The Power of Forward Guidance Revisited

The paper defines the puzzle via a concrete numerical example: “a promise by the central bank to peg interest rates below the natural rate of interest for roughly two years generates explosive dynamics for inflation and output in a workhorse New Keynesian model,” and, in the paper’s own calculation, “a promise by the central bank of a 1 percentage point lower real interest rate for a single quarter… has an 18 times greater impact on inflation when the promise pertains to interest rates five years in the future than when it pertains to the current interest rate” (Introduction, p. 3134). The term “forward guidance puzzle” is attributed to Del Negro, Giannoni, and Patterson (2012), building on results in Carlstrom, Fuerst, and Paustian (2015).

Section I shows this follows directly from the household’s consumption Euler equation, solved forward: because the promised real-rate change alters the relative price of consumption only between the two dates surrounding the change, “consumption growth can only deviate from normal in quarter 20” (the announced date), so consumption “must rise by 1 percent immediately” and stay at that higher, constant level until the shock passes – “a step function” whose height does not depend on how far away quarter 20 is (Sec. I, p. 3137). Formally, solving the intertemporal IS equation forward yields an “undiscounted” sum of expected future real-rate deviations (footnote 7): “there is no discounting in the sum on the right-hand side of this equation,” which is the algebraic source of the puzzle.

The paper argues that “many people face some risk of hitting a borrowing constraint over the next five years. This effectively shortens their planning horizon,” and separately, “people’s desire to maintain a buffer stock of saving for precautionary reasons will temper their response to future interest rate shocks,” since “taking full advantage of the opportunity for intertemporal substitution presented by the future interest rate change requires people to run down their assets,” which “is costly since it leaves them more exposed to future income shocks” (Introduction, p. 3135). As the promised rate change is pushed further into the future, “the change in assets needed to take full advantage of intertemporal substitution grows and the countervailing precautionary savings effect therefore grows stronger, tempering the effects of forward guidance.”

Comparing the incomplete-markets model to its complete-markets counterpart, “in the incomplete markets model the initial increase in output is only about 40 percent as large” as the 25 basis point jump predicted under complete markets, and “the initial response of inflation in the incomplete markets model is again only about 40 percent as large as in the complete markets model” (Sec. III, Figures 3-4, p. 3147-3148). The paper also documents an asymmetry and an aftermath effect not present under complete markets: after the promised rate change passes, “output falls below steady state for some time in the incomplete markets case,” because the shock “leads to a redistribution of wealth away from households with high marginal propensities to consume and toward households with low marginal propensities to consume,” which “lowers aggregate demand… until the distribution of wealth has had time to converge back to steady state” (Sec. III, p. 3148, fn. 17).

Figure 5 shows that “in the complete markets model, output always rises by 25 basis points, regardless of the horizon of the forward guidance. In contrast, in the incomplete markets model, the effect is only about 20 basis points for an announcement about the real interest rate next quarter and falls monotonically thereafter. It is roughly ten basis points for an announcement about the real interest rate five years ahead, and essentially zero for an announcement about the real interest rate ten years ahead” (Sec. III.A, p. 3149). The intuition given is that “the probability of hitting a borrowing constraint before the interest rate change rises with the length of time until the interest rate change occurs,” lowering the expected benefit of responding, while “the cost of responding rises since the amount the household would need to run down its assets… gets larger and larger” – eventually making the trade-off not worth it at all. The paper notes (footnote 20) that a companion paper shows this behaves like an Euler equation with an explicit “discount factor” on expected future consumption growth, in contrast to the standard undiscounted Euler equation.

In a “high risk” calibration that roughly doubles the volatility of idiosyncratic productivity shocks (to match evidence in Guvenen, Ozkan, and Song 2014), “the response of output in this case is only about 20 percent of the complete markets benchmark and the response of inflation only about 32 percent,” i.e. the muting is substantially stronger than in the baseline (Sec. III.B, p. 3150). Conversely, in a “high asset” calibration that raises the model’s asset-to-GDP ratio to match total net worth (3.79) rather than only liquid assets (1.4), “the output response rises to 58 percent of the complete markets benchmark, while the inflation effect rises to 49 percent” – because larger precautionary buffers make households less reluctant to draw them down, moving the model closer to the complete-markets benchmark. A calibration combining both changes falls close to the baseline, since the two effects “largely offset each other.”

Simulating a discount-factor shock calibrated (following Eggertsson and Woodford 2003) to generate an initial 4 percent output decline and a 20-quarter ZLB spell under a naive Taylor-rule policy, the paper compares this “naive” policy to an “extended” policy that holds the nominal rate at zero for additional quarters, chosen “to fully eliminate the initial fall in output in the complete markets model” – which turns out to require holding rates at zero for about 3 extra quarters (23 vs. 20) (Sec. III.C, p. 3153-3154). The paper finds that “while the extended monetary policy fully eliminates the recession in the complete markets case, a substantial recession remains in the incomplete markets model,” and the extended policy is far less successful at preventing deflation: “the initial deflation is only about 30 basis points in the complete markets case,” but “more than 100 basis points in the incomplete markets case,” reflecting higher real rates (since nominal rates are stuck at zero) and contributing to the larger output fall.

The authors explicitly contrast their result with Krusell and Smith (1998), who “compare the response of a complete markets and incomplete markets model to a productivity shock and find that the difference is small,” whereas “we find a large difference between the complete and incomplete markets model in response to forward guidance about real interest rates” (Introduction, p. 3136). They attribute this to the behavior of the real interest rate: “in the flexible price setting considered by Krusell and Smith (1998), these changes in demand are largely undone by adjustments in real interest rates. However, when prices are sticky and real interest rates respond sluggishly to shocks, the effects of market incompleteness and household heterogeneity on demand lead to substantial changes in output relative to the complete markets case” – i.e. the sticky-price setting is essential for market incompleteness to matter this much.

The paper deliberately parameterizes its model “so as to make [redistributional] effects small (since they are not the focus of our paper),” noting that “allowing for such redistributional effects can make the on-impact effects of a shock to the current interest rate quite a bit bigger than in our model,” as in Guerrieri and Lorenzoni (2015) and, for balance-sheet-driven redistribution, Auclert (2016) (Introduction, p. 3136). They also note that Werning (2015) “shows how general equilibrium feedback effects of interest rate changes on household income can under certain conditions result in much larger effects on output than in our model, so that the effect of forward guidance may be unchanged relative to a representative agent benchmark” – signaling that the paper’s muted-forward-guidance result is not a generic feature of all incomplete-markets New Keynesian models, but depends on the specific calibration and general-equilibrium feedbacks (notably, a comparatively weak indirect income effect in their setting, as shown in their Appendix Figure A2) the authors emphasize.

The conclusion states plainly: “allowing for uninsurable income risk and borrowing constraints substantially decreases the power of forward guidance relative to a New Keynesian model with complete markets,” with the horizon-dependence summarized as “forward guidance about interest rates five years in the future has only about 40 percent as large an effect on current consumption as in the complete markets case, while forward guidance about interest rates ten years in the future has essentially no effect on current consumption” (Sec. IV, p. 3156). The authors flag that their analysis, while focused on monetary policy shocks, has broader reach: “movements in real interest rates play an important role in the response of the economy to a wide variety of shocks. The arguments we have made will therefore affect the response of the economy to a wide variety of shocks” – i.e. the precautionary-savings discounting mechanism is not specific to monetary forward guidance but should apply wherever a model relies on the response of aggregate demand to anticipated future real-rate paths.