Monetary Policy Strategies for a Low-Rate Environment

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Monetary Policy Strategies for a Low-Rate Environment Ben S. Bernanke, Michael T. Kiley, and John M. Roberts 2019-009 Please cite this paper as: Bernanke, Ben S., Michael T. Kiley, and John M. Roberts (2019). “Monetary Policy Strategies for a Low-Rate Environment,” Finance and Economics Discussion Series 2019-009. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2019.009. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Monetary Policy Strategies for a Low-Rate Environment By BEN S. BERNANKE, MICHAEL T. KILEY, AND JOHN M. ROBERTS * * Bernanke: Brookings Institution, 1775 Massachusetts Ave., advance to keep rates “lower for longer” (L4L) Washington DC 20036, BBernanke@brookings.edu; Kiley: Federal when the ELB is hit. As Kiley-Roberts and Reserve Board, Washington DC 20551, Michael.T.Kiley@frb.gov; Roberts: Federal Reserve Board, Washington DC 20551, other authors have shown, fully credible L4L John.M.Roberts@frb.gov. We thank Sage Belz and Michael Ng for policies could substantially ameliorate the ELB outstanding research assistance. Views expressed are not necessarily those of the Federal Reserve Board. problem. In particular, during ELB episodes, such policies should lead to lower bond yields The level of nominal interest rates consistent and higher expectations of inflation—both of with a neutral stance for monetary policy which reduce long-term real interest rates—as appears to be much lower than in the past. In a well as to greater optimism about future low-rate environment, the scope for monetary growth, all of which should encourage policy to respond to a slowing economy or spending and economic activity at the ELB. unwanted disinflation may be constrained by The assumption of full credibility of the the effective lower bound (ELB) on nominal policy framework is probably too strong, rates, which (for the case of the United States, however, especially during a period of examined here) we take to be zero. Kiley and transition to a new regime. Imperfect Roberts (2017), using simulation methods credibility poses several problems for L4L similar to those of the present paper, find that, policies. First, if public expectations of future under a policy rule estimated from historical inflation and growth do not respond as hoped data and the assumption that the neutral to central bank announcements, then these nominal interest rate is three percent, monetary policies are likely to be less effective at policy may be constrained by the ELB as much providing accommodation during ELB periods. as one-third of the time. This constraint in turn Second, if policymakers persist with low-rate leads to inferior macroeconomic performance. policies despite their imperfect credibility, the To address this issue, economists have made inflation overshoot that ultimately results could a number of proposals for modifying the lead to a costly un-anchoring of inflation Federal Reserve’s monetary policy framework. expectations or other problems, such as a Although details differ, these proposals buildup of financial risks (Brainard, 2017). generally involve policymakers committing in

Consequently, before adopting L4L policies, based on forecasts of a small-scale VAR model policymakers should be confident that they and, importantly, are assumed not to change would work reasonably well even if they are when the policy rule changes. Arguably, not fully credible with the public. MCAP better describes the situation during a We study this issue using stochastic transition to a new policy rule, about which simulations of FRB/US, the Federal Reserve’s financial market participants may have principal simulation model. For a suite of stronger incentives to be well informed than alternative policy rules, including some leading other agents.2 Under each of these assumptions, L4L rules, and assuming that the normal level we assess the effects of alternative policy of nominal interest rates is three percent, we regimes on macroeconomic outcomes and on consider economic performance under two the frequency of encounters with the ELB. alternative descriptions of expectations Briefly, we find that imperfect credibility formation. First, we consider “model- reduces—but does not eliminate—the consistent expectations” (MCE), under which advantages of L4L rules. In general, the L4L all private agents are assumed to know the rules that perform best strike a balance, structure of the economy, and, in particular, to providing adequate stimulus at the ELB while understand and believe the monetary policy avoiding sizable overshoots of inflation and rule. The MCE assumption, used by Kiley- output. Roberts (2017), seems appropriate for I. Description of Simulations situations in which a policy regime has been in place for some time.1 As noted, our analysis is based on stochastic Second, alternatively, we consider the case in simulations of the Fed’s FRB/US model, a which only asset-market participants have detailed econometric model of the U.S. model-consistent expectations (MCAP). Under economy. For each policy rule and for each of MCAP, bond yields, equity prices, and the the two alternative assumptions about exchange rate are determined in a forwardexpectations formation, we conducted 500 looking manner, but expectations of income, simulations of FRB/US, drawing shocks from inflation, and other nonfinancial variables are those realized over the period 1970-2015 (i.e., 1 2 Hebden and Lopez-Salido (2018) also consider a suite of L4L See Reifschneider and Roberts (2006) and Brayton, Laubach, and policy approaches under MCE in a small model, including versions of Reifschneider (2014) for further discussion both of FRB/US and of the temporary price-level targeting approaches we analyze, and reach these expectational assumptions. broadly similar conclusions to those herein.

