Taxes to Pay for Highways: A Switch from the Tax on Gallons of Fuel Burned to a Tax on Miles Driven Would Be Stupid

Impact of Switching from Fuel Tax on Gallons Burned to Tax on Miles Driven

A.  Introduction

According to a recent report in the Washington Post, a significant and increasing number of state public officials and politicians are advocating for a change in the tax system the US uses to support highway building and maintenance.  The current system is based on a tax on gallons of fuel burned, and the proposed new system would be based on the number of miles a car is driven.  At least four East Coast states are proposing pilots on how this might be done, some West Coast states have already launched pilots, and states are applying for federal grants to consider the change.  There is indeed even a lobbying group based in Washington now advocating it:  The Mileage-Based User Fee Alliance.

There is no question that the current federal gas tax of 18.4 cents per gallon of gasoline is woefully inadequate.  It was last changed in 1993, 23 years ago, and has been kept constant in nominal terms ever since.  With general prices (based on the CPI) now 65% higher, 18.4 cents now will only buy 11.2 cents at the prices of 1993, a decline of close to 40%.  As a result, the Highway Trust Fund is terribly underfunded, and with all the politics involved in trying to find other sources of funding, our highways are in terrible shape. Basic maintenance is simply not being done.

An obvious solution would be simply to raise the gas tax back at least to where it was before in real terms.  Based on where the tax was when last set in 1993 and on the CPI for inflation since then, this would be 30.3 cents per gallon now, an increase of 11.9 cents from the current 18.4 cents per gallon.  Going back even further, the gasoline tax was set at 4 cents per gallon in 1959, to fund the construction of the then new Interstate Highway system (as well as for general highway maintenance).  Adjusting for inflation, that tax would be 32.7 cents per gallon now.  Also, looking at what the tax would need to be to fund adequately the Highway Trust Fund, a Congressional Budget Office report issued in 2014 estimated that a 10 to 15 cent increase (hence 28.4 cents to 33.4 cents per gallon) would be needed (based on projections through 2024).

These fuel tax figures are all similar.  Note also that while some are arguing that the Highway Trust Fund is underfunded because cars are now more fuel efficient than before, this is not the case.  Simply bringing the tax rate back in real terms to where it was before (30.3 cents based on the 1993 level or 32.7 cents based on the 1959 level) would bring the rate to within the 28.4 to 33.4 cents range that the CBO estimates is needed to fully fund the Highway Trust Fund.  The problem is not fuel efficiency, but rather the refusal to adjust the per gallon tax rate for inflation.

But Congress has refused to approve any such increase.  Anti-tax hardliners simply refuse to consider what they view as an increase in taxes, even though the measure would simply bring them back in real terms to where they were before.  And it is not even true that the general population is against an increase in the gas tax.  According to a poll sponsored by the Mineta Transportation Institute (a transportation think tank based at San Jose State University in California), 75% of those polled would support an immediate increase in the gas tax of 10 cents a gallon if the funds are dedicated to maintenance of our streets, roads, and highways (see the video clip embedded in the Washington Post article, starting at minute 3:00).

In the face of this refusal by Congress, some officials are advocating for a change in the tax, from a tax per gallon of fuel burned to a new tax per mile each car is driven.  While I do not see how this would address the opposition of the anti-tax politicians (this would indeed be a totally new tax, not an adjustment in the old tax to keep it from falling in real terms), there appears to be a belief among some that this would be accepted.

But even if such a new tax were viewed as politically possible, it would be an incredibly bad public policy move to replace the current tax on fuel burned with such a tax on miles driven.  It would in essence be a tax on fuel efficiency, with major distributional (as well as other) consequences, favoring those who buy gas guzzlers.  And as it would encourage the purchase of heavy gas guzzlers (relative to the policy now in place), it would also lead to more than proportional damage to our roads, meaning that road conditions would deteriorate further rather than improve.

This blog post will discuss why such consequences would follow.  To keep things simple, it will focus on the tax on gasoline (which I will sometimes simply referred to as gas, or as fuel).  There are similar, but separate taxes, on diesel and other fuels, and their levels should be adjusted proportionally with any adjustment for gasoline.  There is also the issue of the appropriate taxes to be paid by trucks and other heavy commercial vehicles.  That is an important, but separate, issue, and is not addressed here.

B.  The Proposed Switch Would Penalize Fuel Efficient Vehicles

The reports indicate that the policy being considered would impose a tax of perhaps 1.5 cents per mile driven in substitution for the current federal tax of 18.4 cents per gallon of gas burned (states have their own fuel taxes in addition, with these varying across states). For the calculations here I will take the 1.5 cent figure as the basis for the comparisons, even though no specific figure is as yet set.

First of all, it should be noted that at the current miles driven in the country and the average fuel economy of the stock of cars being driven, a tax of 1.5 cents per mile would raise substantially more in taxes than the current 18.4 cents per gallon of gas.  That is, at these rates, there would be a substantial tax increase.

Using figures for 2014, the average fuel efficiency (in miles per gallon) of the light duty fleet of motor vehicles in the US was 21.4 miles per gallon, and the average miles driven per driver was 13,476 miles.  At a tax of 1.5 cents per mile driven, the average driver would pay $202.14 (= $.015 x 13,476) in such taxes per year.  With an average fuel economy of 21.4 mpg, such a driver would burn 629.7 gallons per year, and at the current fuel tax of 18.4 cents per gallon, is now paying $115.87 (= $.184 x 629.7) in gas taxes per year. Hence the tax would rise by almost 75% ($202.14 / $115.87).  A 75% increase would be equivalent to raising the fuel tax from the current 18.4 cents to a rate of 32.1 cents per gallon.  While higher tax revenues are indeed needed, why a tax on miles driven would be acceptable to tax opponents while an increase in the tax per gallon of fuel burned is not, is not clear.

But the real reason to be opposed to a switch in the tax to miles driven is the impact it would have on incentives.  Taxes matter, and affect how people behave.  And a tax on miles driven would act, in comparison to the current tax on gallons of fuel burned, as a tax on fuel efficiency.

The chart at the top of this post shows how the tax paid would vary across cars of different fuel efficiencies.  It would be a simple linear relationship.  Assuming a switch from the current 18.4 cents per gallon of fuel burned to a new tax of 1.5 cents per mile driven, a driver of a highly fuel efficient car that gets 50 miles per gallon would see their tax increase by over 300%!  A driver of a car getting the average nation-wide fuel efficiency of 21.4 miles per gallon would see their tax increase by 75%, as noted above (and as reflected in the chart).  In contrast, someone driving a gas guzzler getting only 12 miles per gallon or less, would see their taxes in fact fall!  They would end up paying less under such a new system based on miles driven than they do now based on gallons of fuel burned.  Drivers of luxury sports cars or giant SUVs could well end up paying less than before, even with rates set such that taxes on average would rise by 75%.

Changing the tax structure in this way would, with all else equal, encourage drivers to switch from buying fuel efficient cars to cars that burn more gas.  There are, of course, many reasons why someone buys the car that they do, and fuel efficiency is only one.  But at the margin, changing the basis for the tax to support highway building and maintenance from a tax per gallon to a tax on miles driven would be an incentive to buy less fuel efficient cars.

C.  Other Problems

The change to a tax on miles driven from the tax on gallons of fuel burned would have a number of adverse effects:

a)  A Tax on Fuel Efficiency:  As noted above, this would become basically a tax on fuel efficiency.  More fuel efficient cars would pay higher taxes relative to what they do now, and there will be less of an incentive to buy more fuel efficient cars.  There would then be less of an incentive for car manufacturers to develop the technology to improve fuel efficiency.  This is what economists call a technological externality, and we all would suffer.

b)  Heavier Vehicles Cause Far More Damage to the Roads:  Heavier cars not only get poorer gas mileage, but also tear up the roads much more, leading to greater maintenance needs and expense.  Heavier vehicles also burn more fuel, but there is a critical difference.  As a general rule, vehicles burn fuel in proportion with their weight: A vehicle that weighs twice as much will burn approximately twice as much fuel.  Hence such a vehicle will pay twice as much in fuel taxes (when such taxes are in cents per gallon) per mile driven.

However, the heavier vehicle also cause more damage to the road over time, leading to greater maintenance needs.  And it will not simply be twice as much damage.  A careful early study found that the amount of damage from a heavier vehicle increases not in direct proportion to its weight, but rather approximately according to the fourth power of the ratio of the weights.  That is, a vehicle that weighs twice as much (for the same number of axles distributing the weight) will cause damage equal to 2 to the fourth power (=16) times as much as the lighter vehicle.  Hence if they were to pay taxes proportionate to the damage they do, a vehicle that is twice as heavy should pay 16 times more in taxes, not simply twice as much.

