Macro Roundup

“…Using a behavioral elasticity that is closer to what would apply to higher income individuals, this tax only generates an additional $5.4 billion in tax revenue in 2019. In other words, applying a mechanical 70 percent tax rate to ordinary income over $10 million generates nearly 89 percent more tax revenue from this group than under current law. Yet, accounting for behavioral responses decreases this revenue gain by 27.8 to 66.8 percent, depending upon the magnitude of the elasticity assumption….”

I. Introduction

The recent statement by Alexandria Ocasio-Cortez, the newly elected representative from New York, that we should tax those earning more than $10 million at a marginal tax rate of 70 percent in order to raise revenues for a Green New Deal, represents just the latest effort to make top income earners pay more by levying even higher taxes on them. In 2016, Hillary Clinton had proposed a surtax on taxpayers with incomes over $5 million, and Bernie Sanders proposed raising the top marginal income tax rate to 54.2 percent.

But will implementing higher tax rates really yield higher revenues? A popular presumption is that there is a mechanical link between higher tax rates and higher revenues. Unfortunately, the answer is not so clear-cut. What is clear is that individuals, particularly higher income individuals, respond to higher tax rates in a multitude of ways. They can shift incomes into non-taxable forms, they can postpone realizing income gains until a time when tax rates are lower, or they can engage in more tax planning activities that result in under-reported incomes. All of this means that the revenue impact of a tax hike is not a straightforward calculation of multiplying the new tax rate by the tax base above the income cutoff. The tax base, and reported taxable incomes, themselves can change.

Research into the behavioral responses to tax changes has been extensive in the field of public finance economics. The technical term for this behavioral response is the elasticity of taxable income (ETI). This elasticity captures the change in reported taxable incomes as tax rates change. It is perhaps not surprising that existing research finds that this elasticity is much higher near the top of the income distribution, meaning that higher-income taxpayers expend greater resources on reducing their tax burdens. That implies that revenue realizations may, in reality, be well below the projections of policymakers, which in turn affects the financing of social programs that are intended to rely on that source of funding. With ordinary income as the tax base, which is defined as wages, salaries, interest, and business income, implementing a 70 percent tax above $10 million would increase the tax revenue generated (assuming no behavioral response) in 2019 by $16.3 billion. However, assuming a conservative behavioral response, the revenue generated is $11.8 billion-27.8 percent less than the static estimation.

Using a behavioral elasticity that is closer to what would apply to higher income individuals, this tax only generates an additional $5.4 billion in tax revenue in 2019. In other words, applying a mechanical 70 percent tax rate to ordinary income over $10 million generates nearly 89 percent more tax revenue from this group than under current law. Yet, accounting for behavioral responses decreases this revenue gain by 27.8 to 66.8 percent, depending upon the magnitude of the elasticity assumption.

Using a dynamic model, the revenue increases are calculated for years until 2027. When applying the tax to only ordinary income, the cumulative revenue gains from 2019 to 2027 from this proposed tax are $212.9 billion under zero behavioral response, $155.3 billion when assuming a conservative behavioral response, and $74.7 billion when assuming a response that is more likely to apply to high-income individuals. In other words, applying a high tax rate on high-income individuals does not necessarily lead to much higher revenue gains in the long-run. Towards the end of this article, I discuss two alternatives to higher incomes tax rates. These are the consumption tax and the carbon tax.

II. Background

The basic question that economics research has tried to answer is how taxpayers respond when tax rates increase. The first area of study to estimate behavioral effects of taxation was on the responsiveness of the labor supply to the tax rate, which we can think of as the “real response” because it leads to real changes in behavior. The hypothesis was that as tax rates increase, the number of hours worked would decrease and thus dampen the revenue effects from taxation. Blundell and MaCurdy (1998) reviewed this literature and found that male labor is mostly unresponsive to changes in the tax rate. However, women, and more generally secondary earners, exhibit more responsiveness.

Moving beyond labor supply changes, a series of papers examined how total taxable income responds to tax changes. For instance, Lindsey (1987) used cross-sectional data around the Economic Recovery Tax Act of 1981 and estimated elasticities of adjusted gross income in the range of 1.05 to 2.75 with a central estimate of 1.6. Feldstein (1995) used panel data on individual tax returns around the Tax Reform Act of 1986 and found elasticities of taxable income (ETI) of 1.1-3.05 with a central estimate around 2.0. Auten and Carroll (1995) used the same regression methodology—but with a much larger panel of data available at the Treasury Department—and found a significantly lower central ETI estimate. To interpret these results simply, a higher elasticity number corresponds to a greater behavioral response.

