With elevated unemployment, estimates of how many jobs a given policy could create have become central to policy debates. This memo describes how the Economic Policy Institute estimates the number of jobs created by various fiscal and other policy changes that inject extra spending into the economy. In brief, EPI uses estimates of the total “fiscal impulse” created by a policy change (how much it adds to or subtracts from the federal budget deficit) and then applies macroeconomic “multipliers” from various sources to measure the impact of the fiscal impulse on economic output (gross domestic product, or GDP). Next, we translate the incremental gain or loss in GDP into the number of jobs supported (or lost) by this increased activity, using rules of thumb that, while simple, are firmly grounded in data and professional forecasting practice. Generally, such job estimates are only significant and relevant during times of elevated unemployment. However, given that unemployment rates have been historically high for years and threaten to be high for years to come, these types of job estimates will likely be useful to inform policy debates for quite some time.
Step one: Estimate the size of the fiscal impulse
The first step is to measure how much the policy increases (or decreases) the federal budget deficit. As explained in Bivens (2010), when the economy has idle resources, increases in the deficit boost spending and economic activity.
Data on the impact of specific policies on the budget deficit is generally pretty easy to find in either Congressional Budget Office documents that “score” policies, administration estimates, or other professional estimates of the fiscal impulse provided by policy proposals.
For the rest of this paper, we’ll use the extended unemployment insurance benefits proposed in the Obama administration’s American Jobs Act to illustrate the process.
The administration has estimated that extending this benefit will add (very roughly) $50 billion to the budget deficit for 2012.1 And we’re finished with step one—we have measured the size of the fiscal impulse. The only important thing to note is that it is increases in deficits that boost economic output closer to its potential while decreases in deficits contract output.
Step two: Translate the fiscal impulse to output gains
The second step is to apply a macroeconomic multiplier to the $50 billion fiscal impulse. Mark Zandi of Moody’s Analytics’ Economy.com (MAEC) has provided some of the most transparent estimated multipliers in this debate, and his overall forecasts for economic developments have been quite accurate in recent years.2 EPI often uses these MAEC multipliers. The Congressional Budget Office also has published a transparent, comprehensive set of multipliers. As shown in Table 1, the MAEC multipliers generally are very close to CBO’s, which signals their robustness. Further, the Council of Economic Advisers (CEA) also has estimated multipliers that, while providing less-detailed policy specifications, are very much in line with CBO and MAEC.
Multipliers, MAEC v CBO v CEA
Additionally, both the MAEC and CBO multipliers tend to closely agree with more generic, less specific multipliers estimated (or simulated) in academic research such as that by Woodford (2010) and Hall (2010). Lastly, the MAEC and CBO multipliers, when applied to fiscal impulses, tend to produce predictions of GDP gains (or losses) that fit very tightly within the overall range of private-sector forecasters’ assessments of various policies. (Figure A depicts general agreement among forecasters regarding the economic effect of the American Recovery and Reinvestment Act.)
In the case of extended unemployment insurance benefits, the $50 billion fiscal impulse is then applied to the estimated macroeconomic multiplier from MAEC. This multiplier is 1.6—meaning that the $50 billion fiscal impulse will generate $80 billion ($50 billion x 1.6) of additional economic activity. The economic intuition behind the math is simple—the bulk of unemployment insurance benefits will be spent on consumer goods such as food and clothing. This re-spending of the unemployment checks supports additional employment in grocery stores and retail clothing outlets. The grocery store cashiers who work more hours because of the demand supported by unemployment insurance checks then have additional income of their own, which they spend, say to buy a new car. The auto worker who works more hours because of increased demand also has more money. And the spending continues to ripple through the economy.
Of course, at each step along the way, some purchasing power leaks out of the economy. Some additional income is saved or taxed rather than spent and some of the dollars are spent on imports rather than items which boost U.S. GDP. This is why the multiplier is nowhere near infinite. The overall macroeconomic multiplier takes into account these “leakages” and shows the final boost to GDP to be expected from the fiscal impulse.
Multipliers and timing
The timing over which the multiplier works is not exact. Generally, a multiplier is assumed to represent a year’s time and no more. So, $50 billion in unemployment insurance benefit increases starting on Jan. 1, 2012, will boost GDP by $80 billion by the end of 2012. Of course, some of this spending power may spill over into 2013 or even beyond, but, the rough rule of thumb is that the multiplier represents the impact of the fiscal impulse over the year following its dispersal. Thus, one can generally interpret our example as indicating that the expansion of unemployment insurance benefits beginning in January 2012 will result in GDP in December 2012 that is $80 billion higher than would otherwise have been the case.
Further, the boosts to GDP provided by one-time fiscal impulses apply with much greater force when the economy is clearly operating below its productive potential (see text box for more on this issue) and do not directly lead to permanently higher growth rates of GDP. However, if this support takes the form of public investments (such as infrastructure improvements) then these investments could lead to higher growth rates, another reason why public investment is an especially attractive way of providing near-term economic support.
Important context for today’s debates: Everything besides finding a job is easier in a liquidity trap
A characteristic, and worrisome, feature of the current economic situation is that the U.S. economy is in a “liquidity trap” (or, alternatively, has reached the interest-rate “zero bound”). Essentially, this just means that the economy continues to have excess capacity (it is not producing as much as it has the capacity to produce) even though the Federal Reserve has moved the primary “policy” interest rate it traditionally lowers to spur economic activity all the way to zero. Nominal interest rates cannot be moved below zero because nobody would acquiesce to holding bonds in such a situation—instead they would just keep their money in cash.
This is a worrisome situation because the Fed’s control of these short-term “policy rates” is the most traditional, and most timely, way that policymakers fight recessions and sluggish growth. This policy lever now has been effectively exhausted and yet unemployment remains high. Policy must now resort to measures that are generally considered unorthodox, and this has proved to be a real problem for the political system—even clearly underpowered policy responses to the Great Recession and its aftermath have been fought by opponents claiming that the responses represent terrible policy overreach.
As much as the liquidity trap intensifies our economic plight, there is one slight upside of such situations—they greatly simplify estimating the impact of some policy changes, especially fiscal policy. Normally, the effect of fiscal policy changes on economic output is rather complex to estimate. Take the example of a temporary cut in the payroll tax rate that is deficit-financed. To get the net effect on economic output, one must measure how much of the payroll tax cut is saved versus spent, and how much is spent on domestic output versus imports. Next, one must measure the impact of the increased budget deficit on putting upward pressure on interest rates. Then, one must measure the sensitivity of private-sector spending to these higher interest rates and subtract, from the increased household spending arising from increased disposable personal income, the private-sector activity “crowded-out” by these higher interest rates.
In a liquidity trap, however, there is no upward pressure being exerted on interest rates by larger budget deficits. The essence of a liquidity trap is that the level of desired savings is so much greater than planned investment that the economy is “trying” to reach negative interest rates. So, under our example, unless the increase in the budget deficit resulting from the payroll tax cut was so large that it completely erased this gap between desired savings and planned investment (and to be clear, deficits over the past couple of years have been nowhere near this large), it will not put meaningful upward pressure on today’s interest rates. And, as a policy matter, the Federal Reserve would actually want interest rates to be lower than the zero rate we have today so would certainly push back against any upward pressure on interest rates that appeared.
Given today’s liquidity trap, the effects of fiscal policy changes become very easy to measure. Hence, the macroeconomic multipliers that we borrow to translate the fiscal impulse to increased economic output should be quite well-estimated. However, this also means that if the U.S. economy begins to move away from the zero lower-bound on interest rates, the size of these multipliers will likely become both smaller and less precisely estimated. This is why we are generally reluctant to estimate a jobs-created number for policies far in the future when the economy may well have escaped the liquidity trap.
Step three: Convert GDP changes to job changes
The last step is to convert the output changes to changes in overall employment. To do this, we rely on rules of thumb that are rough, but are firmly grounded in data and the practice of professional forecasters.
The most intuitive approach would be to simply assume that each 1 percent increase in GDP demands an extra 1 percent increase in labor hours. If this increase in aggregate labor hours came with no increase in the average length of a given worker’s workweek, then this would translate into a 1 percent increase in employment. However, coming out of a recession that saw the length of the average workweek fall significantly, it was likely the case that fiscal support provided during and after the Great Recession would see some of the output gains it spurred going to increased hours of incumbent employees rather than new hires.