(Note that some now argue that the 2 to the fourth power figure found before might be an over-estimate, and that the relationship might be more like 2 to the third power.  But this would still imply that a vehicle that weighs twice as much does 8 times the damage (2 to the third power = 8).  The heavier vehicle still accounts for a grossly disproportionate share of damage to the roads.)

A tax that is set based on miles driven would tax heavy and light vehicles the same.  This is the opposite of what should be done:  Heavy vehicles cause far more damage to the roads than light vehicles do.  Encouraging heavy, fuel-thirsty, vehicles by switching from a tax per gallon of fuel burned to a tax per mile driven will lead to more road damage, and proportionately far more cost than what would be collected in highway taxes to pay for repair of that damage.

c)  Impact on Greenhouse Gases:  One also wants to promote fuel efficiency because of the impact on greenhouse gases, and hence global warming, from the burning of fuels. By basic chemistry, carbon dioxide (CO2) is a direct product of fuel that is burned.  The more fuel that is burned, the more CO2 will go up into the air and then trap heat. Economists have long argued that the most efficient way to address the issue of greenhouse gases being emitted would be to tax them in proportion to the damage they do.  A tax on gallons of fuel that are burned will do this, while a tax on miles driven (and hence independent of the fuel efficiency of the vehicle) will not.

An interesting question is what level of gasoline tax would do this.  That is, what would the level of fuel tax need to be, for that tax to match the damage being done through the associated emission of CO2.  The EPA has come up with estimates of what the social cost of such carbon emissions are (and see here for a somewhat more technical discussion of its estimates).  Unfortunately, given the uncertainties in any such calculations, as well as uncertainty on what the social discount rate should be (needed to discount costs arising in the future that follow from emitting greenhouse gases today), the cost range is quite broad. Hence the EPA presents figures for the social cost of emitting CO2 using expected values at alternative social discount rates of 2.5%, 3%, and 5%, as well as from a measure of the statistical distribution of one of them (the 95th percentile for the 3% discount rate, meaning there is only an estimated 5% chance that the cost will be higher than this).  The resulting costs per metric ton of CO2 emitted then range from a low of $11 for the expected value (the 50th percentile) at the 5% discount rate, $36 at the expected value for the 3% discount rate, and $56 for the expected value for the 2.5% discount rate, to $105 for the 95th percentile at a 3% discount rate (all for 2015).

With such range in social costs, one should be cautious in the interpretation of any one. But it may still be of interest to calculate how this would translate into a tax on gasoline burned by automobiles, to see if the resulting tax is “in the ballpark” of what our fuel taxes are or should be.  Every gallon of gasoline burned emits 19.64 pounds of CO2.  There are 2,204.62 pounds in a metric ton, so one gallon of gas burned emits 0.00891 metric tons of CO2.  At the middle social cost of $36 per metric ton of CO2 emitted (the expected value for the 3% social discount rate scenario), this implies that a fuel tax of 32.1 cents per gallon should be imposed.  This is surprisingly almost precisely the fuel tax figure that all the other calculations suggest is warranted.

d)  One Could Impose a Similar Tax on Electric Cars:  One of the arguments of the advocates of a switch from taxes on fuel burned to miles driven is that as cars have become more fuel efficient, they pay less (per mile driven) in fuel taxes.  This is true.  But as generally lighter vehicles (one of the main ways to improve fuel economy) they also cause proportionately far less road damage, as discussed above.

There is also an increasing share of electric, battery-powered, cars, which burn no fossil fuel at all.  At least they do not burn fossil fuels directly, as the electricity they need to recharge their batteries come from the power grid, where fossil fuels dominate.  But this is still close to a non-issue, as the share of electric cars among the vehicles on US roads is still tiny.  However, the share will grow over time (at least one hopes).  If the share does become significant, how will the cost of building and maintaining roads be covered and fairly shared?

The issue could then be addressed quite simply.  And one would want to do this in a way that rewards efficiency (as different electric cars have different efficiencies in the mileage they get for a given charge of electricity) rather than penalize it.  One could do this by installing on all electric cars a simple meter that keeps track of how much it receives in power charges (in kilowatt-hours) over say a year.  At an annual safety inspection or license renewal, one would then pay a tax based on that measure of power used over the year.  Such a meter would likely have a trivial cost, of perhaps a few dollars.

Note that the amounts involved to be collected would not be large.  According to the 2016 EPA Automobile Fuel Economy Guide (see page 5), all-electric cars being sold in the US have fuel efficiencies (in miles per gallon equivalent) of over 100 mpg, and as high as 124 mpg.  These are on the order of five times the 21.4 average mpg of the US auto stock, for which we calculated that the average tax to be paid would be $202.  Even ignoring that the electric cars will likely be driven for fewer miles per year than the average car (due to their shorter range), the tax per year commensurate with their fuel economy would be roughly $40.  This is not much.  It is also not unreasonable as electric cars are kept quite light (given the limits of battery technology) and hence do little road damage.

e)  There Are Even Worse Policies That Have Been Proposed:  As discussed above, there are many reasons why a switch from a tax on fuel burned to miles driven would be a bad policy change.  But it should be acknowledged that some have proposed even worse. One example is the idea that there should be a fixed annual tax per registered car that would fund what is needed for highway building and maintenance.  Some states in fact do this now.

The amounts involved are not huge.  As was calculated above, at the current federal gasoline tax of 18.4 cents per gallon, the driver of a car that gets the average mileage (of 21.4 mpg) for the average distance a year (of 13,476 miles) will pay $115.87 a year.  If the fuel tax were raised to 32.1 cents per gallon (or equivalently, if there were a tax of 1.5 cents per mile driven), the average tax paid would be still just $202.14 per year.  These are not huge amounts.  One could pay them as part of an annual license renewal.

But the tax structured in this way would then be the same for a driver who drives a fuel efficient car or a gas guzzler.  And it would be the same for a driver who drives only a few miles each year, or who drives far more than the average each year.  The driver of a heavy gas guzzler, or one who drives more miles each year than others, does more damage to the roads and should pay more to the fund that repairs such damage and develops new road capacity.  The tax should reflect the costs they are imposing on society, and a fixed annual fee does not.

f)  The Cost of Tax Collection Needs to be Recognized:  Finally, one needs to recognize that it will cost something to collect the taxes.  This cost will be especially high for a tax on miles driven.

The current system, of a tax on fuel burned, is efficient and costs next to nothing to collect.  It can be charged at the point where the gasoline and other fuels leaves in bulk from the refinery, as all of it will eventually be burned.  While the consumer ultimately pays for the tax when they pump their gas, the price being charged at the pump simply reflects the tax that had been charged at an earlier stage.

In contrast, a tax on miles driven would need to be worked out at the level of each individual car.  And if the tax is to include shares that are allocating to different states, the equipment will need to keep track of which states the car is being driven in.  As the Washington Post article on a possible tax on miles driven describes, experiments are underway on different ways this might be done.  All would require special equipment to be installed, with a GPS-based system commonly considered.

Such special equipment would have a cost, both up-front for the initial equipment and then recurrent if there is some regular reporting to the center (perhaps monthly) of miles driven.  No one knows right now what such a system might cost if it were in mass use, but one could easily imagine that a GPS tracking and reporting system might cost on the order of $100 up front, and then several dollars a month for reporting.  This would be a significant share of a tax collection that would generate an average of just $202 per driver each year.

There is also the concern that any type of GPS system would allow the overseers to spy on where the car was driven.  While this might well be too alarmist, and there would certainly be promises that this would not be done, some might not be comforted by such promises.

D.  Conclusion

While one should always consider whether given policies can be changed for the better, one needs also to recognize that often the changes proposed would make things worse rather than better.  Switching the primary source of funding for highway building and maintenance from a tax on fuel burned to a tax on miles driven is one example.  It would be a stupid move.

There is no doubt that the current federal tax on gasoline of 18.4 cents per gallon is too low.  The result is insufficient revenues for the Highway Trust Fund, and we end up with insufficient road capacity and roads that are terribly maintained.

