On the accuracy of Federal Reserve forecasts of the saving rate

Abstract

This study investigates the unbiasedness and efficiency of the Federal Reserve forecasts of the saving rate for two distinct periods. For 1984–1997, when the saving rate was relatively stable, the forecasts are generally unbiased and efficient. For 1998–2007, when the saving rate experienced significant declines, the forecasts are generally biased (over-predicting) but still efficient. Evidence of efficiency or ‘weakly’ rationality implies that the Federal Reserve forecasts were generated under symmetric (asymmetric) loss for 1984–1997 (1998–2007). Under asymmetric loss where the forecast errors of the same magnitude but of different signs possess different costs, biased forecasts may still be optimal if the bias is strictly due to asymmetric loss. As such, for 1998–2007, the Federal Reserve staff may have assigned high cost (loss) to under-predictions but little or no cost to over-predictions due to, perhaps, perceiving the unusual declines in the saving rate as temporary.

Keywords:

• forecast evaluation
• saving rate
• FOMC
• loss structure

JEL Classification:

• E21
• E27

Notes

¹ See Cohn and Kolluri (2003) for the determinants of household saving in the G-7 countries.

² See McConnell and Perez-Quiros (2000) on the Great Moderation.

³ Real-time data are available at http://www.phil.frb.org

⁴ The calculated ADF test statistic for testing the unit root null hypothesis for Y_t (ΔY_t) is –1.90 (–9.50) with a p-value of 0.332 (0.000). The ADF test equations include a constant with the number of augmented lags determined using the Schwarz criterion