Given that aggregate work hours fell by 30 percent more than aggregate employment, one possible rule of thumb would say that 30 percent of the increase in GDP spurred by output gains arising from fiscal support would be absorbed by longer hours, leaving 70 percent going to increase employment. This would translate into a 0.7 percent gain in employment for each 1 percent gain in GDP. However, this rule of thumb will also predict that each 1 percent gain in GDP should lead to a 1 percent gain in full-time equivalent employees (by definition, the workweek of full-time equivalents cannot change to soak up output gains).
Further, in the past, productivity growth has often accelerated as the economy emerged from recession. This faster productivity growth would result in incremental output gains that would not be met with increased employment but rather with greater efficiency. A rule of thumb factoring in productivity growth could say, for example, that if productivity rose 20 percent faster in the aftermath of recession, each 1 percent of GDP growth would yield only a 0.56 percent gain in employment (with 80 percent of the extra output reflecting changes in labor-hours, and 30 percent of these extra labor hours reflecting longer workweeks rather than new employees). However, in recent recessions productivity growth has been much less cyclical, so this rule of thumb may well be too cautious.
Note that the two considerations (variable length of workweeks and the pace of measured productivity growth) also imply that the rule of thumb linking GDP growth to jobs growth may well change as the economy enters different phases of the business cycle. For example, in the very early parts of recovery there may be great scope for productivity improvements and longer workweeks to absorb output gains. So, it may be that employment growth is very stubborn relative to output growth during these periods. As the length of the workweek and productivity growth tends to approach more-normal values, the employment gain associated with each increment of GDP can increase. This dynamic can explain most, for example, of the pattern displayed in CBO estimates of the GDP and employment gains spurred by ARRA. In 2009, a 1 percent increase in GDP is associated with only a 470,000 increase in employment. By 2012, however, this same 1 percent gain in GDP is associated with a 1.4 million increase in employment.
The correspondence between GDP and employment growth for 2011 that is identified by CBO in their estimates of ARRA’s impact is essentially the one that EPI is currently using. This implies that a 1 percent gain in GDP leads to a 0.9 percent gain in employment, or roughly 1.2 million workers. Previously we had used lower estimates of employment growth associated with each incremental GDP gain, reflecting our belief that both longer workweeks and more rapid measured productivity growth were absorbing a larger share of the output gains. We would also note that the rule of thumb is almost exactly what is implied in recent MAEC estimates of the relationship between employment and GDP growth for 2011. Lastly, this GDP-to-employment growth correspondence is very close to what the Council of Economic Advisers (CEA) has estimated for recent quarters as well (averaging 1.1 million jobs per 1 percent GDP growth in the last quarter of 2010 and first quarter of 2011). (Table 2 depicts the CBO, CEA, and MAEC calculations of the relationship between GDP and employment growth.)
GDP and Employment Changes, CBO and MAEC
What if you specify policies to make GDP more labor-intensive?
It should be noted that one can short-circuit this GDP-to-jobs rule of thumb by simply mandating that certain policies actually use a specified amount of money to create a given number of jobs. So, for example, one can say that the goal of a policy (say direct employment creation by government) is to create 100,000 jobs that pay $40,000 per year in wages and benefits, and hence this will cost $4 billion ($40,000 multiplied by 100,000). It seems to us that this is a generally fair thing to do—as long as the assumptions made are reasonable.
The average full-time equivalent “job” in the U.S. economy is associated with roughly $150,000 in overall GDP. Hence, a stimulus proposal that “costs” $150,000 per job it creates is actually perfectly in line with economy-wide averages. Yet most people don’t realize this. But, besides the wage paid directly to a worker, one needs to factor in nonwage benefits (which tend to inflate compensation to about 25 percent over wage costs), the profits generated by a worker for the firm they work for, the rent that must be paid to landlords for the workplace, and the interest costs that must be paid back by firms that have borrowed to provide the capital equipment used by a worker in performing the job.