bootstrapping residuals of the model). Each and an (2) “inertial” Taylor rule that includes a simulation is of 200 quarters. Results reported lagged interest-rate term, with a coefficient of below are drawn from the second 100 quarters 0.85 in quarterly data. The inertial term implies of each simulation, with the first 100 quarters slower adjustment of policy to economic used as a “burn in” period to establish initial developments. conditions. All simulations assume a neutral The first group of L4L rules we considered nominal interest rate of three percent and an includes three variants of flexible price-level inflation target of two percent. targeting (PLT). The PLT framework has L4L To gain further insight, for each policy rule features that theoretical analyses (e.g., and expectational assumption, we also Eggertsson and Woodford, 2003) suggest could simulated the economic and policy response to help address the ELB constraint. These rules a large shock to consumption, sufficient to are “flexible” in that they take into account drive the economy to the ELB. Those results output gaps and inflation gaps as well as the are discussed briefly below, with more details deviation of the price level from trend (the in the online appendix. “price level gap”). The three such rules we consider are: (1) a basic variant that augments II. Policy Rules the standard, non-inertial Taylor rule with the price-level gap; (2) an inertial version of the We evaluated the performance of ten rule, which adds a lagged interest-rate term; alternative policy rules, broken into four broad and (3) a non-inertial variant in which the price categories: baseline rules and three variants of level gap accumulates only during ELB L4L rules. For algebraic details and additional episodes and is zero otherwise (analogous to references, see the online appendix. The rules temporary PLT; see below). In each of these we studied are listed in Tables 1 and 2 below. variants, the price-level gap enters with the All rules impose a non-negativity constraint same (unit) weight as the output gap, so that (the ELB) on the implied policy rate. these rules can also be interpreted as For a baseline, we considered two variants of responding to nominal income gaps. the standard Taylor rule, which relates the The second group of L4L strategies we policy interest rate to deviations of inflation studied are rules that set an economic threshold from target (the “inflation gap”) and the output (here, defined in terms of inflation gap. The two variants are a so-called (1) performance) as a necessary condition for balanced-approach Taylor rule (Taylor, 1999)

leaving the ELB. The Fed’s Federal Open shadow rate—and that reflects the policy Market Committee adopted a threshold accommodation forgone because of the ELB. approach in December 2012. As they tie policy The actual policy rate is then set to compensate actions directly to observable economic for the forgone accommodation. Specifically, outcomes, threshold strategies may be we consider a policy proposed by relatively easy to communicate. Reifschneider and Williams (2000), in which The first threshold-type policy we considered the policy rate prescribed by the Taylor rule is was temporary price-level targeting. As reduced by cumulative forgone accumulation, discussed by Bernanke (2017), under this rule, until the latter is exhausted. We also consider a policymakers commit to deferring exit from the shadow-rate rule described in Kiley-Roberts ELB at least until any shortfall in inflation over (20017), in which the first difference of the the entire ELB period is fully made up. Away shadow rate depends on the sum of the inflation from the ELB, this rule is the same as the gap and the output gap, weighted by a inertial Taylor rule. parameter α. The actual policy rate is set equal A potential drawback of temporary PLT is to the shadow rate when it is non-negative and that, when the ELB episode is long and the is zero otherwise. Following Reifschneider and associated inflation shortfall is large, it could Roberts (2006), we set α = 0.4. imply a substantial overshoot of inflation after III. Simulation Results the ELB period ends. To mitigate that effect, we also considered variants of temporary PLT Results of the stochastic simulations are in which policymakers limit their inflation reported in Tables 1 and 2. Table 1 provides “lookback” periods to three years or to one results for the case of model-consistent year, respectively. For example, under expectations on the part of all private agents temporary PLT with one-year memory, the (MCE), while Table 2 results reflect the threshold for liftoff from the ELB is that assumption that only asset-market participants inflation over the previous year has been at or have model-consistent expectations (MCAP). above target. Finally, for our third group of L4L policies, [Insert Tables 1 and 2 about here] we consider two variants of so-called shadow- For each rule and expectational assumption, rate policies. Policies of this type define a the tables report: (1) the percentage of time notional policy rate that may be negative—the spent at the ELB; (2) the mean duration, in