What I was surprised by in the research for this blog post was finding that a wide range of signals all pointed to a similar figure for what the gasoline tax should be. Specifically:

  1. The 1959 gas tax of 4 cents per gallon in terms of current prices would be 32.7 cents per gallon;
  2. The 1993 gas tax of 18.4 cents per gallon in terms of current prices would be 30.3 cents per gallon;
  3. The proposal of a 1.5 cent tax per mile driven would be equivalent (given current average car mileage and the average miles driven per year) to 32.1 cents per gallon;
  4. The tax to offset the social cost of greenhouse gas emissions from burning fuel would be (at a 3% social discount rate) 32.1 cents per gallon.
  5. The Congressional Budget Office projected that the gasoline tax needed to fully fund the Highway Trust Fund would be in the range of 28.4 to 33.4 cents per gallon.

All these point in the same direction.  The tax on gasoline should be adjusted to between 30 and 33 cents per gallon, and then indexed for inflation.

The Rate of Return on Funds Paid Into Social Security Are Actually Quite Good

Social Security Real Rates of Return - Various Scenarios

 

A.  Introduction

The rate of return earned on what is paid into our Social Security accounts is actually quite good.  It is especially good when one takes into account that these are investments in safe assets, and thus that the proper comparison should be to the returns on other safe assets, not risky ones.  Yet critics of Social Security, mostly those who believe it should be shut down in its current form with some sort of savings plan invested through the financial markets (such as a 401(k) plan) substituted for it, often assert that the returns earned on the pension savings in Social Security are abysmally poor.

These critics argue that by “privatizing” Social Security, that is by shifting to individual plans invested through the financial markets, returns would be much higher and that thus our Social Security pensions would be “rescued”.  They assert that by privatizing Social Security investments, the system will be able to provide pensions that are either better than what we receive under the current system, or that similar pensions could be provided at lower contribution (Social Security tax) rates.

There are a number of problems with this.  They include that risks of poor financial returns (perhaps due, for example, to a financial collapse such as that suffered in 2008 in the last year of the Bush administration, when many Americans lost much or all of their retirement savings) would then be shifted on to individuals.  Individuals are not in a good position to take on such risks.  Individuals are also not financial professionals, nor normally in a good position to judge the competency of financial professionals who offer them services.  They also often underestimate the impact of high and compounding fees in depleting their savings over time.  For all these reasons, such an approach would serve as a bad substitute for the Social Security system such as we have now, which is designed to provide at least a minimum pension that people can rely on in their old age, with little risk.

But there is also a more fundamental problem with this approach.  It presumes that returns in the financial markets will in general be substantially higher than returns that one earns on what we pay into the Social Security system.  This blog post will show that this is simply not true.

The post looks at what the implicit rates of return are under several benchmark cases for individuals.  We pay into Social Security over our life time, and then draw down Social Security pensions in our old age.  The returns will vary for every individual, depending on their specific earnings profile (how much they earn in each year of their working career), their age, their marital situation, and other factors.  Hence there will be over 300 million different cases, one for each of the over 300 million Americans who are either paying into Social Security or are enjoying a Social Security pension now.  But by selecting a few benchmarks, and in particular extreme cases in the direction of where the returns will be relatively low, we can get a sense of the range of what the rates of return normally will be.

The chart at the top of this post shows several such cases.  The rest of this post will discuss each.

B.  Social Security Rates of Return Under Current Tax and Benefit Rates

The scenarios considered are all for an individual who is assumed to work from age 22 to age 65, who then retires at 66.  The individual is assumed to have reached age 65 in 2013 (the most recent year for which we have all the data required for the calculations), and hence reached age 62 in 2010 and was born in 1948.  The historical Social Security tax rates, the ceiling on wages subject to Social Security tax, the wage inflation factors used by Social Security to adjust for average wage growth, and the median earnings of workers by year, are all obtained from the comprehensive Annual Statistical Supplement to the Social Security Bulletin – 2014 (published April 2015).  Information on the parameters needed to calculate what the Social Security pension payments will be are also presented in detail in this Statistical Supplement, or in a more easy-to-use form for the specific case of someone reaching age 62 in 2010 in this publication of the Social Security Administration.  It is issued annually.

The Social Security pension for an individual is calculated by first taking the average annual earnings (as adjusted for average wage growth) over the 35 years of highest such earnings in a person’s working career.  For someone who always earned the median wage who reached age 62 in 2010, this would work out to $2,290 per month. The monthly pension (at full retirement age) would then be equal to 90% of the first $761, 32% of the earnings above this up to $4,586 per month, and then (if any is left, which would not be the case in this example of median earnings) 15% of the amount above $4,586.  Note the progressivity in these rates of 90% for the initial earnings, then 32%, and finally 15% for the highest earnings.  The monthly Social Security pension will then be the sum of these three components.  Since it is then adjusted for future inflation (as measured by the CPI), we do not need to make any further adjustments to determine the future pension payments in real terms.  The pensions will then be paid out from age 66 until the end of their life, which we take to be age 84, the current average life expectancy for someone who has reached the age of 65.

The historical series of payments made into the Social Security system through Social Security taxes (for Social Security Old-Age pensions only, and so excluding the taxes for Disability insurance and for Medicare) are then calculated by multiplying earnings by the tax rate (currently 10.6%, including the shares paid by both worker and employer).  The stream of payments are then put in terms of 2010 dollars using the historical CPI series from the Bureau of Labor Statistics.

We can thus calculate the real rates of return on Social Security pensions under various scenarios.  The first set of figures (lines A-1) in the chart above are for a worker whose earnings are equal to what median wages were throughout his or her working life.  (A table with the specific numbers on the rates of return is provided at the bottom of this post, for those who prefer a numerical presentation.)  The individual paid into the Social Security pension system when working, and will now draw a Social Security pension while in retirement.  One can calculate the real rate of return on this stream of payments in and then payments out, and in such a scenario for a single worker earning median wages throughout his or her career who retired at age 66 in 2014, the real rate of return works out to be 2.9%.  If the person is married, with a spouse receiving the standard spousal benefit, the real rate of return is 4.1%.

Such rates of return are pretty good, especially on what should be seen as a safe asset (provided the politicians do not kill the system).  Indeed, as discussed in an earlier post on this blog, the real rate of return (before taxes) on an investment in the S&P500 stock market index over the 50 year period 1962 to 2012, would have been just 2.9% per annum assuming fees on 401(k) type retirement accounts of 2.5% (which is typical once one aggregates the fees at all the various levels – see the discussion in section E.3 of this blog post).  But investing in the stock market, even in a broad based index such as the S&P500, is risky due to the volatility.  Retirement accounts in 401(k)’s are generally a mix of equity investments, fixed income securities (bonds of various maturities, CDs, and similar instruments), and cash.  Based on the recent average mix seen in 401(k)’s, and for the same 50 year period of 1962 to 2012, the average real rate of return achieved after the fees typically charged on such accounts would only have been 1.2%.  Social Security for a worker earning median wages is far better.

As noted above, there is a degree of progressivity in the system, as higher income earners will receive only a smaller boost in their pension (at the 15% rate) from the higher end of their earnings.  Thus the rates of return in Social Security for high income earners will be less.  The rates of return they will earn are shown on lines A-2 of the chart.  This extreme case is calculated for a worker who is assumed to have earned throughout his or her entire work life an amount equal to the maximum ceiling on wages subject to Social Security tax (which was $113,700 in 2013).  Note also that anyone earning even more than this will have the same rates of return, as they will not be paying any more into the Social Security system (it is capped at the wage ceiling subject to tax) and hence also not withdrawing any more (or less) in pension.

Such high income earners will nonetheless still see a positive real rate of return on their Social Security contributions, of 1.4% for a single earner and 2.8% if married receiving a spousal benefit.  That is, while there is some progressivity in the Social Security system, it is not such that the returns turn negative.  And the returns achieved are still better than what typical 401(k) retirement accounts earn.

One should also take into account that high income earners are living longer than low income earners.  Indeed, the increase in life expectancies have been substantial in the last 30 years for high income earners, but only modest for those in the bottom half of the earnings distribution.  While I do not have data on what the life expectancies are for a person whose earnings have been at the absolute top of the Social Security wage ceiling over the course of their careers, for the purposes here it was assumed their life expectancy (for someone who has reached age 65) would be increased to age 90 from the age of 84 for the overall population.

In such a scenario, the real rates of return for someone who paid into the Social Security system always at the wage ceiling over their entire life time and then drew a Social Security pension up to age 90 would be 2.2% if single and 3.4% if married with a standard spousal benefit.  These are far better than typical 401(k) returns, and indeed are quite good in comparison to an investment in any safe asset (once one takes into account fees).

