Assessing Market Forecasts of Inflation (at 5 Year Horizon)


Joseph E. Gagnon and Madi Sarsenbayev at the Peterson Institute for International Economics have an interesting article on whether economists or markets are better at forecasting. They write:

Previous posts (here and here) showed that long-term bond yields and long-term inflation compensation derived from bond yields are not good predictors of future inflation. Another post showed that economic forecasters are better at predicting inflation than consumers, but that neither group is especially accurate. This post compares the predictive accuracy of bond markets and economists over both short-term and long-term horizons. It finds that although short-term inflation compensation from bond yields has slightly better predictive accuracy than economists’ projections, economists are actually better at long-term forecasting.

I’ve been using 5 year breakevens to infer expected inflation in several posts. I thought it would be interesting to assess the usual metrics (mean error, root mean squared error) for the standard breakeven, and that adjusted for inflation risk and liquidity premia. (I don’t have a 5 year economists’ forecast to compare against, unfortunately).

Here are the ex post realizations and the corresponding forecasts.

Figure 1: Ex post CPI inflation over 5 years, annualized (black), implied forecast from 5 year Treasury-TIPS spread (red), implied forecast from 5 year Treasury-TIPS spread adjusted for inflation risk and liquidity premia, from DKW (blue), all in %. Source: BLS via FRED, Fed via FRED,  KWW following D’amico, Kim and Wei (DKW) accessed 6/4, and author’s calculations.

The adjustment methodology is described here. Notice that for the better part of the last decade, both forecasts overpredicted inflation. The unadjusted series clearly misses in late 2013, a miss not matched by the adjusted series, thereby highlighting the advantages of that series.

The corresponding forecast errors are shown below.

Figure 2: Ex post CPI inflation over 5 years prediction errors from 5 year Treasury-TIPS spread (red), and from 5 year Treasury-TIPS spread adjusted for inflation risk and liquidity premia, from DKW (blue), all in %. Source: BLS via FRED, Fed via FRED,  KWW following D’amico, Kim and Wei (DKW) accessed 6/4, and author’s calculations.

As a consequence of this pattern, while the mean error associated with the unadjusted series is smaller in absolute value than that for the adjusted (-0.06 pp vs. 0.35 pp), the root mean squared error is larger (0.79 vs 0.57). In addition, the adjusted series explains the actual CPI inflation with a significant slope coefficient of 0.81, R-squared of 0.28; the unadjusted series yields a slope coefficient of 0.09, R-squared of 0.01.

Gagnon and Sarsenbayev also discuss the forecasting performance of unadjusted breakevens at the 2 year and 5 year horizons.

As of today, the 5 year breakeven implies 2.46% inflation on average over the next five years. At the end of May, the adjusted breakeven implied an inflation rate 0.8 percentage points below the unadjusted breakeven.

Figure 3: Five year inflation breakeven calculated as five year Treasury yield minus five year TIPS yield (blue), five year breakeven adjusted by inflation risk premium and liquidity premium per DKW, all in %. Source: FRB via FRED, Treasury, KWW following D’amico, Kim and Wei (DKW) accessed 6/4, and author’s calculations.