In a recent review paper, Slemrod, Saez, and Giertz (2009) conclude that the earlier literature on the ETI, which focused on tax changes in the U.S. around the 1981 and 1986 tax reforms, found significantly larger estimates than more recent studies. They find that credible estimates of the long-run elasticity range from 0.12 to 0.4, with a central estimate of 0.25. However, they also find that the ETI is much higher for high-income individuals who have more access to tax avoidance opportunities, either through sophisticated tax planning, timing decisions, or through the use of deductions and credits.

III. The Elasticity of High-Income Taxpayers

The recent proposal from Rep. Ocasio-Cortez would only apply to the very top income taxpayers. So, it’s important to highlight that the relevant elasticity is the one that would apply to just these individuals. The ETI literature cited above has largely focused on low and middle-income households, with a few exceptions, and therefore the central ETI estimate of 0.25 is to be treated as a conservative estimate when studying higher income taxpayers who have many more opportunities to respond to changing tax rates. For instance, Goolsbee (2000) used panel data on executive compensation from 1991 to 1995 to study the responsiveness of executives’ taxable income to Omnibus Budget Reconciliation Act 1993. He calculated very high short-term elasticities of more than one, which he attributes to the exercise of options in anticipation of tax rate increases. He calculated a lower one-period-ahead elasticity of 0.4.

In a new National Bureau of Economic Research working paper that I co-authored (Gorry et al. 2018), we evaluate this elasticity by focusing on corporate executives and how their compensation responds to tax rates. Corporate executives are a subset of high-income taxpayers, but by some estimates (Carroll, 1998), are a large fraction of all high-income taxpayers. Executive compensation is often the subject of public debate, and so we focused our study on how this type of income responds to changes in tax rates. In addition, over the last several decades, the largest absolute changes in tax rates have taken place at the top of the income distribution, with smaller absolute changes for taxpayers in the middle of the income distribution. This also makes it a more feasible group to study empirically.

Over the last several decades, there has been dramatic growth in executive compensation, largely through the growth of stock options and other incentive-based pay. In an earlier paper (Gorry et al. 2017), we find that this change in the form of executive compensation can be linked to tax changes. For example, Section 162(m) of the Internal Revenue Code enacted in 1993, limited the tax deductibility of any non-performance-based pay over $1 million. This made cash compensation relatively more heavily taxed than stock options, which are inherently performance-based. Equity-based compensation is also advantageous for other reasons, such as the ability to avoid taxation in years with high top marginal tax rates through substitution of cash (salary) for equity-based compensation. In this way, executives can defer taxation on their income in high tax years because equity-based compensation is taxed only when realized (such as through exercise of stock options).

If current tax rates are high, then the share of stock options in the compensation package increases significantly, while the share of salary and other cash compensation decreases. This result implies that there are substantial long-run responses, beyond the exercising of stock options documented in earlier papers. The choice between realizing income and paying taxes today is significantly influenced by not only current personal income tax rates, but also capital gains tax rates and corporate tax rates in the future. This is relevant when thinking about behavioral responses to taxes, and especially the impact on revenues. If taxation is deferred to low-tax years, then revenues are likely to be lower than projected.

These types of responses of taxable income can be thought of as income-shifting responses. In other words, executives respond to anticipated tax rates by shifting income to periods of lower tax burdens. In addition to income shifting responses, executives may also have “real” responses in terms of reduced labor supply, both at the extensive and intensive margins. In Gorry et al. (2018), we use measured differences between the tax rates on current and deferred income to estimate ETI and decompose the total elasticity into real responses (reductions in total reported income due to a reduction in labor supply) and income-shifting responses across tax bases and over time. Our ETI estimate is above one, which is significantly higher than that found in prior studies. Further, decomposing this overall response, we find that much of the behavioral response to taxation comes from the income-shifting response, rather than the real response.

