Monday, June 9, 2008

Tech Panel report on Low/High Cost scenarios

We've discussed in the past the use of the Low/High Cost scenarios for system financing, as the Low Cost are often cited as reason to believe the program will remain solvent in perpetuity without policy changes. Given that, the discussion of the Low/High Cost scenarios in the report the 2007 Technical Panel on Assumptions and Methods may be appropriate. What follows is from the executive summary:

The current approach to uncertainty of projections in the Trustees Report, using high-, medium- and low-cost "scenarios" or "variants" to indicate the range of plausible outcomes is a traditional one whose limitations are well-known. The current practice of assuming that all variables could simultaneously move in a low- or high-cost direction produces estimates that lack both an intuitive and a statistical interpretation. The temptation to assign probabilities to the scenarios or to suggest their degree of likelihood should be resisted. Previous Technical Panels have consistently drawn attention to the limitations of the variant approach. For example, the 1999 Technical Panel noted that using high and low alternatives: (1) assumes trajectories are always high or always low; (2) combines trajectories of various assumptions in rigid ways, for example, all are set at their high-cost value simultaneously; (3) ignores that different aspects of the high and low scenarios have different levels of uncertainty, and (4) does not assign any probability to the forecast ranges. We offer one additional observation: there is no requirement for symmetry of uncertainty—that the forecast be plus-or-minus an equal amount along the projection. Indeed, many key drivers have asymmetrical uncertainty as succeeding chapters will show, and the nature of the uncertainty may well change with the forecast horizon.

Stochastic analysis, on the other hand, produces uncertainty bands that are much easier to interpret. Critically, stochastic analysis can incorporate correlations between variables, and allows ranges to be given a probabilistic interpretation. Although, the actuaries have developed a stochastic model that is used to augment the use of scenarios to analyze uncertainty in the Trustees Report, those results appear more as an addendum than as an integral part of the analysis. The Panel therefore recommends using the results of the stochastic analysis to augment if not supplant the high- and low-cost scenarios, and to communicate the range of uncertainty around the intermediate projections.

This discussion mirrors part of the answer to Angry Bear's questions I posted a while back, in particular the argument that the stochastic model, while not perfect, represents a better way of displaying uncertainty than the Low/High Cost scenarios.

1 comment:

Bruce Webb said...

I thought this was one of the most exciting parts of the Report. The current model is quite crude, typically over the last few years when I deploy Low Cost economic numbers in historic context readers invariably come back pointing to the fertility without explaining to what degree these offset or how they interact with immigration. A more robust formulation along the lines sugggested by the Technical Panel seems very promising, because frankly the current stochastics do not seem to me to be getting the job done, seemingly simply measuring statistical variation effects and not actual probability of the base number of IC assumptions. Or so I surmise, I find the language of the stochastics opaque.