Further, stating that one supports a policy that creates more jobs per unit of GDP than other policies is essentially stating that the policy you support is one of creating low-productivity jobs. This, to be clear, is not necessarily a bad thing—low productivity jobs can be a useful way to soak up idle workers. But it would be fair to ask, per the example above, that if a $4 billion policy is creating jobs for 100,000 workers paid $40,000 per year, “what exactly are these workers doing with no boss, no management, no physical workplace, and no equipment, computers, shovels, uniforms etc. …?”
In the end, although specifying the creation of low-productivity jobs to soak up excess labor is a perfectly fine policy proposal, one should still think hard about the realism of the GDP-to-jobs numbers that one is claiming on behalf of these proposals. And, when making assessments about the cost-effectiveness of stimulus proposals based on the GDP cost per job, one must remember what the economy-wide averages really look like (i.e., much larger than just an average salary).
Why do estimates of analysts sometimes differ?
Generally, EPI estimates of the jobs created by fiscal policy changes are closely in line with others’. Sometimes, however, they differ by significant amounts. The most important thing to remember about this is that the chaining-together of various effects can turn small differences in each calculation along the way into larger end-number differences. These differences will be most pronounced for those generally rare policy changes for which different analysts have significantly different multipliers.
For example, the CBO and MAEC multipliers for identical policies average only 10 percent differences. But for a particularly important policy—the effect of payroll tax cuts—the multipliers differ by essentially 2-to-1. So, if one analyst used the MAEC multiplier (1.1) when estimating the effect of $50 billion in employee-side payroll tax cuts while another analyst used those from the CBO (0.6), the GDP impacts would different by essentially 2-to-1. Right off the bat this means that we’ll find job impacts nearly twice as large using one set of multipliers than another.
Further, different analysts use different rules of thumb to translate GDP changes into employment changes. For example, if one used the correspondence used by the Obama administration in estimating the effect of ARRA on job growth, which says that 1 percent of GDP yields 1 million jobs instead of EPI’s current rule that 1 percent of GDP yields an estimated 1.2 million jobs, it would add another 14 percent difference to estimates of jobs created by given policies.
Timing issues can also cause some discordance in estimates. Some of these timing differences will simply be about when a policy change actually delivers a fiscal impulse. Another, more subtle, difference will concern when within a year the fiscal impulse happens and whether or not job and/or GDP effects are estimated as averages over a whole calendar year or simply as the effect of policy one full year after implementation.
If, for example, an increase of $50 billion in infrastructure spending is legislated to take effect January 1, 2012, but lags in implementation, meaning very little is spent in the first and second quarters and much more is spent in the third and fourth quarters, the measured effect of this policy will vary greatly depending on whether one measures the calendar year average or the end-of-year effect.
As an example, take the following numbers from the CBO on the impact of the ARRA for 2009 (Table 3). According to the calendar-year average, ARRA raised GDP by an estimated 1.4 percent and created 700,000 jobs. Yet if one measured the impact of the stimulus as the difference in GDP and employment at the end of 2009 (i.e., approximately one year after its passage) one would find that it raised GDP by 2.5 percent and boosted employment by 1.4 million people.
GDP and Employment Changes, CBO and MAEC
Are your estimates “jobs” or “job-years”?
Another confusion that sometimes arises in debates over the job-creation impacts of different policies is how this job creation is specified. Job-creation impacts are most commonly expressed as changes in the level of employment at a point in time relative to a counterfactual where the policy was not enacted, or by how many “job-years” would be created relative to the no-policy counterfactual.
The first specification—the change in the level of employment—just expresses the effect of the policy as its impact on the number of jobs in the economy at a given point in time. So, if we are estimating the impact of $50 billion in extended unemployment insurance benefits between January 1, 2012, and December 31, 2012, we would project December 2012 employment levels.
This, for example, is how the CBO has generally expressed its assessment of the effect of ARRA; according to CBO, employment levels at the end of 2010 were 2.4 million higher than they would have been absent ARRA.
“Job-years,” on the other hand, combine data on the impact of a policy on the employment level at different points in time to get a measure of how many job-years are created. So, for example, if a policy change led employment levels to be higher by 100,000 at the end of 2012 than they would have been without the policy (i.e., the “no-policy counterfactual”) and also by 100,000 higher at the end of 2013 relative to a no-policy counterfactual, one can say that the policy resulted in 200,000 job-years.