IV. Impact of ETI on Tax Revenues

Given the findings relating to the elasticity of taxable income specifically for high-income taxpayers, we can now try to forecast the revenue implications. Before I present my own analysis of the Ocasio-Cortez proposal, it is worthwhile to take a step back and see what the optimal revenue maximizing tax rate, r*, would be for the top tax bracket using the formula presented in Saez, Slemrod and Giertz (2009). This optimization rate can be written simply as:

To study the impact of the 70 percent tax rate on revenue gain, I use the Tax-Calculator developed at AEI, which uses the 2011 IRS-SOI Public Use File (PUF) and a recent Census Current Population Survey (CPS) and computes the federal income taxes and Federal Insurance Contribution Act (FICA) taxes for a sample of filing units, beginning in 2013. The model then creates a micro dataset that closely reproduces the multivariate distribution of income, deduction and credit items in 2009, and extrapolates to 2015-2027 levels in accordance with Congressional Budget Office (CBO) forecasts released in the spring of 2016. Additional information on non-filers is taken from the March 2013 Current Population Survey. The following estimates implement tax reforms and use Tax-Calculator version 0.22.2.

I begin by using total taxable income reported by those earning above $10 million as the tax base. The taxable income concept used here includes ordinary income, as well as capital gains and dividends income. While we apply the tax rate and the ETI to this income, it is important to remember that capital gains and dividends are taxed at different rates and the elasticity associated with that income has in some cases been estimated to be higher than 0.25 (Bogart and Gentry, 1993). An increase to a 70 percent tax for taxable income above $10 million decreases the net-of-tax rate on that portion of income by 52.38 percent. Assuming the elasticity of taxable income (ETI) to be 0.25, this implies that taxable income would fall by 13.14 percent. Assuming the elasticity of taxable income (ETI) to be 0.6, this implies that taxable income would fall even more, by 31.43 percent. Using the Open Source Policy Center’s Tax-Calculator model version 0.22.2, we estimate that the tax revenue generated on taxable income over $10 million dollars at the 2019 current law top rate of 37 percent is $104 billion.

Applying a 70 percent tax rate to income over $10 million and assuming no behavioral response, generates an additional $92.8 billion in tax revenue. However, assuming an elasticity of these individuals of 0.25 only generates $67 billion in revenue from a 70 percent tax—approximately $25 billion less than the static estimation. Furthermore, an even stricter behavioral assumption of an elasticity of 0.6 only generates $30.9 billion additional in tax revenue. In other words, applying a 70 percent tax rate to taxable income over $10 million generates nearly 89 percent more tax revenue from this group than under current law. Yet, accounting for behavioral responses decreases this revenue gain by 27.8 to 66.8 percent, depending upon the magnitude of the elasticity assumption.

We calculate the revenue increase from this reform in each year until 2027. Revenue increases annually as one would expect, with major increases in 2027, due to the expiration of many provisions of the Tax Cuts and Jobs Act that lessened the tax burden on the top income bracket from 2019-2026. From 2019 to 2027, as compared to current law baseline estimates, a 70 percent tax on income over $10 million would generate a revenue increase of approximately $1.03 trillion, assuming a static model with no behavioral responses. However, assuming an elasticity of taxable income equal to 0.25 for those earning above $10 million, the cumulative revenue increase is only $748.6 billion. Furthermore, assuming an even stronger response in which the elasticity of taxable income is 0.6, the provision only generates $358.1 billion.

What happens if we exclude capital gains and dividends from income? Using ordinary income, defined as wages, salaries, interest, and business income, implementing a 70 percent tax above $10 million would increase the tax revenue generated (assuming no behavioral response) by $16.3 billion. However, assuming these individuals respond to this reform such that their ETI is equal to 0.25, then only $11.8 billion is generated from this 70 percent tax-27.8 percent less than the static estimation. Furthermore, using an elasticity of 0.6, we see only an additional $5.4 billion in tax revenue. In other words, applying a 70 percent tax rate to ordinary income over $10 million generates nearly 89 percent more tax revenue from this group than under current law. Yet, accounting for behavioral responses decreases this revenue gain by 27.8 to 66.8 percent, depending upon the magnitude of the elasticity assumption.

The revenue increases are calculated for years until 2027. When applying the tax to only ordinary income, the results are evidently smaller than when applying the tax to all taxable income, as this is a smaller tax base. The cumulative revenue gain from 2019 to 2027 from this proposed tax is $212.9 billion under no behavioral assumptions, $155.3 billion when assuming an elasticity of 0.25, and $74.7 billion when assuming an elasticity of 0.6. It is likely that this second set of revenue estimates—which apply the rate increase only to ordinary income—is closer to the true value of the revenue gain from a 70 percent tax on income over $10 million. This is because a simple increase in the federal income tax rate would not actually mean that all taxable income above $10 million is subject to the higher rate.

