Monday, March 31, 2008

The cost of "sitting on" Social Security

In a Seattle Times column by Froma Harrop, Dean Baker is quoted regarding his preferred plan for Social Security: wait and see how big a problem it turns out to be.

What should we do about Social Security? "I would just say, 'Let's sit on this,' " Baker answers. If come 2030 Americans see problems looming, he adds, "we can do something."
Here is the Social Security Trustees best guess of what "sitting on it" until 2030 would mean. The current 75-year Social Security deficit is 1.7% of payroll, meaning that and immediate and permanent increase of 1.7% in the payroll tax, from 12.4% to 14.1%, would be sufficient to restore solvency for 75 years. (An equivalent immediate and permanent benefit reduction would do the same.)

With each passing year, an additional year of deficits is added to the 75 year calculation, thereby increasing the size of the deficit. This tends to increase the shortfall by around 0.06% of payroll each year. Therefore, our best guess of the 75-year actuarial deficit in 2030 would be around 1.32% of payroll higher than today's, for a total of around 3% of payroll. This would require an immediate tax increase of 3%, to 15.4%, or equivalent reduction in benefits.

Moreover, while there is a great deal of uncertainty regarding Social Security's future finances, this uncertainty isn't a reason for delaying action. Remember that things could turn out worse than the Trustees project, not only better than projected, and people would be willing to act sooner or even over-balance the program in order to avoid this unlikely but adverse outcome.


Bruce Webb said...

"With each passing year, an additional year of deficits is added to the 75 year calculation, thereby increasing the size of the deficit. This tends to increase the shortfall by around 0.06% of payroll each year. Therefore, our best guess of the 75-year actuarial deficit in 2030 would be around 1.32% of payroll higher than today's, for a total of around 3% of payroll."

That may be your best guess but recent history has not been kind to that projection, something helpfully pointed out in 2005 by EPI in the following table: Changes in Trustees' projections over time. The .06 change due to the change is just a structural recognition that the economy will be larger 76 years out than it is today which increases the deficit in nominal terms. However this effect has been more than offset by other changes in 10 of the last reporting years and by about half in one of the other two. Only in 2006 was the overall negative change more than this built in bias downwards. To complete the data series: 2005 1.92%, 2006 2.02%, 2007 1.95%, 2008 1.7%, all of this despite a built in -.06 due to this change in valuation period..

Which really was Dean's point. Reformers have been telling us for years that 'We can't afford to wait, the problem will just get bigger and more expensive'. Except that it hasn't. In fact even in 2006 when we experienced a fairly sharp upward adjustment (mostly due to a change in assumed interest) of .10 percent the Cost of Inactivity was negative for everyone less than 20 years from retirement, the fix expressed in payroll terms of doing something the previous year was greater than the projected extra tax going forward. Doing Nothing left money in my pocket in excess of my increased liability. And then subsequent drops in payroll gap from 2.02% to 1.95% to 1.7% in the years since validated Nothing as a plan for Social Security at least in the short term. It doesn't really matter if you want to take your benchmark at the 2.23% of 1997 or the 2.02% of 2006, until or unless the payroll gap re-reaches those levels there is no reason to move, each year of non-activity is the equivalent of a tax cut equal to the projected payroll tax gap for the current year. That cumulatively is a lot of money since 1997. Which partially explains why Dean called his 1999 book Social Security: the Phony Crisis

'Nothing'. The numerically proven plan for Social Security since 1997.

Andrew G. Biggs said...

One thing I find interesting about the folks who deny the Social Security problem is their difficulty dealing with CBO's projections. On one hand, CBO projects a significantly smaller deficit than the Trustees do, which should make them attractive. But, while smaller, the deficit as projected by CBO is still quite large, which hurts things for the folks who claim there isn't a problem at all.
Put another way, there isn't just one set of projections showing significant problems for Social Security, there are two -- conducted by very different groups using very different models. Are BOTH CBO and the Trustees stupid, or part of a conspiracy? I wonder.

Arne said...

If you look at this chart,, in the CBO report, you will see that the variation in actual revenues over the past 10 years exceeds the 80 percent limits from the model. That is simply unreasonable.
Both the SSA and the CBO have unreasonable confidence in their prediction of what is "most likely". I still find their consistant systematic errors of the last decade (including from last year to this year) reason to believe that systematic errors will persist.

Bruce Webb said...

Thanks Arne.

What I find interesting is that advocates of Crisis proudly point to the Stochastic projections as somehow being dispositive between Low Cost and Intermediate Cost while ignoring that actual results at least to 2004 have been falling at or beyond the upper confidence interval. At some point this kind of result has to shake your confidence in the overall model, and after a period of time even in the modelers. I reached that point with the 2003 Report which for the first time introduced Infinite Horizon numbers into the Report and added the equal novelty of the Stochastic Projections. I am not in Arne's league as a data cruncher but this sentence sent up a red flag:
"Each time-series equation is designed such that, in the absence of random variation, the value of the variable would equal the value assumed under the intermediate set of assumptions."
Correct me if I am wrong, but it would seem that this test is simply assuming what was put into question, that these fully numbers were in fact fully informed neutral projections with no bias up or down, and just show that you don't get to either Low Cost or High Cost through random variation.