quarters, of ELB episodes; (3) the mean output macroeconomic volatility offsets this benefit. gap; (4) the mean inflation rate; (5) the root In particular, the flexible PLT approaches mean square deviation (RMSD) of the output stabilize inflation but worsen output gap gap from its target of zero; (6) the RMSD of volatility, indicating that FRB/US does not inflation from its target of 2.0 percent. These share the close connection between price and are the same statistics reported by Kiley- output stability present in many New Roberts (2017). In addition, the last column of Keynesian models (Eggertsson and Woodford, each table reports an overall loss measure, 2003). Flexible PLT also performs very poorly equal to the sum of the squared deviations of under the MCAP assumption, presumably inflation and output from their respective reflecting the failure of the public’s inflation targets. Key takeaways from Tables 1 and 2 expectations to adjust to the policy regime. include the following: Third, although all three variants of threshold First, as in previous studies, traditional policy approaches based on the price-stability rules, such as Taylor rules, perform relatively mandate (lines 6-8) perform reasonably well poorly when the neutral nominal interest rate is under both expectational assumptions, the best low, as assumed here. For example, under both outcomes are achieved by variants of expectational assumptions, both variants of the temporary PLT with shorter inflation lookback Taylor rule (lines 1-2 of the tables) exhibit periods (shorter “memory”). Notably, frequent encounters with the ELB, with both temporary PLT with 1-year memory (line 8) output and inflation significantly below target, delivers significant improvements over on average. The inertial variant of the Taylor traditional rules under both expectational rule performs worse than the standard, non- assumptions. One reason may be that variants inertial rule. of temporary PLT with shorter memory are less Second, the performance of traditional policy prone to the risk of a large overshooting of rules is not much improved by the inclusion of inflation. Interestingly, despite their lower-forthe price level gap as a determinant of policy longer motivation, temporary PLT rules with rates (lines 3-5). For example, in terms of the shorter memories also result in less time spent loss function, flexible PLT (lines 3-4) does not at the ELB and shorter average ELB episodes. do noticeably better than the Taylor rules under For those concerned that long periods of zero MCE. Although output and inflation are closer rates raise the risk of financial instability or to target on average under PLT rules, greater

other adverse side effects, this feature is study (for each policy regime and expectational attractive. assumption) the responses of policy and the Fourth, shadow-rate rules do quite well in our economy to a single large aggregate demand simulations. The Reifschneider-Williams shock. Specifically, we studied the effects of a shadow-rate rule (line 9) performs similarly to negative shock to consumer expenditures on the temporary PLT approaches with shorter nondurables and non-housing services that, memory. The Kiley-Roberts change rule (line under the standard Taylor rule and MCE, leads 10) performs very well, which may owe to the to a decline in output of nearly 8 percent and an sizable long-run policy responses to output and ELB period lasting 20 quarters. Figures in the inflation deviations implied by the rule.3 online appendix show the simulated responses Overall, our simulations confirm earlier to this shock of the output gap, inflation, the results that, relative to traditional policy rules, federal funds rate, inflation expectations, and L4L rules can deliver better economic the ten-year Treasury interest rate for each rule outcomes when the neutral interest rate is low and each expectational assumption. Figure 1 and the ELB is accordingly a concern. As below is illustrative: It shows the behavior of expected, we also find that the advantage of inflation during and after the ELB episode, L4L policies relative to traditional policies is under MCAP and for three policy rules: the generally somewhat less under the MCAP inertial Taylor rule, inertial flexible PLT, and expectational assumption. Importantly, temporary PLT with one-year memory. though, a number of L4L policies retain a substantial advantage over traditional rules even when expectations outside the financial sector do not adjust to a change in the policy framework. IV. An Aggregate Demand Shock To gain more intuition about the performance FIGURE 1: EFFECTS OF AGGREGATE DEMAND SHOCK ON INFLATION of alternative rules, we also used FRB/US to 3 Henderson and McKibbin (1993) are an early example of analyses demonstrating that rules with sizable responses to output and inflation perform well in FRB/US relative to traditional policy rules.