C.  Social Security Rates of Return Assuming Higher Social Security Tax Rates

The rates of return calculated so far have been based on what the actual historical Social Security tax rates have been, and what the current benefit formula would determine for future pensions.  But as most know, at current tax and benefit rates the Social Security Trust Fund is projected to be depleted by about 2034 according to current estimates.  The reason is that life expectancies are now longer (which is a good thing), but inadequate adjustments have been made in Social Security tax rates to allow for pay-outs which will now need to cover longer lifetimes.  The problem has been gridlock in Washington, where an important faction of politicians opposed to Social Security are able to block any decision on how to pay for longer life expectancies.

There are a number of ways to ensure Social Security could be adequately funded.  One option, which I would recommend, would be simply to lift the ceiling on wages subject to Social Security tax (which was $113,700 in 2013, $118,500 in 2015, and will remain at $118,500 in 2016).  As discussed in section E.2 of this earlier blog post, it turns out that this alone should suffice to ensure the Social Security Trust Fund remains adequate for the foreseeable future.  The extra funding needed is an estimated 19.4% over what is collected now (based on calculations from an earlier post on this blog, but with data now a few years old), and it turns out that ending the wage ceiling would provide this.  At the ceiling on wages subject to Social Security tax of $113,700 in 2013, the share of workers earning at this ceiling or more was just 6.1%, but due to the skewed distribution of income in favor of the rich, untaxed wages in excess of the ceiling accounted for 17.3% of all wages paid.  That is, Social Security taxes were being paid on only 82.7% of all wages.  If the taxes were instead paid on the full 100%, Social Security would be collecting 21% more (= 100.0 / 82.7).

The extremely rich would then pay Social Security taxes at the same rate as most of the population, instead of something lower.  It should also be noted that it is the increase in life expectancy of those at the upper end of the income distribution which is driving the Social Security system into deficit at the current tax rates, as they are the ones living longer while those in the lower part of the income distribution are not.  Thus it is fair that those who will be drawing a Social Security pension for a longer period should be those who should be called on to pay more into the system.

To be highly conservative, however, for the rate of return calculations being discussed here I have assumed that the general Social Security tax rate will be increased by 19.4% on all wages below the ceiling, while the ceiling remains where it has been.  These calculations are for historical scenarios, where the purpose is to determine what the rates of return on payments into Social Security would have been had the tax rates been 19.4% higher on all, to provide for a fully funded system.  Finally, note that while these scenarios assume a higher Social Security tax rate historically, they also set the future pension benefits to be paid out to be the same as what they would be under the current benefit rates.  That is, the pay-out formulae would need to be changed to leave benefits the same despite the higher taxes being paid into the system.

The real rates of return would then be as shown in Panel B of the chart above.  While somewhat less than before, the real returns are still substantial, and still normally better than what is earned in a typical 401(k) plan.  The returns for someone earning at the median wage throughout their career will now be 2.4% if single and 3.6% if married (0.5% points less than before).  The returns for someone earning at or above the ceiling for wages subject to Social Security taxes would now be earning at the real rate of 0.8% if single and 2.2% if married for the age 84 life expectancy (0.6% points less than before), or 1.6% and 2.9% (for single and married) if the life expectancy of such high earners is in fact age 90 (also 0.6% points less, before round-off).

The real rates of return all remain positive, and generally good compared to what 401(k)’s typically earn.

D.  Conclusion

As noted above, the actual profile of Social Security taxes paid and pension received will vary by individual.  No two cases will be exactly alike.  But the calculations here indicate that for someone with median earnings, and still even in the extreme case of someone with very high earnings (where a degree of progressivity in the system will reduce the returns), the rates of return earned on what is paid into and then taken out of the Social Security system are actually quite good.  They generally are better than what is earned in a typical 401(k) account (after fees), and indeed often better than what would earn in a pure equity investment of the S&P500 index (and without the risk and volatility of such an investment).

Social Security is important and has become increasingly important.  Due to the end of many traditional defined benefit pension plans, with a forced switch to 401(k) plans or indeed often to nothing at all from the employer, Social Security now accounts (for those aged 65 or older) for a disturbingly high share on the incomes of many of the aged. Specifically, Social Security now accounts for half or more of total income for two-thirds of all those age 65 or older, and accounts for 100% of their income for one-quarter of them. And for the bottom 40% of this population, Social Security accounted for 90% or more of their total income for three-quarters of them, and 100% of their income for over half of them.

The problem is not in the Social Security system itself.  It is highly efficient, with an expense ratio in 2014 of just 0.4% of benefits paid.  Private 401(k) plans, with typical expenses of 2.5% of assets (not benefits) each year will have expenses over their life time that are 90 times as great as what Social Security costs to run.  And as seen in this post, the return on individual Social Security accounts are quite good.

The problem that Social Security faces is rather that with longer life expectancies (most importantly for those of higher income), the Social Security taxes being paid are no longer sufficient to cover the payouts to cover these longer lifetimes.  They need to be adjusted. There are several options, and my recommendation would be to start by ending the ceiling on wages subject to Social Security taxes.  This would suffice to solve the problem.  But one could go further.  As discussed in an earlier blog post (see Section E.2), not only should all wages be taxed equally, but one should extend this to taxing all forms of income equally (i.e. income from wealth as well as income from wages).  If one did this, one could then either cut the Social Security tax rate sharply, or raise the Social Security benefits that could be paid, or (and most likely) some combination of each.

But something needs to be done, or longer life spans will lead the Social Security Trust Fund to run out by around 2034.  The earlier this is resolved the better, both to ensure less of a shock when the change is finally made (as it could then be phased in over time) and for equity reasons (as it is those paying in now who are not adequately funding the system for what they will eventually drawdown).

 

============================================================

Annex:  Summary Table

Real Rates of Return from Social Security Old-Age Taxes and Benefits

A)  Social Security Scenarios – Current Rates

  1)  Earnings at Median Throughout Career

   a)  Single

2.9%

   b)  Married

4.1%

  2)  Earnings at Ceiling Throughout Career

   a)  Single

1.4%

   b)  Married

2.8%

  3)  Earnings at Ceiling, and Life Expectancy of 90

   a)  Single

2.2%

   b)  Married

3.4%

B)  Social Security with 19.4% higher tax rate

  1)  Earnings at Median Throughout Career

   a)  Single

2.4%

   b)  Married

3.6%

  2)  Earnings at Ceiling Throughout Career

   a)  Single

0.8%

   b)  Married

2.2%

  3)  Earnings at Ceiling, and Life Expectancy of 90

   a)  Single

1.6%

   b)  Married

2.9%

C)  Comparison to 401(k) Vehicles

  1)  S&P500 after typical fees

2.9%

  2)  Average 401(k) mix after typical fees

1.2%

More on the High Cost of the Purple Line: A Comparison to BRT on the Silver Spring to Bethesda Segment

Comparison of Purple Line to BRT Cost, Silver Spring to Bethesda

This is a quick post drawing on a report in today’s Washington Post on the implementation of bus rapid transit (BRT) in Montgomery County, Maryland.  The article notes that one of the early BRT routes planned in the county would run from Burtonsville to Silver Spring down US Highway 29, with an estimated capital cost of $200 million.

This would be a distance of 10.2 miles, so the cost would be $19.6 million per mile on average.  This BRT line is currently slated to stop in Silver Spring, but it would be straightforward to extend it along East-West Highway for a further 3.7 miles to Bethesda. Assuming the same average cost per mile, the capital cost of this addition would be $72 million.

The current plan is for the Purple Line light rail line to cover this same basic route, connecting Silver Spring to Bethesda.  As I have discussed in earlier blog posts, the Purple Line is incredibly expensive, even if one ignores (as the official cost estimates do) the environment costs of building and operating the line (including the value of parkland destroyed, which is implicitly being valued at zero, as well as the environmental costs from storm water run-off, habitat destruction, hazardous waste issues, higher greenhouse gas emissions, and more).  The current capital cost estimate, following the service and other cuts that Governor Hogan has imposed to bring down costs, is $2.25 billion.  This also does not include the costs that Montgomery County will cover directly for building the Bethesda station and well as the cost of a utilitarian path to be built adjacent to the train tracks.  The Purple Line would also cost more to operate per rider than the Montgomery County BRT routes are expected to cost, so there is no cost savings from lower operating costs.

The Purple Line would be 16.2 miles long in total.  Using just the $2.25 billion cost figure, this comes to $139 million per mile.  This is extremely high.  Indeed, the Columbia Pike streetcar line in Arlington County, which was recently cancelled due to its high cost, would have cost “only” $117 million per mile despite it being built through a high density urban corridor for most of its entire route.