Taxable income includes capital gains and dividends, which are taxed under a different rate and only upon realization. Therefore, individuals can defer realization in high tax years, which implies that this income base has a different elasticity. Nonetheless, we present results using both tax bases and find a similar narrative of revenue base erosion when accounting for behavioral responses—a key factor politicians often ignore when promoting the benefits of tax increases.

Along the same lines, recent modeling by the Tax Foundation shows that when accounting for both the elasticity of taxable income as well as the capital gains elasticity, the proposal would lose approximately $63.5 billion between 2019-2028 (Pomerleau and Li, 2019).

V. Alternative Solutions?

Given the uncertainty surrounding higher income taxation, an alternative is a progressive consumption tax. The problem with income taxation is that it penalizes saving and investment decisions. It may also influence choices relating to work and hours of work. However, a consumption tax moves away from that approach by taxing people’s expenditures rather than income. Typically, higher income households spend more on goods than lower income households. Therefore, by charging a tax on the items that people spend their income on, rather than all income, we do away with a tax on savings and investment. Such an idea (specifically an X-tax) has been discussed and developed in Viard and Carroll, among others.

As described by Carroll and Viard (2012), the X tax is a two-part tax, with separate components for households and businesses. The household component taxes only wages, leaving any income from saving untaxed. Higher-paid workers are taxed at higher rates. The business component (which applies to both corporations and non-corporate businesses) taxes business cash flow rather than income. The key difference is that, investment costs are immediately fully deductible rather than depreciated over a period of years. The tax rate on businesses would equal the top rate paid by the highest wage earners. The X tax would avoid the saving and investment penalties that are imposed by individual and corporate income taxes. In particular, the repeal of the corporate income tax would bring more capital to the U.S., which would increase labor productivity and wages.

Some estimates have found that replacement of the income tax system by a consumption tax may increase the long-run size of the U.S. economy by as much as nine percent (Auerbach, 1996), although the actual gains are likely to be smaller. In principle, the X tax can be as progressive, or more progressive, than the current tax system, by imposing high tax rates on businesses and high-paid workers. In practice, though, it may be difficult to match the current system’s measured progressivity for households at the very highest income levels, such as those in the top 0.1 percent (Carroll, Ganz, and Viard, 2008). In addition, the X tax may face political obstacles because there is no household tax on capital income.

Another idea worth considering is a carbon tax. This seems particularly relevant here, and one that may gain more support across both sides of the political aisle. The tax hike in Representative Ocasio-Cortez’s proposal is aimed at achieving a fundamental move towards cleaner energy and decarbonization. The Green New Deal states that within 10 years, the country will be fully powered by renewable energy sources. If the goal is to achieve greenhouse gas reductions and encourage investments in renewables, this can be achieved through a carbon tax.

As I have written earlier (Morris and Mathur, 2014), several studies show that a carbon tax can achieve environmental goals while at the same time raising significant amounts of revenue to fund other policy goals. A $16 tax per metric ton of CO2 could raise more than $1 trillion over 10 years. In addition, a carbon tax is likely to be more efficient than subsidies for clean energy technologies for several reasons. First, it is very hard to target subsidies toward the most cost-effective abatement, both because the government does not know which technologies will be most cost effective and because doing so can make a program prone to political favoritism. Second, it is nearly impossible to preclude subsidizing abatement that would happen anyway.

Clean energy subsidies can also have the perverse effect of increasing the overall supply of energy and making it cheaper, partly offsetting the benefits of the subsidies. In short, it is easier to be cost effective in discouraging things we do not want than encouraging things we do want. A carbon tax also promotes pollution-abating innovation, both because it increases the relative returns from new, clean energy technologies as compared to fossil-fuel dependent methods in a more predictable way and because it incentivizes innovation across an array of potential activities. A carbon tax can be straightforward to administer if designed properly.