This sentence also raises some questions.
"Due to time constraints, results presented reflect the intermediate assumptions and the methods of the 2002 Trustees Report. Updating to the assumptions and methods of the 2003 Trustees Report is not expected to materially change the results."
What time constraints? This suggest to me a rush job. And raises another question. In footnote 1 of Prof. Samwick's LMS Plan he is described or describes himself as:
"Andrew Samwick is Professor of Economics and Director of the Nelson A. Rockefeller Center for Public Policy at Dartmouth College. From 2003 to 2004, he was Chief Economist on the staff of President Bush’s Council of Economic Advisers, where his responsibilities included Social Security."

Somebody made the decision to introduce these new innovations, one apparently at the last minute, into the 2003 Report, and I have a hard time believing that decision was made at staff level in the OACT. On the other hand those PV numbers are fairly critical in selling LMS in the face of actual 75 year numbers which seemed to be a perfectly acceptable measure right through the 2002 Report. Something is not quite right in this picture.

Bruce Webb said...

As to CBO numbers I personally don't use them because they simply open one up to 'gotcha' moments, you look like an idiot citing 2046 when any half-informed yo-yo can point out the official number of 2041 as represented by the official Trustees' Report.

Just as a tactical matter I find it easier to confine all my data points within the official models. As often say I am not a data cruncher, I am just a data pointer using the near universally accepted data set, i.e. the Reports.

Andrew G. Biggs said...

I'm not sure if this table will come through correctly, but it shows the CBO projected annual balances relative to GDP for the 10th/90th percentiles from 2004-2007, then includes the actual balance as a percent of GDP from these years (this taken from various Trustees Report figures and calculated by me). It shows the actual balance smack in the middle of the projected area.

10th 90th Actual
2004 44% 74% 66%
2005 49% 78% 59%
2006 47% 85% 64%
2007 42% 88% 57%

I haven't looked only at revenues, since what matters is revenues relative to costs (eg, annual balances or TF ratios). Nominal dollar values are surely harder to predict than balances relative to payroll/GDP, but I'll check it out when I can.

Re the stochastic simulations, they came about in response to work by CBO and recommendations from the 1999 Technical Panel (see, p. 75ff). The 2003 Technical Panel said "The Trustees are to
be commended for the inclusion of projections based on a stochastic model for the first time in the
2003 Report." (see, p. 14).

I don't believe Andy Samwick had any hand in the stochastic modeling. He was at CEA at the time, which like other White House offices doesn't take part in the Trustees process. The main staff person on that end would have been Kent Smetters from Treasury (now at the Wharton School), but these things don't come about until consensus has been reached among the Trustees and staff. The Trustees were actually late to the stochastic game, since CBO was already doing it and Ron Lee at U of Calif had also done a lot of work.

Bruce is right that the stochastic model is by design based on the intermediate assumptions from the Trustees, so the median outcome will be very close to the Trustees Intermediate projection. (The median outcome is usually a tiny bit worse due to interactions between the variables, but not much.) So if you think the average rate of real wage growth will be 1.x% rather than 1.1%, then you'd want to have a higher value inputted into the stochastic model. But no one has presented (to me, at least) a particularly convincing argument that real wage growth will be that much higher than forecast.

Anonymous said...

what Biggs leaves out is that by the time the 3% tax increase would be needed...that's one and a half percent for each the worker and the employer... or about ten dollars per week,,, wages will have increased about 200 dollars per week.

he also neglects to mention it would be pointless to increase the tax today... all that would do would be to increase the Trust Fund debt.

and of course the Big Fact that the experts can't seem to understand is that the money will be needed (if it is needed at all) to pay for the longer life expectancy of the people who will be paying the tax... in other words they will get their money back. with interest.

nope. far better to create an imaginary accounting system and then flap around the barnyard yelling the sky is falling the sky is falling... there's a 3% "deficit".

or note that the "looming deficit" can be closed by increasing the payroll tax one tenth percent per year starting in 2030. this turns out to be 2 cents per week per week.

Arne said...

My experience with statistics is in Statistical Process Control on a manufactuing line. I understand that if someone is talking about a 90 percent confidence limit, then there is an expectation that 10 percent of the time the actual result will fall outside the limit. If I look backwards from the 2004 chart, I see that about 50 percent would fall outside the envelope (for revenues per GDP), so 4 data points inside the envelope does not make me believe the variance has decreased.

Andrew G. Biggs said...

Arne, I may have misinterpreted your earlier comment. The pre-2004 figures on the CBO chart are historical, so there isn't an uncertainty range shown for them. After 2004, the CBO model projects uncertainty ranges for all the sub-variables, then adds them together to show income and cost as shown in the chart. What I tried to show was that in the range they predicted for which we now have historical data -- from 2004 through today -- the outcomes were pretty reasonable. Obviously this isn't hard to do over the short term so it's not that big a deal, but that was what I was getting at.

Arne said...

In SPC we use historical data to get an estimate of variance. We then use that variance to determine confidence limits. If I look at what kind of envelope would include 90 percent of historical data and compare it to the envelope that represents the stated confidence limits for the predicted years, I see a contradiction.