The distance from Silver Spring to Bethesda on the Purple Line will be 4.4 miles if it is built. This is longer than the direct route by road since it will follow a more indirect path passing up and around the direct route.  Assuming the cost of this 4.4 miles is the same on average as for the rest of the Purple Line (it might be higher due to the need to build some major bridges, including over Rock Creek), the cost would come to $612 million.

The choice therefore is between spending $612 million to build this segment of the Purple Line from Silver Spring to Bethesda, or spending $72 million by extending the BRT.  The Purple Line cost is 8.5 times as much, and government could save $540 million ( = $612m – $72m) by terminating the Purple Line in Silver Spring and using BRT service instead.

As an earlier blog post argued, new thinking is necessary if we are to resolve the very real transportation issues we face in this region.  This is one more example of what could be done.  A half billion dollar savings is not small.

 

The Purple Line: New Thinking is Needed to Address Our Transportation Problems

The Washington Post in recent weeks has published a number of pieces, both articles and editorials, on purportedly massive economic benefits from building the Purple Line (a 16 mile light rail line in the Maryland suburbs of Washington, DC).  See, among others, the items from the Post herehere, here, and here.  Their conclusion is based on uncritical acceptance of the results of a consultant’s report which concluded that the funds invested would generate a return (in terms of higher incomes in the region) of more than 100% a year.  But transit projects such as this do not generate such huge returns.  This should have been a clear red flag to any reporter that something was amiss.

As discussed in an earlier post on this blog, the consultant’s report was terribly flawed. There were obvious blunders (such as triple counting the construction expenditures) as well as more complex ones.  While one should not expect a reporter necessarily to have the expertise to uncover such problems, one would expect a good reporter and news organization to have sought out the assessment of a neutral observer with the necessary expertise.  This was not done.

The consultant’s report was so badly done that one cannot conclude anything from it as to what the economic impact would be of building such a rail line.  But more importantly, the report did not even ask the right question.  It looked at building the Purple Line versus doing nothing.  But no one is advocating that we should do nothing.  We have real transportation issues, and they need to be addressed.

The proper question, then, is what is the best use of the scarce funds we have available for public transit.  If an alternative approach can provide better service for more riders at similar or lower cost, then the impact of building the Purple Line versus proceeding with that alternative is negative.  As discussed below and as an illustration of what could be done (there are other alternatives as well), instead of the Purple Line one could provide free bus service on the entire county wide local bus systems.  These bus systems cover poor communities as well as the rich (the Purple Line will pass through some of the richest zip codes in the nation), and for the cost of the Purple Line the bus systems could be expanded to provide free bus service to as many as four times the ridership the Purple Line is projected to carry (with such projections, based on past experience, likely to prove optimistic).  The local bus systems already carry twice the projected ridership of the Purple Line.

With the aim of providing an alternative view to the discussion being carried, I submitted the attached to the Post for consideration as a local opinions column.  The Post decided, however, not to publish it, so I am making it available here.

 

—————————————————————————————————-

Solving Our Transportation Problems Will Require New Thinking, Not Old White Elephants

As Governor Hogan comes to a decision on the Purple Line, there is increasing pressure from proponents arguing the 16 mile light rail line will yield huge developmental benefits.  A recently released and highly publicized study by the consulting firm TEMS concluded the line would raise incomes by $2.2 billion a year, for a capital cost that TEMS took to be just $1.9 billion (the cost estimate is now higher, at close to $2.5 billion).  Thus the return, TEMS says, will be well in excess of 100% a year.

As the saying goes, when something sounds too good to be true, it usually is.  The TEMS study was badly flawed.  More importantly, it did not address the right question.  What we should be asking is how best to address our very real transportation needs, given the limited public resources available.

There are numerous problems with the TEMS study.  To start with more obvious blunders:  It estimated impacts during the construction period as if the entire Purple Line would be built in Montgomery County, would be built again in Prince George’s, and built again in Washington, DC (even though it will not even touch Washington).  That is, it triple counted the construction expenditures and therefore its income and jobs impacts.  It also assumed that all the inputs (other than the rail cars) would be sourced in Washington or these counties.  But steel rail cannot come from here:  Neither Washington nor its suburbs have any steel mills.

There were more fundamental problems as well.  Among them was the fallacy of cause and effect.  In their statistical analysis, TEMS found that higher income neighborhoods are associated with lower transportation costs.  From this they jumped to the conclusion that lower transportation costs led to those higher incomes.  That is not the case.  Rich people live in Georgetown and, being close to downtown, transportation costs there are relatively low.  But moving to Georgetown does not suddenly make you rich due to low transportation costs.  Rather, one can afford to buy a home in Georgetown if you are already rich.

Perhaps the most basic problem is that TEMS assumed it was either the Purple Line or nothing.  But no one is advocating doing nothing.  We face real transportation issues, and they need to be addressed.  Unfortunately, alternatives have not been seriously examined.  Part of the problem has been narrow-minded thinking that has failed to consider broader alternatives than solely a line on this fixed corridor.

As an example of what might be done, consider the locally run RideOn and TheBus systems in these counties.  These two systems already carry double the projected ridership of the Purple Line.

The annual operating cost of the Purple Line is expected to be $55 million a year.  This is in addition to the $2.5 billion capital cost.  Taking just half of that annual operating cost, net of fares expected to be collected due to the Purple Line, one could cover the full amount currently collected in fares on the entire county-wide RideOn and TheBus systems.  That is, one could provide free bus service on the entire systems for just half of the cost of operating the Purple Line.

Some might say that with zero fares, there would be the “problem” that many more riders will want to take these buses.  But that would be fantastic.  One would need to cover the additional costs, but this could be done.  Note first that filling empty seats on buses costs nothing, and there are a lot of empty seats now.  There is then the second half of the annual operating cost of the Purple Line, and finally the $2.5 billion capital cost.  Taken together, these funds could cover the full costs (including costs covered in county and state budgets) of doubling the scale of the RideOn and TheBus systems.

We can therefore have the Purple Line, serving riders on a narrow 16 mile corridor that runs through some of the most affluent areas of these counties, or for the same cost provide free bus service on systems that could carry four times as many riders.  Furthermore, the bus systems serve not just affluent areas but also the poorest communities of the counties.  For the poor, earning at or close to the minimum wage with perhaps two jobs to get by, daily bus fares of $6 or $8 or more are not insignificant.

We need to be open to broader options for how to address our transportation crisis.  The debate on the Purple Line has not done that.  And by treating the issue as the Purple Line or nothing, proponents are increasing the likelihood that the outcome will be nothing.

—————————————————————————————————

Note on Sources:

a)  The TEMS consultant report, March 2015, commissioned by Montgomery County, Prince George’s County, and the Greater Washington Board of Trade.

b)  Current cost estimates for building and operating the Purple Line, along with ridership projections, are from the most recently issued Federal Transit Administration Purple Line Profile Sheet, November 2014.

c)  Cost and ridership data for the local transit systems (RideOn and TheBus) are from the National Transit Database of the Federal Transit Administration.

The TEMS Study of the Economic Impact of the Purple Line: A Good Example of a Badly Flawed Report

A.  Introduction

A review of a recently released report, purportedly on the economic impact of the Purple Line, should be of interest not only to those with a direct interest in the Purple Line project itself, but also to those interested in how such work is now used as part of a political process to influence decisions on major public projects.  It is a badly flawed report. Nonetheless, its results were announced with great fanfare, and treated without question by news organizations such as the Washington Post.

The Purple Line is a proposed light rail line which would run in a 16 mile arc through suburban Washington, DC, from east of the city to its north.  It is a controversial project, due to its high financial as well as environmental costs while serving relatively few riders.

An earlier analysis on this blog calculated that the full cost per trip on the proposed system would be an estimated $10.42 (and double this per day for a round-trip).  But the system would take in only 38 cents in average additional fares per trip (since a large share of the riders will be free or reduced rate transfers from the bus or existing rail systems), leading to a subsidy of over $10 per ride.  And this analysis assumes that the current cost and ridership projections will hold true.  Such projections have generally proven to be highly optimistic on other light rail projects.

Despite the high cost, there are significant vested interests pushing strongly for the project.  In particular, land developers along the proposed corridor would see the value of their properties rise, possibly by the hundreds of millions of dollars.  And local government authorities (in particular those of Montgomery and Prince George’s Counties) have come out in favor:  Almost all of the cost would be borne by the State of Maryland or the Federal Government, and the subsidy payments from the State of Maryland would be locked in (under the proposed PPP contract) for 30 years beyond the estimated 5 year construction period.