A carbon tax assigns an exact dollar figure per metric ton of carbon, which places a greater level of transparency on U.S. efforts to reduce emissions, and may in turn foster more international agreements on environmental policies. And some economic research suggests that given the different structures of the uncertainties in the incremental benefits and costs of an extra ton of GHG abatement, setting a carbon price trajectory may be a better bet than setting annual country-level emissions targets. Of course, over the long run, it is important to ensure that cumulative emissions of all countries do not exceed levels that would risk undue damages.

A concern with the carbon tax is that a disproportionate share of the burden falls on lower income households. In my work with Adele Morris (2014), we show that the regressivity of a carbon tax can be offset by using 10 to 15 percent of the revenues to provide lump sum rebates to the bottom 20 percent of households. Another alternative is to use carbon tax revenues to expand a program like the Earned Income Tax Credit (Mathur and Morris, 2017), which is targeted at lower income households.

To conclude, there are several alternatives to taxing higher income households that would either yield more revenues or more directly get at spurious consumption undertaken by these households. A tax on the rich may be a winning political strategy, but there is little by way of sound economic policy to recommend it.

Aparna Mathur (@aparnamath) is a Resident Scholar in Economic Policy Studies at the American Enterprise Institute. Aparna would like to thank Erin Melly for excellent research assistance.

Aparna Mathur, “A 70% Tax Rate on the Rich May Be Smart Politics, But Is Not Smart Economics,”Bloomberg Tax, January 25, 2019, https://news.bloombergtax.com/daily-tax-report/insight-a-70-tax-rate-on-the-rich-may-be-smart-politics-but-is-not-smart-economics

A 70% Tax Rate on the Rich May Be Smart Politics, But Is Not Smart Economics

John Hartley, “Public Debt In A Time Of Growth (And Low Interest Rates,” Behavioral Macreconomics, January 20, 2019, http://behavioralmacroeconomics.blogspot.com/2019/01/public-debt-in-time-of-growth-and-low.htmlPublic Debt In A Time of Growth (And Low Interest Rates)

Jon

Thanks,

That's not to say, that higher levels of debt-to-GDP couldn't be sustainable (or that there wouldn't be good reasons like increased investments in public education), just that this offsetting "Blanchard" effect which is a feature of the "specialness" of government debt that disappears at a certain point (between 50 and 100% of GDP) and shouldn't be a point of justification for additional debt increases.

Absent any shocks to default expectations, the “Blanchard effect” predicated on r-g seems like a real phenomenon to be taken seriously in lower. There could be other reasons behind the historical post-war r-g < 0 phenomenon such as the high productivity of the 20thcentury which may not continue into the 21stcentury indefinitely perBob Gordon’s arguments.

Of course, if some event suddenly shifts expectations negatively toward an increased likelihood toward non-repayment of public debt, this sends r skyrocketing (the G7 examples exclude default cases like Greece and Argentina of the world, which hit default at very different levels of Debt-to-GDP).

The real interest rate however is largely invariant to Debt-to-GDP (historicallyreal interest rates haven't changed much over the past 700 years):

What is causing the Blanchard effect to be? More or less, it's the same story of Reinhart-Rogoff's "Growth In A Time of Debt" observation that growth begins declines non-linearly with GDP hitting a flashpoint somewhere around 100%:

It appears the "Blanchard effect" (where r < g and grow out debt) ends somewhere between 50 to 100% of GDP (depending on the country). The US, which has the world's reserve currency, appears to be something of an exception which might be at the higher end of that range.

Here is what the r-g differential looks like for the G7 countries at higher levels of debt-to-GDP:

We can certainly grow our way out of the existing stock of debt no problem as Blanchard observes, but what if deficits (G-T) persistently grow at rates higher than r-g? In other words, is r-g scalable at higher levels of debt-to-GDP? r-g aren't exogenous and r and g both depend a lot on the growth rate of G-T.

Paul Krugmanover at the NYTcorrectly observes and makes the related point that we didn't pay back the debt accumulated from WW2 (when debt-to-GDP was close to levels today around 100%) through taxes or spending cuts. Instead, the U.S. grew its way out through g, something that we we’re able to do in part because r-g < 0 for most of the 20thcentury (I’ll add while debt-to-GDP levels remained below 100% with a clear path to paying our way out.

It is true that there are only three ways to reduce public debt (other than default): 1) lowering fiscal deficits (G-T), 2) higher economic growth (g), and 3) seignorage (the latter of which we can and should ignore for inflationary reasons). While we often talk about #1 in fiscal policy/public debt discussions while #2 is either ignored or poorly quantified. Blanchard finally gives #2 the proper attention that it deserves.