To go back to ARRA, CBO estimates that by the end of 2009 ARRA had led to an additional 1.4 million jobs. This means that the number of job-years spurred by ARRA by the end of 2010 was clearly higher than the 2.4 million increase in the employment level reported by CBO. If ARRA’s impact had begun on January 1, 2009, one could just add the 1.4 million increase in jobs by the end of 2009 to the 2.4 million increase by the end of 2010 to get 3.8 million job-years supported by ARRA by the end of 2010.
Impact on levels versus rates of change
Another common source of confusion related to issues of timing concerns the effect of fiscal policy changes on the level of output and employment versus its rate of change. What matters for measuring the impact of a policy on the level of output or employment is simply the flow of spending in the time period examined. As long as this flow is positive, then levels of output and employment will be higher than they otherwise would have been. What matters for measuring the impact of a policy on the rate of change of output or employment is how much greater or smaller this period’s flow of spending is compared with the last period.
So, again using CBO data on the effectiveness of ARRA, we present the CBO’s estimates on the impact of ARRA on GDP levels in each quarter examined. Then, we calculate a counter-factual “no ARRA” level of GDP for each quarter and then calculate the growth rate of actual GDP and the path that GDP would have followed without ARRA. Figure B shows these two paths, with the horizontal axis at zero representing the path without ARRA: It is clear that the level of GDP is higher in each period (at least through the second quarter 2011—and it is never lower), but, that ARRA only appreciably boosted the growth rate for the first four quarters of implementation. The reason for this is that while ARRA spend-out continues even today, the rate of spend-out is falling, and it is this rate of spend-out that determines its impact on the growth-rate.
This helps to clarify something important: Analysts in recent months have frequently pointed to the “fiscal drag” on growth stemming from the winding down of spending related to ARRA (as well as other fiscal drags). This might make some people think that somehow ARRA was bad for the economy, or, at least neutral—that it boosted growth for a year and has suppressed it since. This impression would be incorrect. ARRA was clearly good for the economy; we have higher incomes today (and have had higher incomes since its implementation) because of ARRA, and our income levels will never be lowered because of it. But, because the rate of spending has increased, this means that ARRA is actually leading to lower growth-rates in the coming quarters than would have occurred absent its passage.
1. The $50 billion figure is used in this paper as an example. A new report from the Congressional Budget Office (2011b) provided a spending figure of $44 billion for the unemployment insurance provisions in the American Jobs Act of 2011. There is also an additional $1 billion for the cost of suspending the Extended Benefit “look back.”
2. Measured relative to other forecasters—just about all forecasting firms and institutions missed the severity of the Great Recession.
Bivens, Josh. 2010. Budget for Recovery: The Need to Increase the Federal Deficit to Revive a Weak Economy. Economic Policy Institute Briefing Paper #253. Washington, D.C.: EPI.
Congressional Budget Office. 2011a. Estimated Impact of the American Recovery and Reinvestment Act on Employment and Economic Output from April 2011 Through June 2011. http://www.cbo.gov/ftpdocs/123xx/doc12385/08-24-ARRA.pdf
Congressional Budget Office. 2011b. Cost Estimate for S. 1549, American Jobs Act of 2011. Washington, D.C.: CBO, October. http://cbo.gov/ftpdocs/124xx/doc12470/s1549.pdf
Council of Economic Advisers. 2011. The Economic Impact of the American Recovery and Reinvestment Act of 2009 Seventh Quarterly Report. Washington, D.C.: Executive Office of the President. http://www.whitehouse.gov/sites/default/files/cea_7th_arra_report.pdf
Hall, Robert. 2010. “By How Much Does GDP Rise if the Government Buys More Output?” Working Paper. Palo Alton, Calif.: Standford University.
Zandi, Mark. 2011. “U.S. Macro Outlook: Compromise Boosts Stimulus.” Moody’s Analytics Economy.com. http://www.economy.com/dismal/article_free.asp?cid=195470
Woodford, Michael. 2010. “The Simple Analytics of the Government Expenditure Multiplier.” Paper prepared for an Allied Social Science Associations meeting, Atlanta, Ga., Jan. 3–5. http://www.columbia.edu/~mw2230/G_ASSA.pdf