In this environment of controversy, a consultant’s report was commissioned and recently released with great fanfare.  The report (dated March 2015) is titled “Purple Line Preliminary Impact Study:  Update”, and was prepared by the firm Transportation Economics & Management Systems, Inc. (TEMS).  The conclusions from the report were provided publicly in a presentation to business leaders on April 20, reported on by the Washington Post that day, and used also that same day as the basis for an editorial by the Post advocating that the Purple Line should be built (a position the Post has long taken).

There are numerous and major flaws with the report.  This blog post will go through some of the more important ones.  But first it will summarize several of the red flags that should have signaled to any serious analyst and news organization that there could be problems with the report, and that a more careful reading would have been warranted before its conclusions were widely publicized.

A number of the problems with the report are quite technical.  I would not suggest that a general news reporter would have the technical knowledge necessary to have discovered these himself or herself.  But the red flags are obvious, and should have signaled to the journalist that there could very well be issues here, and that if he did not have the skills to assess the report, then he should have consulted with some neutral third party to do such an assessment for him.  The Purple Line project is controversial, and an experienced reporter and news organization should have recognized that a report such as this, commissioned by and released by advocates for the project, may not be one to take on faith.  But this was not done.

B.  Red Flags

Some obvious issues should have raised attention:

1)  Gigantic Returns:  The TEMS report concludes that building the Purple Line will lead to an “Increase in income to local households of $2.2 billion per year” (Chapter 8, Conclusion).  This is astounding.  The cost estimate to build the line used in the study (from the August 2013 Final Environmental Impact Study) was only $1.9 billion (expressed in 2014 prices).  (Note:  The most recently published estimate, from November 2014, puts the expected cost a good deal higher, at $2.45 billion in current dollars, or about $2.3 billion in terms of 2014 prices.  But the TEMS study used the earlier cost estimate.)

A $2.2 billion increase in annual incomes on a one-time $1.9 billion cost implies an annual rate of return of 116%!  One is generally content with annual rates of return of perhaps 16%, or even 10%, on public projects.  Yet this one claims a return of 116%.  This should have been an immediate flag that something is questionable in what was done.  As the adage goes, if something is too good to be true, it probably is.

2)  Implausibly Precise Statistical Results:  While a more technical issue, any observer conversant with basic statistics and regression analysis would have been surprised to see that the t-statistic was as high as 250 in the cross-section regressions in the simple model of travel demand (Exhibit 4.2 of Chapter 4).  The t-statistic is a measure of how tight the data fits around the estimated coefficients of a regression equation.  Any t-statistic greater than 2.0 is generally taken to imply the coefficient is statistically different from zero (with a 95% confidence).  In cross-section regressions, one is normally happy to find t-statistics of 2 or 3.  But in the statistical regression reported on here, for the estimated number of trips by commuters between two geographic zones as a function of just two simple variables, the t-statistics varied between 200 and 250.

Such precision in results in statistical work such as this is highly surprising.  In the real world there are many other determinants of travel demand between two zones than just the two variables used in the TEMS study (one for the cost of such travel, and one a constructed variable based on population, incomes, and employment).  While I do not have access to the data they used to determine what is going on, any statistician would be highly suspicious of such precise results.

3)  Who Sponsored and Paid for the Report:  An assessment of any report such as this starts with finding out who commissioned and paid for it.  The Washington Post article and editorial both state that the report was commissioned by Montgomery and Prince George’s Counties, the two Maryland counties through which the Purple Line will run.  No one else was mentioned.  Yet a report on the same presentation that day by the Gazette (a local newspaper of Suburban Maryland) noted that the Greater Washington Board of Trade was also a commissioner of the work, along with the two counties.  The Board of Trade is an industry group, whose members include construction companies and property developers, a number of whom will benefit directly if Maryland proceeds with this project.

Good journalism would have called for full disclosure on who sponsored and paid for the report.  If this was misunderstood at the time, a correction should have been reported later.  And an obvious question at the presentation of the report would have addressed not only who commissioned the report, but also what was the total cost and how was that cost shared among the sponsors.  Given the tight budgetary situation of all governments these days, it would not be surprising if a disproportionate share of the costs came from the Board of Trade (and some sub-set of its members who might have an interest in the outcome).

C.  Problems With the Report

Once one delves into the details of the work done, a number of flaws become clear.  This section will summarize a few of them.  The sequence followed is that of the report, starting with the theoretical construct, through the statistical work, and then the results.  However, this sequence unfortunately means that the more important issues are the ones further down on the list, rather than at the top.  I hope the reader will be patient.

1)  Confusion in the Theoretical Framework:  There are two major parts to the report. The first seeks to estimate what it terms to be the long-term supply side impacts of the project, while the goal of the second is to estimate the immediate impacts on the region from the construction spending itself.  We will focus first on the report’s supply side analysis, starting with the theoretical framework presented.  A separate section below will review how the immediate impacts were estimated.

The report provides an elaborate theoretical framework (in Chapter 2) for the approach they say they are taking, but there are issues.  It starts by saying they will work through a supply side analysis to determine how a transportation investment such as the Purple Line will increase productivity and output, and assert that this will be equivalent to (the “mirror image” of) the more traditional approach of valuing transportation investments by how much cost and time they save for drivers and riders.  But in fact this will not be the case. Measured levels of household incomes simply do not include as one element the time saved (or as a negative, the time consumed) in travel.  Yet the TEMS report uses the standard measures of household incomes and other such economic variables in their statistical work.  Thus the TEMS approach and the traditional approach of valuing the benefits from transportation investments will not be mirror images of each other.  They will produce totally different results.

They also define what they call “economic rent”, to be a function of the variables: population structures, industrial structures, education levels, cultural characteristics, and “transportation efficiency”.  They do not further define the five variables other than transportation efficiency, but argue that they will be largely unchanged over a period of 10 to 20 years or so.  Thus any changes over such a period will only be due to changes in transportation efficiency.  Actually, this will not be the case, as any geographic area will see its population and incomes changing over time.  But while incorrect, it should not matter to their analysis.  One could interpret their approach as looking for the partial effect of transportation efficiency on what they call economic rents.

But there are problems with how this is implemented.  First, they take as their measure of “transport efficiency” a weighted average cost of automobile travel (for both time and financial costs) from a specific geographic zone to all other geographic zones in the region. Why they should include only automobile travel in a study looking at the impacts of a light rail line is not clear.  But more of a concern is that “economic rent” is measured by a series of what they call “proxies” (specifically:  employment, household income density, and residential property value density), and that they assume that each of these variables is separately a function of transportation costs (and transportation costs alone).

This is a simplistic framework.  It is not at all clear why the specific variables they define as “proxies” for economic rent do indeed capture what economic rent really is. They merely assert they do.  Economic rent corresponds to the value of land in a particular location. Land rent, with all else equal, will generally be higher in more central locations with lower transportation costs.  But land rent is not synonymous with employment or with household incomes, for example.  Thus while there may well be a relationship between land rents and transportation costs, it is not at all clear why there should be the same such relationship between household incomes and transportation costs.

There are therefore issues with the theoretical framework used.

2)  Flawed Statistical Analysis:  I noted above that at least certain of their statistical results appear to be too good to be true.  But there are other issues as well.

One mistake is to assume that a relationship that might apply at a broad geographic scale will apply in the same way in a more limited jurisdiction.  Their basic statistical work is based on an analysis of the relationship between their socio-economic proxies and average transportation costs over a set of 299 geographic zones in the Washington and Baltimore metropolitan areas.  This is a large area, stretching from the Pennsylvania border to south of Fredericksburg, Virginia (a distance of over 150 miles), and from the Shenandoah Valley in Virginia to the Chesapeake Bay.

Distances such as this matter a good deal in deciding where to live and commute.  There will not be many people commuting from Fredericksburg to Baltimore and beyond, or from Warrenton to Annapolis.  Even if one found a nice house and neighborhood in such areas, the cost of commuting will dominate in the decision not to live there.  And a statistical regression, when properly done, should pick up such relationships and show that the commuting costs of course matter.

But there is then a problem is assuming that the same statistical relationship will apply similarly, and with the same parameters, when examining housing and commuting choices on a much smaller scale.  If your commute would be five miles, say, from one possible home location, and seven miles from another, the difference in commuting times might not be all that important.  Rather, one might choose the location that is further away based on how much one likes the specific house or neighborhood, where your friends live, and other such factors.  It would be a mistake to assume the statistical relationship with transportation costs will be the same.