If you attended the annual ASSA meetings this year, you may havewatchedorreadOlivier Blanchard's American Economic Association presidential address entitled "Public Debt and Low Interest Rates".

Blanchard observes that for most of the post-war period, real GDP growth (g) has been higher than real interest rates (r). He further makes the point that if real growth is higher than real interest rates (r-g < 0), we can grow our way out of an existing debt stock with relative ease. Blanchard argues this is reason for us to be "less worried" about public debt and that debt is "less bad" than we think. There’s also some nice modelling using a Diamond OLG framework if you have the time to read all the way through.

Before our lunch, wanted to send you mylatest blog postwhich is a rebuttal of the Blanchard argument that since r - g < 0 (real interest rate being less than real GDP growth implying we can grow our way out of debt), we can add more federal debt (Larry Summers and Jason Furman now have a Foreign Affairs piece out titled "Who's Afraid of Budget Deficits"echoingthis). Looking forward to lunch and would be great to hear your thoughts!:

Average Per Capita Income (2001-2017) By MSA:

Wage Growth (Annual % Change):

FRED, “Personal Income Per Capita,” Federal Reserve Bank Of St Lous, Accessed January 28, 2019, https://fred.stlouisfed.org/series/A792RC0A052NBEA

"....Many countries provide financial incentives to spur innovation, ranging from tax incentives to research and development grants. In this paper, we study how such financial incentives affect individuals' decisions to pursue careers in innovation. We first present empirical evidence on inventors' career trajectories and income distributions using de-identified data on 1.2 million inventors from patent records linked to tax records in the U.S. We find that the private returns to innovation are extremely skewed - with the top 1% of inventors collecting more than 22% of total inventors' income - and are highly correlated with their social impact, as measured by citations....The highest-paid 1% of inventors (whose annual incomes exceed $1.6 million per year) earn more than 22% of total inventors' income. The distribution of income among patent-holders is as skewed as the distribution of income in the population as a whole. Individuals with highly cited patents have much higher incomes, suggesting that the private benefits of innovation are correlated with their social returns...Inventors tend to have their most impactful innovations around age 40 and their incomes rise rapidly just before they have high-impact patents. We then build a stylized model of inventor career choice that matches these facts as well as recent evidence that childhood exposure to innovation plays a critical role in determining whether individuals become inventors. The model predicts that financial incentives, such as top income tax reductions, have limited potential to increase aggregate innovation because they only affect individuals who are exposed to innovation and have no impact on the decisions of star inventors, who matter most for aggregate innovation. Importantly, these results hold regardless of whether the private returns to innovation are known at the time of career choice. In contrast, increasing exposure to innovation (e.g., through mentorship programs) could have substantial impacts on innovation by drawing individuals who produce high impact inventions into the innovation pipeline. Although we do not present direct evidence supporting these model-based predictions, our results call for a more careful assessment of the impacts of financial incentives and a greater focus on alternative policies to increase the supply of inventors....Using this model, we compare the impacts of two types of policies on innovation: increasing private financial returns (e.g., by cutting top income tax rates) and increasing exposure (e.g., through mentorship programs). The model implies that the potential to increase innovation by reducing top taxes is limited, for three reasons.First, such policies only affect the subset of individuals who have been exposed to the possibility of an inventor career. Second, if the returns to innovation are forecastable at the point of career choice, such policies would only induce inventors of marginal quality to enter the field rather than star inventors. In our data, the mean annual income of those with patents in the top 1% of the citation distribution is more than $1 million between ages 40-50. The decisions of these star inventors are unlikely to be affected by small changes in financial incentives, making aggregate quality-weighted innovation relatively insensitive to tax rate...Third, if the returns to innovation are uncertain at the point of career choice, the elasticity of innovation with respect to top income tax rates is likely to be small in a standard expected utility model because tax changes only a ect payouts when inventors have very high incomes and low marginal utility...."

Alex Bell, Raj Chetty, Xavier Jaravel, Neviana Petkova and John Van Reenen, "Do Tax Cuts Produce More Einsteins? The Impacts of Financial Incentives vs. Exposure to Innovation on the Supply of Inventors," Center For Economic Performance, January 2019, http://cep.lse.ac.uk/_new/publications/abstract.asp