Figure 1 illustrates a general lesson from target, resulting in volatility in inflation these simulations, which is that robust policy expectations and interest rates (not shown). frameworks share two characteristics: First, In contrast, in our simulations, temporary they must provide sufficient stimulus at the PLT rules with short memory and shadow-rate ELB. Second, on the other hand, they must not rules keep inflation close to target both during involve so much stimulus as to result in and after ELB episodes, under both MCE and excessively large overshoots in output and MCAP assumptions. Figure 1 illustrates for the inflation after the liftoff from the ELB. case of temporary PLT with 1-year memory, Taylor rules do poorly on the first of these under MCAP. As the figure shows, inflation criteria, at least for the scenario we consider. returns quickly to target but does not overshoot. Figure 1 illustrates for the MCAP case: As the In general, robust rules will balance the need to figure shows, following the hypothesized provide enough stimulus at the ELB with the aggregate demand shock, the inertial Taylor imperative of avoiding excessive overshoots, rule provides insufficient stimulus, with under both forward-looking and backwardinflation remaining below target more or less looking expectational regimes. indefinitely. As figures in the online appendix V. Conclusion show, for Taylor rules, the same pattern holds for inflation expectations and nominal interest Our principal finding is that, when neutral rates, under MCE as well as MCAP. interest rates are low and the ELB is a potential In our simulations, L4L rules generally problem, imperfect credibility of the policy provide more stimulus at the ELB than regime reduces but does not eliminate the traditional Taylor rules. Under MCE, most advantages of using “lower-for-longer” such rules return inflation and output to target policies. However, to deliver good results, such quickly. Under the MCAP results show in policies should be calibrated to balance the Figure 1, however, some of these rules also imperatives of providing sufficient stimulus at lead to significant overshoots of inflation, the the ELB and avoiding undesirably large concern raised by Brainard (2017). For the case overshoots of inflation and output. of flexible PLT with inertia, although inflation We do not view our results as definitive on returns to target more quickly than under the the question of which specific type of L4L Taylor rule, inflation ultimately overshoots its policy is best, however. That choice depends not only on details such as model specification

and parameter values but also on broader 2003. "The Zero Bound on Interest Rates and questions such as the ease of communicating Optimal Monetary Policy." Brookings alternative policies to the public. For example, Papers on Economic Activity, Fall 2003, arguably, threshold-type rules are easier to pages 139-235. explain and communicate than shadow-rate Hebden, James and J. David Lopez-Salido. rules, an advantage that in practice might 2018. "From Taylor's Rule to Bernanke's outweigh the modest advantage of the latter in Temporary Price Level Targeting." Finance our simulations. Likewise, any change in a and Economics Discussion Series 2018-051, well-established policy framework has costs, Board of Governors of the Federal Reserve which should be considered in the evaluation of System (US). alternative strategies. Henderson, Dale W. and Warwick McKibbin. 1993. "A Comparison of Some Basic References Monetary Policy Regimes for Open Economies: Implications of Different Bernanke, Ben S. 2017. “Monetary Policy in a Degrees of Instrument Adjustment and Wage New Era.” Presented at conference on Persistence." Carnegie-Rochester Rethinking Macroeconomic Policy, Peterson Conference Series on Public Policy. 39: 221– Institute of International Economics, October 318. 12-13. Kiley, Michael T. and John M. Roberts. 2017. Brainard, Lael. 2017. “Rethinking Monetary “Monetary Policy in a Low Interest Rate Policy in a New Normal.” Presented at World.” Brookings Papers on Economic conference on Rethinking Macroeconomic Activity, Spring 2017, 317-72. Policy, Peterson Institute of International Reifschneider, David L. and John M. Roberts. Economics, October 12-13. 2006. “Expectations Formation and the Brayton, Flint, Thomas Laubach, and David Effectiveness of Strategies for Limiting the Reifschneider. 2014. "The FRB/US Model: Consequences of the Zero Bound on Interest A Tool for Macroeconomic Policy Rates.” Journal of the Japanese and Analysis." FEDS Notes. Washington: Board International Economies, 314-37. of Governors of the Federal Reserve System, Taylor, John B. 1999. “Introduction.” April 3. https://doi.org/10.17016/2380- Monetary Policy Rules, University of 7172.0012 Chicago Press, 1-14. Eggertsson, Gauti B. and Michael Woodford.