Yet the authors of this report do assume this.  They assume that the relationship they estimate based on the region wide data stretching over 150 miles and many hours of potential commuting time will apply similarly at the scale relevant to riders deciding whether or not to take the Purple Line.  The Purple Line will only be relevant to largely local riders, living and/or working within a few miles of the 16 mile long rail line.  Statistically, the authors made the mistake of presuming that relationships in a data set that is largely “out-of-sample” will apply similarly in the more limited scale relevant to the Purple Line.

There are other issues as well.  As already noted, the t-statistics for their travel demand model estimations are implausibly high.  It is also odd that the estimated slope coefficients in their regressions relating employment, household income density, and property value density (in Exhibit 5.4), and later housing density and housing units density (in Exhibit 5.7), as a function of average transportation costs, are all in the relatively narrow range of -3.30 to -3.97.  By the way the equations were structured, these coefficients are all what economists call “elasticities”, meaning that a 1% decrease in average transportation costs in the zone will lead to increases of between 3.30% and 3.97% in the various socio-economic variables.  It is surprising that these response rates are all so close to each other, for such very different variables as employment, household income densities, property values, and so on.  While I cannot say what might be causing this without knowing more on precisely what was done, the similarity in response rates over such disparate variables is probably a flag that something was not done properly in the statistics.

There is also a, possibly related, technical statistical issue in that they assume in one set of relationships that their socio-economic measures (income, etc.) are a function just of their average transportation cost figures (equation 12), while in another equation (equation 6) they postulate that travel demand will be a function of certain constructed socio-economic variables (which are themselves built up from the basic set of socio-economic variables) and average transportation costs. This implies in their system that the variables they are using (the socio-economic variables and average transportation costs) to explain travel demand are not in fact independent of each other.  When this is the case, ordinary least squares regressions will not work, and one needs to utilize a more sophisticated statistical approach.

3)  The Elasticity Estimates Are Just Not Plausible:  While the similarity across the elasticity estimates is curious, it is more important to recognize the implications of the values themselves.

Using the case of the response of household income density to transportation costs, the equation the TEMS study estimated found an elasticity of -3.79.  That is, for a 1% fall in transportation costs in the area, household income density will rise by 3.79%.  Some of this might come from higher average household incomes in the area and some by more homes being built in the area, both of which will increase the income of the area.

This would be a huge response, if true.  Transportation costs (private plus intracity public transit) on average accounts for about 15% of the consumer price index (BLS data on the CPI weights).  Median household income along the Purple Line is roughly $80,000 (based on a simple average of the median household incomes at the four major stations where there are now regular MetroRail lines).  15% of $80,000 is $12,000 spent directly on transportation costs.  To this one should add the value of time spent commuting (as an additional cost).  Based ultimately on Census Bureau data, a study found that residents of Washington, DC, spend an additional 11% of their working hours each week on commuting.  Applying this 11% to the $80,000 median household income, the total cost of transportation for an average household is 26% of $80,000, or $20,800.

The TEMS regression results, if they are to be believed, imply that a 1% reduction in transportation costs ($208 = 1% x $20,800) will lead to a 3.79% rise in household incomes ($3,032 = 3.79% x $80,000) through either a rise in per household incomes or in the number of households in the zone or by some combination.  This implies that a subsidy of just $208 per household for what they spend on transportation will lead to a rise in household incomes in the area by $3,032!

This would be amazing, if true.  A small $208 cost would be converted into more than a $3,000 gain in annual incomes!  And with government income tax rates averaging roughly 25% (the figure the TEMS study uses), the government tax take would rise by over $750. Only 28% of this increase in the tax take could then be used to pay for a further $208 subsidy, and one would have the equivalent of a perpetual motion machine (or in this case a perpetual wealth machine).

Unfortunately, it is not likely that there will be such a response to transportation investments.  Perpetual wealth machines do not exist.  The parameter estimates are simply implausible.  The reason why the result may have been found (assuming the statistics was done properly, which is itself not clear) will be discussed immediately below. The implausible parameter values also explains why the TEMS study found such purported high returns (of 116% a year) for an investment as costly and as inefficient as the Purple Line.  But as the next section will discuss, the interpretation was wrong.

4)  Lower Transportation Cost Is Not the Main Cause of Higher Incomes – Correlation Is Not Causation:  The regression equations summarized in Exhibits 5.4 and 5.7, regress variables such as employment, household income density, and so on, on average transportation costs in the zone.  But it is a well known principle in regression analysis that such regressions do not demonstrate causation.  Rather, they can only show correlation.

Nevertheless, the TEMS report asserts that the correlations found in their regressions do show that employment, household income density, and so on, will rise as a direct result of average transportation costs falling.  The percentage rise will be in accordance with the elasticities estimated, they assert, and will be a consequence of the higher productivity of the economy that lower transportation costs leads to.

But it is not at all clear that the causation goes in the direction the TEMS report asserts. The correlations may rather be showing that people with high incomes prefer to live in areas where transportation costs (and commuting times, which are part of transportation costs) are relatively low.  In the Washington, DC, area, to take an example, the Georgetown neighborhood is a high income area in the city, close to the central downtown office zone, and hence an area with relatively low transportation costs.  Many rich people who can afford it like to live in the area, and home prices are high reflecting this preference.  But the residents of Georgetown did not become rich because transportation costs are on average relatively low there.  Rather, rich people have sought to live in Georgetown for, among other reasons, the relatively low cost of getting to work from there.

Thus one finds in the regression results a correlation between high incomes (and the other variables estimated) and relatively low average transportation costs.  But the residents did not become rich as a result of some reduction in transportation costs. They were already rich, which allowed them to move into an area such as Georgetown.

Thus it is incorrect to conclude, as the TEMS study does (see the beginning of Chapter 6, page 36), that building the Purple Line will “create more than 27 thousand jobs; will increase property value (sic) by 12.8 $ billion (sic) and the household income (sic again) is estimated to increase by $2.2 billion”.  Building a rail line (or any other transportation improvement) will not itself raise household incomes in such a way or create thousands of jobs.  Rather, the correlation observed (and assuming the statistical analysis was done correctly) can arise due to the choices people make between living in one neighborhood and another.

Note also that a decision of a relatively high income households to move to a location such as Georgetown in preference to a location further away from their job, will lead not only to higher income households concentrating in Georgetown, but to a symmetrical reduction in such households in the other locations they chose not to move to.  Similarly for property values:  Home prices will be bid up in Georgetown, and will see a reduction relative to what they would otherwise be in other locations.  But this is arising not because lower transportation costs is making people richer in Georgetown (that is, not due to a supply side effect increasing productivity, as the TEMS study asserts), but due to shifts in location preferences.

This is important.  A reduction in transportation costs is not making the region richer through some supply side effect, and certainly not in accordance with regression coefficients such as those found (with an elasticity of -3.79 for income, for example). Rather, the regression equations (and assuming again that the statistics were done properly, even though there are questions on that) are picking up at best a locational preference that shifts households from one location to another, and has limited or no effect on household incomes or property values in the region as a whole.

5)  The Multiplier Analysis Fails on Several Counts:  In addition to the “supply side” analysis reviewed above, the TEMS study undertook to estimate the immediate impact on employment and incomes in the areas immediately surrounding the Purple Line corridor during the construction period.  It was this analysis that led to the stated figure in the news reports that the project would create 4,000 jobs per year during the construction period (see here and here for example).

The multiplier analysis is decidedly not supply side analysis, but rather a purely demand side assessment of how much incomes and jobs would rise to produce what goes into the project.  And in a multiplier analysis, one takes into account not only what is used directly in the project, but also the production of the inputs that go into what is used directly and then the inputs into the inputs, and so on.

When unemployment is high and factories are underutilized, a multiplier analysis can be of interest.  An earlier post on this blog discussed what the fiscal multiplier means at the national level, and how the value of the multiplier will differ across countries and under different conditions, in particular whether one is assessing the multiplier at a time of high unemployment or low.  It can certainly be a useful tool if properly applied.  But one needs to be careful in how it is applied, and here the TEMS study fails.

There are multiple issues:

a)  The TEMS study failed to recognize that the major share of the inputs to the project will come from outside the region:  The expenditures that are the basis for the multiplier analysis come from the FEIS, which was finalized in August 2013.  The FEIS study has the capital cost figures in 2012$, and the TEMS authors puts them into 2014$. The capital cost estimate in the FEIS would then be $1.9 billion in 2014$ (it is now projected to be higher).  From this, the TEMS authors subtracted the cost of the train vehicles of $0.2 billion, as these vehicles would be built somewhere outside the Washington, DC, metropolitan region (the initial set of streetcar / light rail line vehicles purchased for a new line in Washington, DC, indeed came from the Czech Republic). Thus building such cars would have no multiplier effects here.  This was correct.

