment would be high if unemployment paid better than working.  To see this, try drawing a graph of the relationship between unemployment and the size of UI benefits that satisfies the properties (a) and (b): it must turn or jump sharply toward high unemployment as the benefit approaches the amount of pay from working.