TABLE 1— STOCHASTIC SIMULATION RESULTS UNDER MCE Mean ELB Mean Mean RMSD of duration of RMSD of frequency output inflation inflation Loss ELB output gap (percent) gap rate rate (quarters) 1. Taylor 38.3 10.9 -1.1 1.2 3.5 2.2 17.2 2. Taylor (inertial) 33.6 20.7 -1.4 1.0 3.9 2.4 20.7 3. Flexible price-level target 32.6 8.5 -0.4 2.0 3.6 1.5 15.2 3. Flexible price-level target (inertial) 24.6 13.8 -0.6 2.0 4.4 1.5 21.8 5. Flexible temporary price-level target 17.6 12.9 0.3 2.4 3.4 1.6 14.5 6. Temporary price-level target 16.3 12.5 0.0 2.3 3.1 1.7 12.6 7. Temporary price-level target (3 yr memory) 15.6 11.2 0.3 2.4 2.7 1.6 9.6 8. Temporary price-level target (1 yr memory) 15.1 9.4 0.2 2.3 2.5 1.5 8.5 9. Reifschneider-Williams 28.1 10.1 0.2 2.1 2.4 1.6 8.0 10. Kiley-Roberts change rule 37.0 16.9 -0.1 2.1 1.9 1.4 5.7 Notes: Results based on 500 simulations of 100 quarters each. Loss = (cid:2869)(cid:2869) Σ(cid:2895) Σ(cid:2898) (cid:4670)(cid:3435)(cid:2024) (cid:3398)(cid:2024)∗(cid:3439) (cid:2870) (cid:3397)(cid:1877)(cid:3550)(cid:2870)(cid:4671) for t, j period-simulations. (cid:2898)(cid:3012) (cid:3037)(cid:2880)(cid:2869) (cid:3047)(cid:2880)(cid:2869) (cid:3047),(cid:3037) (cid:3047),(cid:3115) Source: Authors’ calculations. TABLE 2— STOCHASTIC SIMULATION RESULTS UNDER MCAP Mean ELB Mean Mean RMSD of duration of RMSD of frequency output inflation inflation Loss ELB output gap (percent) gap rate rate (quarters) 1. Taylor 39.1 13.3 -1.4 1.6 3.6 1.5 15.2 2. Taylor (inertial) 39.6 30.1 -2.2 1.4 4.9 1.8 27.4 3. Flexible price-level target 36.1 14.1 -0.1 2.0 5.7 1.5 34.3 4. Flexible price-level target (inertial) 44.0 34.2 -2.8 1.0 7.4 2.4 60.3 5. Flexible temporary price-level target 20.8 22.0 1.0 2.3 5.0 1.7 25.0 6. Temporary price-level target 21.7 16.1 0.2 2.2 3.8 1.3 16.1 7. Temporary price-level target (3 yr memory) 8.3 7.8 1.6 2.6 3.8 1.3 16.2 8. Temporary price-level target (1 yr memory) 11.2 8.6 1.0 2.4 3.0 1.2 10.7 9. Reifschneider-Williams 19.4 9.3 0.5 2.3 3.2 1.2 11.4 10. Kiley-Roberts change rule 42.3 21.8 0.0 2.0 2.1 1.0 5.3 Notes: Results based on 500 simulations of 100 quarters each. Loss = (cid:2869)(cid:2869) Σ(cid:2895) Σ(cid:2898) (cid:4670)(cid:3435)(cid:2024) (cid:3398)(cid:2024)∗(cid:3439) (cid:2870) (cid:3397)(cid:1877)(cid:3548)(cid:2870)(cid:4671) for t, j period-simulations. (cid:2898)(cid:3012) (cid:3037)(cid:2880)(cid:2869) (cid:3047)(cid:2880)(cid:2869) (cid:3047),(cid:3037) (cid:3047),(cid:3037) Source: Authors’ calculations.