The following forecasts were produced by a statistical model characterized by a stochastic regime-drift (red curve in the figure below) that moves between two regimes. The first features a general acceleration of economic activity; the second features general deceleration. Below we shall refer to these as robust-growth and sluggish-growth. Regime changes occur stochastically with probabilities driven by an "observable tension" index defined as the geometric sum of deviations of actual GDP growth from "potential" GDP growth. The parameters of the drift process in any given regime are stochastic and drawn from a stable distribution. Full details can be found here.

The specification of the drift process was recently changed in order to provide additional flexibility. The motivation for the change can be traced back to the fact that the old model repeatedly failed to identify the current contractionary regime. In sharp contrast, the new model indicates that the economy transitioned into contraction in 2003:III.

The differences between the behavior of the drift process under each model are illustrated in the figure below for the period 2000:I-2008:I. (Click on the image for a larger version)

The panels represent the estimated drift processes (red curves) both with (right panels) and without (left panels) a regime-shift at 2003:III. The minor differences between right and left panels for the old model contribute to explaining why it fails to recognize a regime-shift in 2003:III. In contrast, the new model allows for greater flexibility and sharper differences and easily identifies the regime-shift.

On 01/29/2010, the Bureau of Economic Analysis announced that GDP grew at an annual rate of 5.73% in the fourth quarter of 2009.

According to our latest estimates, which take this development into account, the economy is in a state of sluggish growth.  The current probability assigned to this scenario is 79%.  The most likely date at which this regime began is the third quarter of 2003.

The confidence intervals use an estimate of the variance of the process produced over the "Great Moderation" period. Use of a larger variance (as estimated using post-war data) produces progressively larger confidence intervals (up to 40% larger).