But then the TEMS study assumed that the entire remaining $1.7 billion would be used to purchase items for the Purple Line from production in Montgomery County, Prince George’s Country, or Washington, DC.  This is of course not true.  There are no steel mills in Washington, DC, or its Maryland suburbs that produce steel rails.  There are no plants that produce the sophisticated electronics that goes into the communications and other systems of the control centers (Siemens of Germany is one of the main global suppliers of such systems).  The overhead power lines are not made from copper and other materials mined locally.  And so on.  The primary and perhaps sole local component would be the share of the $1.7 billion paid to local labor for the installation.  This will be a significant cost item, of course, but far less than the full $1.7 billion.

It is thus a gross error to have assumed that the purchase of the steel rails, the communications equipment, the overhead power lines, and much of the rest, will lead to local multiplier impacts in the Washington region from their production.  Their production is elsewhere.  Thus the true multiplier impacts in the Washington region, even if one accepts their methodology, will be nowhere close to those they estimate.

But it gets worse.

b)  The Construction Cost Estimates Were Triple-Counted, Once Each for Montgomery County, Prince George’s County, and for Washington, DC:  The TEMS study concluded that there would be an additional $7.0 billion in gross regional product as a consequence of the $1.7 billion in construction expenditure for the Purple Line.  This implies a multiplier of 4.1 (= $7.0 billion / $1.7 billion).  Such a multiplier would be huge.  At the national level, one might expect a multiplier of 2 to 3 when unemployment is high, and many economists have argued that it might be more like 1.5.  It really depends on the degree of unemployment and other conditions.  But no one says it will be more than 4.

Furthermore, the multiplier at the national level will be much higher than the multiplier at a regional level.  If my income goes up due to employment on some project, I will spend that income not only on goods and services produced in the immediate Washington, DC, region, but also on pork from Iowa, wines from California, vegetables from Florida, cars from Michigan (or Germany), and so on.  Hence the local multiplier will be far below what it will be at the national level, and will be smaller the smaller one defines the local region (less for the city of Washington, DC, than for the Washington, DC, metropolitan region, for example).

So how did the TEMS authors arrive at such a high multiplier of 4.1?  They made a big blunder.  Examination of the tables showing their calculated Gross Regional Product figures for Montgomery County, Prince George’s County, and Washington, DC (Exhibits 7.10, 7.11, and 7.12) shows increased construction sector product of $1.66 billion in each case.  But this is (to three significant digits) the estimated total construction expenditure assumed for the Purple Line (the $1.7 billion figure is rounded from a more precise figure of $1.656 billion that one can obtain by reproducing the process they followed to arrive at their $1.7 billion).  The individual figures for Montgomery County, Prince George’s County, and Washington, DC, differ very slightly (in the fourth digit) since the feedback effects in the input-output matrices used for the multiplier analysis will differ a bit across these jurisdictions.

The TEMS authors triple counted the expenditures on the Purple Line.  Not only did they assume the entire $1.7 billion non-vehicle cost of the line would be spent locally, but they presented figures based on $1.7 billion being spent in Montgomery County, $1.7 billion being spent again in Prince George’s County, and $1.7 billion being spent again in Washington, DC (and the Purple Line will not even touch Washington).

The results for the multiplier analysis are therefore completely wrong, even if one takes their methodology for granted.  They made a big blunder.  But what is perhaps even more worrying is that the multiplier they reported of 4.1 was clearly far too high for what one would expect in any such analysis at a regional level.  Despite what should have been a big flag that something was amiss, the results were reported without the authors reviewing how they had arrived at such a large and implausible number.

c)  The Multiplier Methodology is Mechanical, and Implies That Cost Overruns are Good:  Finally, one should note that a multiplier methodology such as that used here, even if applied without the mistakes that were made, is a mechanical one.  One takes construction expenditures, at whatever level they are, and multiplies out the implied levels of employment, regional product, and personal incomes that follow based on this multiplier approach.

An implication of this is that every time the cost goes up, the calculated “benefits” rise also.  Indeed, under a multiplier analysis such as that done here, the benefits will rise in proportion.  If the project ends up costing twice as much, then the “benefits” in terms of higher jobs and incomes will be twice as much.  But this is of course silly.  Cost overruns are not good.

The problem is that the wrong question is being asked.  A project is not a good one because it requires more (rather than less) labor to build it.  Higher costs are not a good thing.  Rather, the objective of a transportation investment is to provide transportation services, and the question that should be asked is what is the lowest cost and most efficient way to provide those services.  If one can achieve the transportation aims with a project that only costs half as much, then one should follow that approach rather than the more expensive one.  And if one then has additional budget resources available through following the lower cost approach, one can then consider undertaking other projects, for transportation or whatever.  In the end, the number of jobs involved will be similar if similar amounts are spent.

6)  The most basic flaw in the TEMS study was that it was asking the wrong question:  The question the TEMS study sought to address was what the economic impacts would be of building this project compared to doing nothing.  But this was the wrong question.

No one is advocating that nothing should be done to address the very real transit issues in the area of the Purple Line corridor.  The issue, rather, is how best to address the transit needs.  Any assessment of the Purple Line should not be relative to doing nothing, but rather relative to what the best other alternative would be.  If the best other alternative is superior to the Purple Line, then the actual impact of building the Purple Line (instead of the alternative) is negative.

The Alternatives Analysis / Draft Environmental Impact Statement (AA/DEIS), did look at a number of bus alternatives.  All turned out to be far cheaper than light rail both in total amount and per rider (see Summary Table 6-2 of Chapter 6 of the AA/DEIS).  The most cost effective (in terms of cost per new rider) was a simple upgrade of the regular bus system, with a cost per new rider that was 60% less than the light rail alternative chosen. Furthermore, a bus system can be easily scaled up or down, with frequency and routes adjusted depending on ridership and changing development patterns.  A light rail system is fixed, and fixed forever.  It is also basically either all the way on or all the way off.  There is little flexibility.

It should also be noted that the true alternative should have recognized that not just buses provide transit to riders in this corridor.  One also has the existing MetroRail system. The four larger stations of the Purple Line would be at intersections with four MetroRail stations, and existing MetroRail service would often require less time for the journey than the Purple Line would.  Light rail lines are slow.  For example, the FEIS highlights (see Table 9-1 of Chapter 9 of the FEIS) that in the year 2040, a bus journey from Bethesda to New Carrollton (the two end points on the Purple Line) would require 108 minutes, while the Purple Line light rail would require 63 minutes, a saving of 42% they state.  But the FEIS failed to recognize that no rational person would take the Purple Line for such a journey, since one could make the same trip by MetroRail (today and in 2040) in just 51 minutes.  The Purple Line would take substantially longer for this journey than simply taking the existing MetroRail service.  Nevertheless, having failed to take into account the MetroRail alternative, the FEIS (and then the TEMS study as well) calculated benefits as if a transit rider would save 45 minutes ( =108 – 63) from Bethesda to New Carrollton by taking the Purple Line rather than the “no build” alternative of a bus following the same route.

The alternative considered in the FEIS to the light rail line was therefore a straw man.  They did not take into account the MetroRail alternative, which would be as fast or faster for many of the riders, nor did they consider seriously what an upgraded bus system could do.  And much could be done to upgrade bus service from the second class system it has been treated as, through use of a combination of redesigned routes, express routes on some corridors, perhaps bus rapid transit on some routes, and more.  But even the straw man they did consider was far more cost effective than the light rail alternative chosen.

D.  Conclusion

There are major flaws in the TEMS study, both in its structure and in its implementation. Some are outright blunders, such as the triple counting in the multiplier analysis by treating the Purple Line as if it were to be built completely in Montgomery County, completely again in Prince George’s County, and completely again in Washington, DC.  But even without such mistakes, the approach taken has major issues, such as from confusing correlation with causation, failure to recognize that the bulk of the inputs would come from elsewhere, the statistical issues, and more.

While a number of the issues are technical, there were also easy to spot clear red flags that something was wrong.  A public project such as this does not generate an annual rate of return of 116%.  One does not get fantastically precise statistical results with real world data.  These and other results should have served as flags first to the authors of the study that something was wrong, second as a warning to those commissioning the study that the results looked odd, and third as a signal to the journalists covering the release that they should consult with some neutral third party who would have the necessary expertise to advise on whether there might be issues.  When something looks too good to be true, it usually is.

But such a review was not done, and the results were announced as if they were valid.