Monday, April 28, 2008

Paranoic Empiricism or Empirical Paranoia?

This post will take some time to develop, consider this installment one.

In 1997 I finally got my hands on the Annual Report of the Trustees of Social Security to find out why reported dates of Trust Fund depletion seemed to be moving. Well I found that answer, in short better than expected economic growth was driving that date out in time. But it was the numbers behind the numbers that provided the real surprises.

First I learned that the Social Security Trustees actually produce three different economic and demographic projections purporting to cover the full range of possible economic outcomes. The standard projection, the one always cited in the papers is officially called Intermediate Cost. It is accompanied by a more optimistic model called Low Cost and a more pessimistic model called High Cost. In looking at Low Cost two things jumped out. First its outcome was better than most people would have thought possible, if those numbers came home Social Security would be fully funded through the 75 year actuarial window. The second was even more interesting, the numbers needed going forward were not much different than the actual numbers of 1996, if the near to medium future looked like the near to medium past then we were essentially home free. Now in point of fact 1996 was at the time considered to be a pretty good year, having essentially identical numbers representing the top of a possible range of economic outcomes did not seem odd.

But then the 1998 Report came out and instead of the 2.5% Real GDP projected by Intermediate Cost (hereafter IC) or the 3.2% of Low Cost (LC) the actual number came in at 3.8%. Which was interesting enough in that it showed the short term model was too conservative, what the Trustees saw as the top end of the range clearly wasn't. But what was really interesting was the outcome of Low Cost. Other things being equal a better than expected first year number should have the mathematical effect of moving the curve up, 1998 Low Cost should have shown a better end result than 1997, fully funded should have yielded to somewhat over funded. But it didn't. While IC showed a full 3 year improvement with Trust Fund Depletion moving out from 2029 to 2032, LC stayed static at fully funded yet not overfunded.
Why was this? Well on examination in the face of a better than expected 1997 the reaction of the Social Security Trustees (or actually the staff actuaries in the Off of the Chief Actuary) was to adjust future growth numbers down, including a slightly lower number for the current year. It seemed a little odd that the net effect was equilibrium but odder things have happened.

But when lightning strikes twice you start to get jittery. Because in 1999 it happened again, growth numbers came in not only in advance of IC but also of LC, in fact strongly so with much the same result as in the 1998 Report, depletion moving out 2 years to 2034. But once again the outcome of Low Cost didn't budge, it still showed fully but not overfunded Social Security through the 75 year window. And on inspection you could see a further downward adjustment in future numbers. I began to smell a rat. In the face of two years of much better than expected numbers what would make you more pessimistic about the future?

Third time is a charm, the 2000 Report came out and the pattern was confirmed. Much better growth than even the 'optimistic' LC projected, Trust Fund depletion moving out in time by another 3 years to 2037 but LC stubbornly staying at its same equilibrium result.

Which is where the paranoia part comes in. Empirically the simplest explanation of the number series varying outside what would appear to be your confidence interval without a change to your upper limit is that it is in fact a deliberate limit, that the data was being fit to the curve. Now in a policy context this wasn't necessarily a bad thing, in fact it established Low Cost as a useful touchstone, outcomes closer to LC showing less need to address Social Security at all while any outcome better than IC at a minimum lowering the risk of doing Nothing short term. The problem is that the Trustees insisted this was not what was happening at all, instead these numbers were simply the best available information as analyzed by the career staff at the Social Security Administration. Moreover these numbers were examined and utilized by many other agencies including OMB and CBO, deliberate manipulation would imply a government wide conspiracy to conceal the true financial situation facing Social Security. And all of this mind you within a presumedly Social Security friendly Clinton Administration.

A empirical conclusion that seemed obvious based on pure examination of the number series crashed into political reality, there was no way to pull this off, it all just had to be a coincidence. Until it happened again, and again, and again. In all Reports from 2001-2007, now produced by the Social Security unfriendly Bush Administration, the result of LC remained constant: fully funded but not overfunded. It would appear that statistical lightning could strike the same effective point eleven years in a row. Yeah and I could let the money ride and hit 11 sevens in a row at a Craps table in Vegas, the smart money is against that happening.

Finally the pattern broke in 2008 with results to be seen below:
Shape of Low Cost This year Low Cost for the first time ever shows what an overfunded Trust Fund would look like, I can finally ditch paranoia and using the Trustees own numbers begin to explain the risks of an over funded system, an outcome that until 2007 had officially been deemed impossible by two Administrations. Just excuse me if I hold onto some of my paranoia, I formed a conclusion in 2000, openly posted it in Nov 2004 under What is the Low Cost Alternative and had it stand up to test with the 2005, 2006 and 2007 Report. I may have been wrong but I wasn't crazy, or at least not much, Occam's Razor suggested that Low Cost was being artificially constrained, careful examination of the numbers year over year showed clear signs of under motivated changes in data points for future years and it beggers belief that everything just happened to return the exact same equilibrium result.

My belief anyway.

No comments: